You’ve optimised for Google. You’re ranking. But when your prospects ask ChatGPT or Perplexity for recommendations, your brand doesn’t show up. Instead, they see your competitors cited as sources, woven into the answer itself. This is the new visibility problem. AI-powered search isn’t replacing Google overnight, but it’s changing how buyers research and make decisions. And the rules for showing up are different. Rankings don’t guarantee citations. You need to be retrievable, quotable, and verifiable or you’re invisible in the answers that matter. This guide covers how AI citation actually works, what makes content citable, and a practical framework for improving your chances of being cited. We’ll look at the data, walk through specific tactics, survey the tools available for tracking your AI visibility, and be honest about what we still don’t know.

Key Takeaways

• The problem: 58.5% of Google searches now end without a click. When AI Overviews appear, CTR for the #1 result drops by 34.5%. Your content is being consumed, but you’re not getting the visit.
• The new metric that matters: Citation Share how often your brand is cited in AI-generated answers compared to competitors. If you’re not the footnote, you’re invisible.
• What gets cited: AI systems assemble evidence, not rankings. They favor content that directly answers questions, contains specific data, and comes from verifiable sources. The GEO research shows that adding citations, statistics, and quotations can boost visibility by up to 40%.
• The 4-part framework for AI citability:

1. Answer-first architecture (BLUF): Put the direct answer in the first 50-100 words of each section. AI extracts what’s clearest, not what’s buried.
2. Data density: Unique stats, benchmarks, and scoped claims get cited. Vague benefit statements get paraphrased away.
3. Entity authority: 86% of AI citations come from sources brands can control (websites, listings, reviews). Consistent messaging across LinkedIn, G2, and Crunchbase makes you confirmable.
4. Technical readiness: Server-rendered HTML, schema markup, and optionally llms.txt help AI systems retrieve and parse your content.

• What to do first: Audit your top 10 pages for BLUF structure. Add one unique data point to each. Align your entity footprint. Start manual AI monitoring while you evaluate tools.
• The uncomfortable truth: We’re still early in understanding what drives citations. These are best practices based on available research, not guarantees. The AI systems change fast, so focus on fundamentals that seem durable.

I. The “Ghost Traffic” Problem

Here’s the uncomfortable shift I’ve been watching: for a growing share of informational searches, the search session ends without a click because the answer is already on the page (or in an AI summary).

A large-scale clickstream study by SparkToro using data from Datos (a Semrush company), analysing millions of devices between September 2022 and May 2024, found that 58.5% of U.S. Google searches ended as zero-click (no result clicked), and only 360 clicks per 1,000 searches went to the open web. In the EU, it was 374 open-web clicks per 1,000 searches with 59.7% ending in zero clicks.

And when AI Overviews show up, things get worse. Ahrefs analysed 300,000 keywords 150,000 with an AI Overview present and 150,000 informational keywords without and found they correlate with a 34.5% lower CTR for the #1 ranking page (comparing March 2024, before the U.S. rollout of AI Overviews, to March 2025, after). The average position-one CTR dropped from 7.3% to just 2.6% year-over-year for keywords that triggered AI Overviews.

So the pain isn’t just “zero-click.” It’s what I call ghost traffic: your expertise is being consumed but you’re not getting the visit, the pixel, the retargeting pool, or the attribution.

The 2026 KPI Shift: CTR → Citation Share

When buyers ask an AI, the “winner” isn’t the #1 blue link it’s the source inside the answer.

Citation Share (a practical definition):

Your brand/domain citations for a topic cluster ÷ total citations shown across your tracked prompts for that cluster.

If you aren’t the footnote, you’re increasingly invisible in the buyer’s journey.

II. The Mechanics of AI Citation: Why Some Sites Win

If you’ve ever wondered why some pages get cited and others don’t, here’s the short version.

Most answer engines follow a retrieve → synthesise → cite pattern (often called RAG, retrieval-augmented generation). They pull documents from an index, decide which chunks are useful, then generate a response grounded in those sources.

In other words: they don’t “rank pages,” they assemble evidence.

What Tends to Get Cited?

1) Consensus + utility beat keyword matching

Generative engines look for sources that:

• directly answer the question
• contain specific, usable facts
• are consistent with other reputable sources

2) “Data beats vibes”

In the GEO research literature (Aggarwal et al., Princeton/Georgia Tech/Allen Institute/IIT Delhi, published at KDD 2024), adding citations, quotations, and statistics can materially increase visibility in generative answers. The GEO paper reports visibility lifts up to 40% on the Position-Adjusted Word Count metric, with the top-performing methods (Cite Sources, Quotation Addition, and Statistics Addition) achieving 30-40% improvement across diverse queries.

That maps to a simple reality: LLMs like “hard edges” (numbers, definitions, scoped claims) because they’re easier to justify and attribute.

Important caveat: This research is still early. We’re learning what drives citations in real-time, and AI systems change rapidly. These are best practices based on available research, not guarantees.

III. If You Only Have 4 Weeks: A Prioritisation Roadmap

Before diving into the full framework, here’s what to do if you need to show progress fast:

Week 1-2: Audit Your Top 10 Pages for BLUF Architecture

Pull your highest-traffic informational pages. For each one, check: does the first 100 words directly answer the implied question? If not, rewrite the intro. This is the highest-leverage change you can make.

Week 3-4: Add One Piece of Original Data to Each Key Page

Even if it’s from an internal customer survey, product usage stats, or a simple benchmark you’ve run. Replace vague benefit statements with specific, measurable claims.

Month 2: Align Your Entity Footprint

Update LinkedIn, G2, Crunchbase, and your website About page to use the same positioning language. This is often overlooked but directly impacts how confidently AI systems cite you.

Ongoing: Set Up Basic AI Monitoring

Start with manual testing (see Section VII) while you evaluate tools. Query AI systems weekly with your core keywords and track whether you’re cited.

IV. The 4-Step Framework for AI Citability

Step 1: The “Answer-First” (BLUF) Architecture

BLUF = Bottom Line Up Front.

AI systems (and impatient SaaS buyers) reward sections that deliver the core answer immediately.

Rule of thumb: Put the likely answer to the user’s prompt in the first 50–100 words of each major section.

Before (fluffy):

“In today’s evolving landscape, it’s important to understand how AI is transforming content discovery…”

After (BLUF):

“To get cited in AI answers, your page needs a direct, quotable answer near the top, unique data worth attributing, consistent entity signals across the web, and machine-readable structure (schema + crawlable HTML).”

Make it scannable:

• Use question-based H2s (e.g., “How to get cited in ChatGPT Search?”)
• Follow with a 1–2 sentence answer
• Then expand with bullets, steps, examples

This isn’t just style. It’s extractability.

Why BLUF Matters for AI Systems

When an AI model processes your page, it doesn’t read like a human who’s willing to scroll. It’s looking for the clearest, most direct answer to surface. Content buried below three paragraphs of preamble often gets skipped in favour of a competitor who puts the answer first.

The GEO research found that stylistic changes like improving fluency and readability (Fluency Optimisation and Easy-to-Understand methods) resulted in visibility boosts of 15-30%. This suggests generative engines value not only what you say but how clearly you say it.

Step 2: Data Density & “The Citation Magnet”

If you want citations, you need citation-worthy assets.

Think in terms of primary-source value:

• Original benchmarks
• Proprietary stats
• Mini-surveys
• Teardown comparisons
• Pricing/packaging matrices
• Process metrics from real deployments

Ahrefs’ research confirms that 99.2% of keywords triggering AI Overviews are informational in intent. That means a massive amount of “how-to” content is getting summarised instead of clicked.

Your defence is to publish what AI can’t easily rewrite without attributing: unique numbers and concrete claims.

Actionable tip: Replace “benefit statements” with “measurable outcomes.”

❌ “Our tool saves time.”

✅ “Teams cut procurement cycles by 22% (median across 41 workflows).”

Even better: Show your work

When you make a claim with data, specify:

• Sample size
• Timeframe
• What the metric actually measured
• A methodology note or link

This is how you become the “source of truth,” not another paraphrasable blog.

Example: Before and After

Before (generic, paraphrasable):

“Email personalisation improves engagement and helps companies connect better with their customers.”

After (specific, citation-worthy):

“Personalised subject lines increased open rates by 26% across 1.2M emails sent in Q3 2025 (n=47 B2B SaaS companies, median list size 12K). Segmented sends outperformed batch sends by 3.1x on click-through.”

The second version gives an AI something concrete to attribute. The first version is just vibes.

Step 3: The Entity Authority Loop (Off-Page AEO)

Even if your content is perfect, models still ask: “Should I trust this source?”

Yext analysed 6.8 million citations across major AI models (ChatGPT, Gemini, and Perplexity) between July and August 2025 and found 86% of citations come from sources brands can manage or strongly influence. The breakdown: websites generated 44% of citations (2.9M), listings 42% (2.9M), and reviews/social 8% (545K). Forums like Reddit accounted for just 2% once location context and query intent were applied.

Importantly, citation patterns vary by intent and context. For unbranded objective queries often the most discoverable first-party websites and local pages made up nearly 60% of citations.

Practically: your website is necessary, but your entity footprint is what makes you confirmable.

Your Entity Authority Checklist (B2B SaaS Edition)

Consistent name + description across:

• LinkedIn company page
• Crunchbase
• G2 / Capterra (or relevant review sites)
• Founder/exec profiles
• Notable podcasts/newsletters/guest posts

Same positioning language everywhere (category, target customer, core outcome)

Real-world proof: Customer quotes, case studies, partner pages

Actionable tip: Update Organisation schema + align your third-party profiles to match it (same brand name, URL, logo, socials).

The goal: when an AI cross-checks your brand, it finds a consistent story.

Why Entity Signals Matter

AI systems don’t just look at your page in isolation. They cross-reference. If your website says you’re a “marketing automation platform” but your G2 listing says “email marketing software” and your LinkedIn says “sales enablement tool,” that inconsistency makes you less likely to be cited confidently.

Think of it this way: the AI is trying to write a sentence that says “According to [Brand], [claim].” If it can’t be confident about what Brand actually is or does, it’ll pick a competitor whose story is clearer.

Step 4: Technical AI Readiness

You can’t get cited if you can’t get retrieved.

Start with basics:

• Important content visible in server-rendered HTML
• Fast, crawlable pages
• Logical internal linking (so “orphan” pages aren’t invisible)

Then add machine-readable clarity.

A) llms.txt (the emerging “AI discovery file”)

/llms.txt is a proposed standard (outlined at llmstxt.org) to provide an LLM-friendly map of your site what you are, what matters, and where the canonical pages live.

It’s early, adoption is uneven, and it’s not a magic switch. Google’s Gary Illyes stated in July 2025 that Google doesn’t support llms.txt and isn’t planning to. But it’s a low-effort way to reduce ambiguity for AI systems that choose to read it, and adoption is growing among developer-focused sites (Stripe, Anthropic, etc.).

Best-practice guidance from the llms.txt specification:

• Keep it concise (aim for a curated list, not a sitemap)
• Use factual language (avoid marketing hype)
• Link to authoritative pages
• Update when offerings change

Example skeleton:

# Genrank 
> Become the source AI engines like ChatGPT, Google Gemini, Perplexity, Grok, and Claude trust and cite

## What we do
- AEO Audits: Comprehensive content analysis across five dimensions with 0-100 scoring
- Segment Audits: Analyze groups of similar pages by URL pattern for template-level optimization
- Content Editor: AI-powered tool for generating optimized definitions, summaries, and meta descriptions

## Key pages
- https://genrank.co/ (overview)
- https://genrank.co/docs/ (documentation)
- https://genrank.co/blog/ (articles and research)
- https://genrank.co/pricing/ (plans)

## Contact
- https://genrank.co/contact/

B) Structured Data (Schema)

Schema helps “other applications” (not just Google) interpret your content.

For B2B SaaS, start with:

• Organisation (who you are)
• SoftwareApplication (what your product is)
• FAQPage for tightly scoped Q&A content

Important nuance: Google announced in August 2023 that it would reduce the visibility of FAQ rich results, limiting them to “well-known, authoritative government and health websites.” They also fully deprecated HowTo rich results in September 2023.

So don’t implement FAQ schema expecting Google SERP bling that ship has largely sailed for most businesses. But the schema can still help AI parsers understand your page structure, and some practitioners argue that FAQ schema has become more important for AI search even as it became less visible in traditional SERPs.

C) Crawlability and Render Hygiene

AI systems rely on being able to fetch and parse your content. Make sure:

Content is in the initial HTML: If your key content only appears after JavaScript executes, some crawlers will miss it. Server-side rendering or static generation is preferred.

Check your robots.txt: Some sites inadvertently block GPTBot, ClaudeBot, or other AI crawlers. While you may have legitimate reasons to block them, understand the trade-off.

Clean HTML structure: Semantic HTML (proper heading hierarchy, paragraph tags, lists) helps AI parse your content.

V. A Mini-Teardown: What Gets Cited vs. What Gets Ignored

Let’s look at two hypothetical versions of the same content and see why one gets cited.

Version A: The “Standard Blog Post”

Title: “Why Customer Retention Matters”

Opening paragraph:

“In today’s competitive business landscape, customer retention has become more important than ever. Companies that focus on keeping their existing customers often find that it leads to better business outcomes. Let’s explore why retention should be a priority for your organisation.”

What’s wrong: No specific claim an AI can cite. Vague (“more important than ever”), no data, no unique perspective.

Version B: The “Citation Magnet”

Title: “Customer Retention Benchmarks: What Good Looks Like in B2B SaaS (2025 Data)”

Opening paragraph:

“The median net revenue retention for B2B SaaS companies with $10M-50M ARR is 105%, based on our analysis of 127 companies in our portfolio. Top-quartile performers hit 115%+ NRR. Below 95% NRR, growth becomes significantly harder you need to acquire 1.3 new customers just to replace each churned customer’s revenue.”

Why this works:

• Specific numbers that can be attributed
• Defined scope (ARR range, sample size)
• A clear, quotable insight
• Original data that AI can’t find elsewhere

When an AI is asked “What’s a good net revenue retention rate for SaaS?”, Version B gives it something to cite. Version A just adds noise.

VI. The GEO/AEO Tools Landscape: What’s Available Today

One of the biggest challenges in GEO is measurement. Here’s an honest look at what’s available:

Enterprise Platforms

seoClarity (Clarity ArcAI)

• Tracks AI Overviews, ChatGPT, Perplexity, Gemini, Copilot, and Google AI Mode
• Enterprise-grade with actionable insights layer
• Pricing: Custom quotes, typically $2,500-4,000+/month for full suite
• Best for: Large enterprises needing comprehensive tracking + traditional SEO in one platform
• Caveat: Steep learning curve, significant setup time

Ahrefs Brand Radar

• Tracks brand mentions across 6+ AI platforms (ChatGPT, AI Overviews, Perplexity, Gemini, Copilot)
• 190M+ prompts in database, zero setup required
• Integrates with existing Ahrefs ecosystem
• Pricing: Add-on to Ahrefs subscription ($199/month per index, or $699/month for all indexes)
• Best for: Teams already using Ahrefs who want AI visibility data alongside traditional SEO
• Caveat: Some users report accuracy issues for ChatGPT/Perplexity specifically; functions more as research database than simple tracker

Mid-Market / SMB Tools

Otterly.ai

• Tracks ChatGPT, Perplexity, Google AI Overviews, Gemini, Copilot, AI Mode
• GEO audit tool included (one of the more detailed audits available)
• Brand visibility index, citation tracking, sentiment analysis
• Pricing: Lite $29/month (10 prompts), Standard $189/month (100 prompts), Premium $489/month (400 prompts)
• Best for: SMBs and agencies wanting a focused AI visibility tool without enterprise complexity
• Caveat: Smaller user base, some feature gaps vs. enterprise tools

SE Ranking

• AI Overview tracking within broader SEO platform
• Good if you want AI + traditional SEO monitoring unified
• Pricing: Starts around $65/month
• Best for: Teams wanting one tool for both traditional and AI search

Specialised / Emerging Tools

Profound: AI visibility across Copilot, Perplexity, Claude with competitive analysis. Enterprise pricing ($3,000+/month).

Scrunch AI: AI citation and sentiment tracking. Emerging player.

Rankability: AI visibility tracking plus optimisation recommendations. Growing platform.

LLMrefs: Tracks visibility across 10+ AI platforms including Claude and Grok. Keyword-focused approach.

The Free Option: Manual Monitoring

If you’re not ready to invest in tools, manual testing still works:

1. Create a spreadsheet of 20-30 prompts relevant to your business
2. Query each AI system (ChatGPT, Perplexity, Gemini, Google with AI Overviews) weekly
3. Record: Were you cited? What was said? Who else was cited?
4. Track changes over time

It’s tedious, but it builds intuition and costs nothing. Start here before buying tools.

A Note on What We’re Building

Full disclosure: I’m building Genrank to solve this problem. We’re focused specifically on citation tracking and share-of-voice analytics for AI search helping you understand not just if you’re cited, but why (or why not), and what to change.

We’re still early (waitlist stage), so if you need something today, the tools above are your best options. If you want early access to what we’re building, you can join the waitlist at genrank.co.

VII. Measuring Your GEO Progress

Without measurement, you’re guessing. Here’s how to track progress:

Manual Testing Protocol

Query AI systems weekly with prompts relevant to your business:

• “What tools help with [your category]?”
• “How do I [problem your product solves]?”
• “What’s the best [your product type] for [use case]?”
• “[Competitor] vs [your brand]”
• “What companies offer [your service type]?”

Document:

• Were you cited? (Y/N)
• If yes, what exactly was said about you?
• Who else was cited?
• How were you positioned relative to competitors?

Proxy Metrics (While AI Attribution Matures)

Most analytics tools can’t attribute AI citations directly yet. Watch these instead:

Direct traffic: May indicate AI-driven discovery (users hear about you via AI, then navigate directly)

Branded search volume: AI mentions can drive searches for your brand name

AI referrals: Some platforms now show up in referral traffic (chatgpt.com, perplexity.ai). Set up segments in GA4.

What “Good” Looks Like

There’s no universal benchmark yet, but directionally:

• Being cited for 20%+ of your core topic prompts is strong
• Positive or neutral framing (vs. negative mentions) matters
• Appearing alongside top competitors (not being absent) is the baseline

VIII. Common Mistakes to Avoid

The GEO research paper found that some traditional SEO tactics actively hurt AI visibility:

Keyword stuffing decreased visibility by 10%. AI systems penalise content that feels optimised rather than genuinely helpful.

Generic claims don’t get cited. “We help businesses grow” gives an AI nothing to attribute. Specific claims get citations.

Inconsistent information undermines trust. If your blog says one thing and your product page says another, AI systems may cite neither.

Burying the lede. Putting your answer after three paragraphs of context means AI might extract a competitor’s clearer answer instead.

Ignoring entity signals. Perfect on-page content with a messy off-page footprint (inconsistent listings, outdated profiles) reduces citation confidence.

IX. What We Don’t Know Yet

Honesty is important here: this field is evolving fast, and there’s a lot we’re still learning.

We don’t fully understand citation ranking. Within an AI response, why does Source A appear before Source B? The ranking factors aren’t clear.

Platform-specific optimisation is murky. Do ChatGPT and Perplexity weight different signals? Probably, but the research is thin.

The impact of AI-specific markup (llms.txt, etc.) is unproven. It makes theoretical sense, but there’s no rigorous study showing it improves citations.

Long-term stability is unknown. Will the tactics that work today still work in 6 months? AI systems iterate quickly.

The best approach is to focus on fundamentals that seem durable (clear content, specific data, entity consistency) while staying adaptable.

X. Conclusion: Win by Being Verifiable

The content teams that win in 2026 won’t be the ones who publish the most.

They’ll be the ones who are the most:

• Extractable (BLUF + structure)
• Specific (data density)
• Confirmable (entity footprint)
• Machine-readable (llms.txt + schema + crawlable pages)

And importantly: the ones who measure their AI visibility instead of guessing.

Quick “AI Citability” Audit Checklist

On-Page

• Each section answers the question in the first 50–100 words
• Includes at least one unique stat, benchmark, or table per key page
• Claims include scope (who/when/how measured)
• Clear definitions + constraints (no vague generalities)
• At least one “quotable sentence” per major section
• Proper heading hierarchy (H1 → H2 → H3)
• Content visible in initial HTML (not requiring JavaScript)

Off-Page / Entity

• Consistent brand story across LinkedIn + Crunchbase + review sites
• Organisation schema matches your public profiles
• At least 3 credible third-party mentions/reviews/case studies
• Same positioning language used everywhere
• Key executives have updated, consistent profiles

Technical

• Important content is in server-rendered HTML
• /llms.txt exists and is concise + factual (optional but recommended)
• Schema implemented for Organisation + key content types
• AI crawlers not blocked in robots.txt (or blocked intentionally with understanding of trade-off)
• Pages load quickly (under 3 seconds)
• Clean internal linking (no orphan pages)

Measurement

• Weekly manual AI prompt testing set up
• Baseline citation share documented
• AI referral segment created in analytics
• Tool evaluation underway (or monitoring in place)

Resources and Further Reading

Zero-Click Search Study (SparkToro + Datos, 2024)
https://sparktoro.com/blog/2024-zero-click-search-study-for-every-1000-us-google-searches-only-374-clicks-go-to-the-open-web-in-the-eu-its-360/

AI Overviews Reduce Clicks by 34.5% (Ahrefs, 2025)
https://ahrefs.com/blog/ai-overviews-reduce-clicks/

GEO: Generative Engine Optimization (Princeton, Georgia Tech, Allen Institute, IIT Delhi)
https://arxiv.org/abs/2311.09735

Yext AI Citations Research (2025)
https://www.yext.com/research/article/ai-citations-user-locations-query-context

llms.txt Specification
https://llmstxt.org/

Google’s FAQ/HowTo Rich Results Changes (2023)
https://developers.google.com/search/blog/2023/08/howto-faq-changes

Ahrefs Brand Radar
https://ahrefs.com/brand-radar

seoClarity AI Overview Tracking
https://www.seoclarity.net/ai-overviews-tracking

Otterly.ai
https://otterly.ai/

For two decades, content strategy revolved around keywords. We built campaigns around search volume, difficulty scores, and keyword placement. The assumption was simple: match the words people type, and you'll rank.

AI systems don't work this way. They don't search for strings of text, instead they search for concepts. When someone asks ChatGPT or Google's AI Overview a question, the system isn't matching keywords. It's querying a knowledge graph to find the most confident relationships between entities.

This is the shift that separates traditional SEO from Answer Engine Optimization: moving from optimizing for keywords to optimizing for entities. If your content strategy is still built entirely around keyword research, you're optimizing for a system that's being replaced.

At Genrank, we've built entity analysis into our platform because we've seen this pattern consistently across our analysis of over 500,000 generative queries: content can rank well organically but get ignored by AI systems. The page might be the best document on a topic, but if the brand isn't established as a trusted entity for that topic in the AI's knowledge graph, it won't get cited.

How AI knowledge graphs work

The knowledge graph is the AI's conceptual model of the world. It's a network of interconnected entities: people, companies, products, concepts, and the relationships between them.

When an AI synthesizes an answer, it's not crawling the web in real-time. It's querying this internal model to find relationships it can trust. "Genrank" isn't just a word to the AI, it's an entity with attributes (software company, founded when, does what) and relationships (related to AEO, competes with X, created by Y).
Trust in the knowledge graph is established through consensus and clarity. The AI needs to be confident that an entity is what it claims to be, and that it's genuinely authoritative for the topics it covers. If the AI can't make these connections confidently, it defaults to better established entities even if they're less relevant to the specific query.

This explains a frustrating pattern we see in our data: pages that rank #1 organically but never appear in AI-generated answers. The page won the keyword game, but the brand hasn't won the entity game.

What we check for entity strength

In Genrank's entity analysis, we evaluate several factors that determine how clearly AI systems can identify and trust your brand.

Entity identification and coverage. We detect the primary and secondary entities present on a page, and flag when core entities expected for the topic are missing. If you're writing about a comparison but only define one side, or explaining a process without naming the key concepts, the AI has gaps to fill and it might fill them with someone else's content.

Entity clarity and disambiguation. We identify ambiguous terms or overloaded concepts that could confuse AI systems. If your content uses "AEO" without ever defining it, or switches between "answer engine" and "AI search" inconsistently, the AI's confidence drops. We recommend explicitly distinguishing similar entities: X vs Y, concept vs product, your definition vs the industry definition.

Entity-intent alignment. We map entities to generative intent types: definition, comparison, procedure, eligibility. If someone asks "what is X" and your page about X is structured as a sales pitch rather than a definition, there's a misalignment. We detect when the entities discussed don't match the likely user question.

Entity relationship analysis. We analyze how entities on a page are connected. this includes: part-of, used-for, depends-on, differs-from. AI systems expect certain relationships for certain topics. If you're explaining a concept without showing how it relates to adjacent concepts, we suggest adding or clarifying those relationships.

Knowledge graph alignment. We compare your on-page entities and terminology against commonly cited sources. If your terminology differs from how established sources discuss the topic, your content may not match what the AI expects. We score based on consistency with established conceptual models.

The entity cluster concept

A useful way to think about entity strategy is the entity cluster. The set of concepts that define your brand's topical authority.

Your core entity is your brand, product, or key thought leader. This needs to be defined with absolute clarity. The AI should have no ambiguity about what your company is, what it does, and what makes it distinct.

Related entities are the concepts, frameworks, and terms that your brand creates or is closely associated with. These form the cluster around your core authority. For a company in the AEO space, this might include specific methodologies, metrics, or frameworks you've developed.

Adjacent entities are the broader concepts in your space that you want to be associated with. These are the topics where you want the AI to consider you as a credible source, even if you didn't originate the concept.

The goal is to strengthen the relationships between your core entity and related entities, while building credible connections to adjacent entities. When someone asks about a topic in your cluster, the AI should see your brand as a natural, authoritative source.

Entity disambiguation in practice

The most direct way to communicate your entity to AI systems is through structured data, specifically, the sameAs property in JSON-LD schema.

The sameAs property links your entity to authoritative external sources that verify your identity. When the AI sees that your Organization schema links to your Wikipedia page, Wikidata entry, and official social profiles, it can confidently disambiguate you from other entities with similar names.

```
{
  "@context": "https://schema.org",
  "@type": "Organization",
  "name": "Your Company",
  "url": "https://yourcompany.com",
  "sameAs": [
    "https://en.wikipedia.org/wiki/Your_Company",
    "https://www.wikidata.org/wiki/Q12345",
    "https://www.linkedin.com/company/yourcompany",
    "https://twitter.com/yourcompany"
  ]
}
```

This creates a consensus signal. The AI can verify that the entity claiming to be "Your Company" on this page is the same entity recognized by Wikipedia, Wikidata, and LinkedIn. Confidence increases, and citation likelihood increases with it.

In our AEO scoring, we check for this disambiguation and flag when it's missing. We also recommend building Wikipedia and Wikidata presence if you don't have it. Google's Knowledge Graph incorporates data directly from both sources [2], making them high-authority signals that significantly improve entity recognition.

Consistent entity mentions

Beyond structured data, the AI builds trust by seeing entities mentioned consistently across your content and across the web.

This requires editorial discipline. If you've created a framework or metric, use the exact same name every time. If it's "Citation Value Score," don't sometimes call it "CVS" without first establishing that abbreviation. If it's "Answer Engine Optimization," don't switch to "AI search optimization" mid-article.

Inconsistency creates ambiguity, and ambiguity reduces confidence.

The same principle applies to how you reference other entities. When you mention a concept, be precise. When you reference research, cite it properly. When you discuss a competitor or adjacent player, use their canonical name. SparkToro's research on zero-click search [1] is valuable precisely because they've consistently owned that terminology. When people discuss zero-click trends, SparkToro is the entity the AI associates with that concept.

Entity pages as knowledge graph anchors

Every core concept in your entity cluster should have a dedicated page on your site. This page acts as the anchor for that entity in the knowledge graph.

The structure should be clear:

A definitive answer block early in the content. A concise definition or explanation in the first 1-2 paragraphs that the AI can extract directly.

Complete JSON-LD schema. Organization or Person schema with `sameAs` properties linking to authoritative sources.

Clear relationship mapping. How this entity relates to other entities in your cluster. What it's part of, what it's used for and how it differs from similar concepts.

Cited sources. Links to authoritative external sources that support your claims about the entity.

When the AI searches for information about a concept in your cluster, this page should be the single most authoritative source. Not because it ranks well, but because it clearly and unambiguously defines the entity and its relationships.

Internal linking as relationship programming

Internal links do more than help users navigate. They signal relationships to AI systems.

Every time you mention a related entity on your site, link to the page that defines it. This creates an explicit relationship signal: "When I mention Citation Value Score, I'm referring to this specific concept, defined here."

Over time, this builds a web of entity relationships on your own site that mirrors the structure you want in the AI's knowledge graph. The AI learns that your site treats these concepts as connected, and that you're a comprehensive source for the entire cluster.

What this means for content strategy

The shift from keywords to entities changes how you plan content.

Start with entity mapping, not keyword research. Before creating content, map the entities in your space. What are the core concepts? How do they relate? Which entities do you want to own, and which do you want to be associated with?

Create anchor pages for each entity. Every concept in your cluster needs a definitive page. These pages are the foundation of your entity authority.

Build relationship density. When you create new content, explicitly connect it to your entity cluster. Reference your core concepts, link to your anchor pages, and show how new topics relate to established ones.

Maintain consistency. Use the same terminology every time. Update your anchor pages when definitions evolve. Treat your entity cluster as a living system that needs maintenance.

Invest in external entity signals. Wikipedia, Wikidata, Google Knowledge Panel are useful external sources to validate your entity to AI systems. If you don't have presence on these platforms, building it should be a priority.

The keyword era rewarded content that matched search queries. The entity era rewards content that builds a coherent, trustworthy conceptual presence. The AI isn't looking for the best page on a topic, it's looking for the most trusted entity for that topic.

References

[1] SparkToro, "2024 Zero-Click Search Study: For every 1,000 US Google Searches, only 360 clicks go to the Open Web. In the EU, it's 374," https://sparktoro.com/blog/2024-zero-click-search-study-for-every-1000-us-google-searches-only-374-clicks-go-to-the-open-web-in-the-eu-its-360/

[2] Wikipedia, "Knowledge graph," https://en.wikipedia.org/wiki/Knowledge_graph. Notes that Google's Knowledge Graph incorporates "JSON-LD content extracted from indexed web pages, including the CIA World Factbook, Wikidata, and Wikipedia."

AI search is changing what it means to be "discoverable" on the internet. For two decades, content discovery meant ranking a page, winning a click, and converting that visitor on your site. Now, large language models and AI-powered search experiences increasingly resolve intent inside the interface, often without sending traffic to the original source. This is a structural shift from a Discovery Web (finding where information lives) to a Synthesis Web (getting the synthesised answer immediately). The new game is not just rankings and clicks, but whether your brand is understood, trusted, and included in the answers people consume.

Key Takeaways

• A majority of Google searches end without a click, which weakens "traffic" as the default unit of content value (Semrush/Datos Zero-Click Study, July 2024).
• AI Overviews expanded access rapidly, reaching over 1 billion global monthly users after rollout and international expansion (Google, Oct 2024).
• When AI summaries appear, click-through rates often drop materially, including a measured 34.5% lower CTR in one large dataset analysis (Ahrefs, Apr 2025).
• The optimisation target is shifting from keywords to conceptual proximity (embeddings and vector similarity), where "being close in meaning" becomes visibility (Wikipedia: Embedding (machine learning)).
• If you cannot measure how AI systems describe and cite your brand, you are operating with an incomplete view of demand. Genrank exists to close that visibility gap.

The Broken Social Contract of the Web

The old deal: publish → rank → click → monetise

The web ran on a simple exchange. Publishers invested in creating information. Search engines indexed it. Users clicked through. Publishers earned attention, leads, and ad revenue.

That contract shaped everything: how we wrote, how we formatted, how we measured success. "Traffic" became the proxy for impact because traffic was the mechanism that turned knowledge into business outcomes.

The collapse: when the answer becomes the interface

The contract breaks when the "answer" is delivered inside the search layer.

Google's AI Overviews are the most visible mainstream example. In May 2024, Google announced that AI Overviews would begin rolling out broadly in the U.S. (Google, May 2024).

By October 2024, Google said AI Overviews were expanding to 100+ countries and would have more than 1 billion global users every month (Google, Oct 2024).

The deeper issue is not one product feature. It is the redefinition of value.

If the interface resolves intent without a click, then the click stops being the unit of value.

The thesis: from a Discovery Web to a Synthesis Web

This is the shift:

• Discovery Web: "Where can I find the information?"
• Synthesis Web: "Give me the information now."

Information Foraging Theory is useful here. When the "cost" of finding information drops, user behaviour changes because people optimise for faster, lower-friction paths to answers (Nielsen Norman Group, 2019).

AI interfaces make the cost of an answer feel close to zero. That changes the path people take, which changes where visibility is created.

The Science of the Shift: From Index to Vector

The "legacy" way: an index is a library of addresses

Classical search is a masterpiece of indexing.

A simplified mental model: Google's index is a vast catalog of addresses. You type a query. The system retrieves and ranks documents that appear relevant, historically using signals like matching terms, link authority, and behavioural feedback.

The key point is that the core output is still a list of places to go.

The "neural" way: the web becomes coordinates

Neural information retrieval changes the representation.

Instead of treating pages as bags of keywords, AI systems encode text into embeddings, which are vectors in a high-dimensional space that preserve semantic relationships (Wikipedia: Embedding (machine learning)).

This is why the metaphor shifts from "addresses" to "coordinates."

Your site stops being just a URL. It becomes a set of vectors representing what you mean.

That changes ranking from "keyword match" to "vector match."

Cosine similarity is the new "ranking factor"

In vector search, similarity is commonly measured using cosine similarity:

cos(θ) = (A · B) / (‖A‖ ‖B‖)

If your content vector A is not close to the user intent vector B, you do not exist in that retrieval space.

Cosine similarity is a standard similarity measure for embeddings and vector representations (Wikipedia: Embedding (machine learning)).

This is not abstract theory. Dense retrieval systems and modern QA pipelines operationalise this. For example, Dense Passage Retrieval formalises document and query encoding into dense vectors for retrieval (Karpukhin et al., 2020).

The insight: you are ranking for conceptual proximity

In the Synthesis Web, you are no longer only competing for a keyword.

You are competing to be the closest meaning to the prompt.

That has two implications for content strategy:

Clarity compounds. If the internet consistently describes you the same way, your "conceptual coordinate" becomes stable.

Filler dilutes. If your content is verbose, generic, or inconsistent, you create semantic diffusion (and you drift away from the intent vectors that matter).

This is also where "knowledge distillation" becomes a useful metaphor. Distillation describes transferring knowledge from a larger model (or ensemble) into a smaller model (Hinton, Vinyals, Dean, 2015).

In practice, LLMs absorb patterns from the web and compress them into internal representations. You do not get to negotiate with that compression. You can only influence the training data and the public narrative the model observes.

The Rise of Answer Engine Optimisation (AEO)

AEO is the discipline that emerges when the output is no longer "ten blue links," but an assembled answer.

Genrank defines AEO as optimising content so AI systems can understand it, trust it, and cite it. That framing matters because it shifts the success metric from clicks to inclusion.

Search Engine Journal has also framed AEO tactically as building for AI citations and visibility, emphasising that these systems leave clues if you know what to measure (Search Engine Journal, Nov 2025).

AEO vs SEO: what actually changes

This matters more when traffic is structurally pressured. In the U.S., 58.5% of Google searches ended in zero clicks in 2024, and in the EU it was 59.7% (Semrush/Datos, July 2024).

The three pillars of AEO

Pillar 1: Directness (raise signal-to-noise)

AI systems extract. Humans skim. Both reward clarity.

This is why BLUF ("bottom line up front") works. It is one of the rare optimisations that improves machine readability and human comprehension.

Action steps (start with verbs):

• Name the question your section answers in the H2 or H3.
• State a 40–80 word direct answer immediately under the heading.
• List the key constraints, edge cases, or definitions in bullets.
• Expand with examples only after the answer is clear.
• Remove intros that do not change understanding.

Pillar 2: Corroboration (build third-party validation)

In classic SEO, links were votes.

In AI visibility, mentions are memory anchors.

When multiple independent sources describe your product in similar terms, AI systems gain confidence. That is why comparison posts, tutorials, and community explanations matter more than polished landing pages.

This is also why user behaviour data is pointing toward brand trust and multi-source validation. Forrester found that 89% of B2B buyers have adopted generative AI, with nearly 95% planning to use it in at least one area of future purchases (Forrester Buyers' Journey Survey, 2024).

If buyers are researching through AI, then the sources the AI trusts become your new distribution layer.

Pillar 3: Structured facticity (give models "training rails")

Structured data is not just for rich snippets anymore. It is machine-readable scaffolding.

Schema.org states that, as of 2024, over 45 million web domains markup pages with over 450 billion Schema.org objects (Schema.org, 2024).

Google's own documentation frames structured data as a way to help systems interpret content and enable search features (Google Search Central docs).

If you want to be legible to machines, this is the lowest-friction layer you control.

Action steps (start with verbs):

• Implement Organisation schema on your homepage and about page.
• Add Article/BlogPosting schema on blog content.
• Use FAQPage schema only when you truly have Q&A blocks.
• Validate your markup with Google's testing and Search Console workflows (Google Search Central docs).
• Standardize your product description string everywhere you publish (site, directory listings, founder interviews, docs).

For JSON-LD as a format, JSON-LD is explicitly designed as a Linked Data format that is easy for humans to write and machines to parse (JSON-LD).

The Measurement Crisis: The "Dark Matter" of Traffic

Traditional analytics are honest, but incomplete.

They tell you what happened on your site. They cannot tell you what happened inside the AI layer before a user ever arrives (or never arrives).

The scale of the click gap is already visible in search behaviour. Semrush/Datos data suggests the open web is receiving a minority share of outcomes when you look at searches that end without clicks (Semrush/Datos, July 2024).

At the same time, CTR pressure is measurable when AI summaries appear. Ahrefs found a 34.5% lower average CTR for top-ranking pages when an AI Overview was present, based on a 300,000 keyword analysis (as of April 2025) (Ahrefs, Apr 2025).

And independent reporting summarised by Search Engine Land points to CTR reductions ranging from 34% to 46% across studies when AI summaries appear (as of September 2025) (Search Engine Land, Sept 2025).

The visibility gap (what your dashboard cannot see)

Here is the uncomfortable scenario:

1. 10,000 people ask an AI system, "What's the best tool for tracking AI citations for content?"
2. The model answers confidently.
3. It either omits you or misrepresents you.
4. Your analytics show zero, because nobody clicked.

This is why I think of it as dark matter. The impact is real, but your instruments cannot detect it.

Gartner's February 2024 press release made this directional argument explicit, predicting traditional search engine volume would drop 25% by 2026 due to AI chatbots and virtual agents (Gartner, Feb 2024).

Whether the exact number lands is less important than the strategic message: measurement frameworks built for the click economy will underreport reality.

A working concept: Synthesised Visibility

I use "Synthesised Visibility" as a practical metric category:

• How often your brand is included in AI-generated answers for your category
• How accurately you are described
• How frequently you are cited (and which sources are used)
• Where you are absent (despite being strong in traditional SEO)

That is the visibility layer that traffic analytics miss.

Genrank: The Compass for the Synthesis Web

Genrank is not trying to be "another SEO tool."

SEO tools measure what search engines do with pages.

Genrank measures what AI systems do with meaning.

Our focus is the observation layer for the AI era: mapping the conversations you do not get to see, then turning them into actions you can take.

What Genrank is built to show

At a high level, Genrank helps you answer:

• How does AI describe Genrank today?
• Which sources are shaping that description?
• Where is the model confident, vague, or wrong?
• For which prompts are we included, excluded, or misclassified?
• Which third-party mentions move the needle?

This is the difference between "I hope we show up" and "I know where we stand."

How to use Genrank in a content-led go-to-market motion

Action steps (start with verbs):

1. List the 25–50 prompts that represent real buyer questions in your category.
2. Measure how often Genrank (and your competitors) appear in answers to those prompts.
3. Identify narrative gaps (missing features, wrong positioning, outdated comparisons).
4. Publish corrective content designed for AEO: direct answers, corroborated claims, structured data.
5. Earn third-party validation through comparisons, guest posts, directory profiles, and community discussions where your category is explicit.

The goal is not to "game" an LLM. The goal is to make the public internet unambiguous about what your product is, and to do it in the formats these systems can reliably extract.

Conclusion: The Land Grab for Mindshare

Platform shifts create land grabs.

Desktop to mobile reshaped distribution. Search to social reshaped attention. Now AI interfaces are reshaping discovery into synthesis.

The rule that is being rewritten in real time is simple: if you are not legible to answer engines, you can be present on the web and still be invisible in practice.

You can either become a footnote in an AI's training history, or the source it keeps reaching for because you are clear, corroborated, and structurally easy to trust.

If you want to see what the AI sees, join the Genrank waitlist and follow along as we build the visibility layer for the Synthesis Web:

If you want to see what the AI sees, join the waitlist and follow along as we build the visibility layer for the Synthesis Web.

A question I hear more and more is how ChatGPT decides which products to recommend. The short answer is that it doesn’t rank tools the way Google ranks pages. It learns from how the internet explains, compares, and contextualises products over time.

People treat ChatGPT like a search engine you can “rank” on, or worse, like another distribution channel you can hack with a checklist. That mindset usually leads to shallow tactics and disappointing results.

Large language models don’t work like Google. They don’t crawl the web looking for keywords to index. They learn by observing how humans explain things to other humans.

Once you internalise that, the path to turning ChatGPT and other AI engines into a high-intent conversion channel becomes much clearer.

Below is how I’ve seen it work in practice.

1. Make sure your product is everywhere people explain things (this is how AI engines discover products)

LLMs are trained on explanations, not landing pages.

They learn from tutorials, comparisons, walkthroughs, forum answers, long-form blog posts, and videos where someone is trying to genuinely explain how something works or why they chose one tool over another.

If your product consistently shows up in those contexts, you dramatically increase the chances that it will surface naturally in AI-generated answers when users ask ChatGPT for recommendations.

Think about the kinds of questions real users ask:

• How do you create birthday cards with [your product]?
• What is the best alternative to [your product]?
• How do I use [your product] to get more leads?

If your product appears in thoughtful, context-rich answers to those questions, AI models notice.

This is why educational content matters far more than promotional content.

Documentation, tutorials, and real use cases teach the internet how to talk about your product. Sales pages mostly teach it how to ignore you.

Beyond your own content, you should actively encourage third parties to write about your product in explanatory ways:

Comparison articles on independent blogs

• “X vs Y” breakdowns on major platforms
• Reddit threads where users discuss real pros and cons
• Product Hunt comments that go beyond launch hype
• YouTube videos that actually show the product in use

It’s less about chasing backlinks and more about teaching the internet what your product actually does and when it should be used.

2. Be consistent and authoritative with your language

Language consistency matters more for LLMs than it ever did for traditional SEO.

Models like ChatGPT are pattern-recognition machines. They build statistical associations between concepts, phrases, and entities. When you describe your product in ten different ways, you dilute those associations.

This is where many startups accidentally sabotage themselves.

They rotate taglines. They change positioning every few months. They describe the same feature with different language depending on the channel.

From a branding perspective, that already creates confusion. From an LLM perspective, it creates semantic diffusion.

You want one clear description of what your product is and who it is for—and you want to repeat it everywhere.

If your product’s core positioning is something like:

“Genrank helps marketers create content AI engines trust and cite”

And that phrasing appears consistently across documentation, blog posts, interviews, comparison articles, and community discussions, those words start to fuse together in the model’s internal representation.

This is not SEO writing. It’s closer to writing for statistical memory.

The goal is not keyword density. The goal is conceptual clarity.

3. Be available in public

This sounds obvious, but it’s often overlooked.

ChatGPT and other AI engines cannot read your private Slack threads, internal Notion docs, or closed-source repositories. They learn from public, crawlable content.

If your best explanations live behind a login, they might as well not exist.

You want your product discussed in places that are both public and conversational, where AI crawlers can observe real human explanations.

• Reddit
• Hacker News
• Product Hunt
• Public GitHub issues and discussions
• Blogs that allow indexing
• YouTube descriptions and transcripts

These environments matter because they contain natural language written for other humans, not for search engines.

When someone explains why they switched from one tool to another, or how they solved a specific problem using your product, that’s exactly the kind of signal LLMs learn from.

Private knowledge compounds internally. Public knowledge compounds externally.

If you care about AI visibility, you need both.

4. Make your product a topic of discussion by third parties

LLMs don’t just learn what your product is. They learn where it sits in an ecosystem.

This is why third-party comparisons are so powerful and why products that are frequently compared tend to be recommended more often by AI engines.

When your product is repeatedly mentioned alongside others in the same category, models begin to understand:

• What problem space you belong to
• Who your competitors are
• What differentiates you
• When you should or shouldn’t be recommended

Content like:

• “Genrank vs Otterly: which AI SEO tool should you use?”
• “Why Genrank works best for content marketers”
• “Top AI tools for content teams in 2025”

helps models triangulate your position.

The key is that this content should not all come from you.

When multiple independent sources describe similar trade-offs, the signal becomes stronger. Cross-references increase confidence.

From the model’s perspective, repeated agreement across sources looks like truth.

5. Keep your product information fresh

LLMs don’t “forget” in the way humans do, but they do learn distributions of meaning over time.

Every meaningful update to your product creates a new semantic anchor:

• New features
• New integrations
• New partnerships
• New use cases

If those updates are publicly documented and explained, they reshape how your product is understood.

This is why changelogs, release posts, and technical walkthroughs matter more than most teams realise.

Once a model associates your product with a certain capability, that association can persist for a long time. Updating the narrative requires repeated, consistent signals.

Silence is not neutral. It freezes perception.

From an AI perspective, each update is another data point that reinforces relevance and keeps your product present in the model’s understanding of the category.

Artificial intelligence has been through plenty of “next big things.” A decade ago everyone was talking about deep learning; by 2023 we were drowning in large language models. Now, in the middle of 2025, the discourse has shifted again.  This time the buzzword is agentic AI systems built from multiple models and tools that can plan, act and learn with minimal supervision.  As one Deloitte forecast noted, about a quarter of companies already using generative AI expect to pilot agentic systems in 2025, rising to half by 2027.  That sort of growth implies more than hype.  To understand why so many technologists and executives are paying attention, let’s take a deeper look at what AI agents are, how they are being used today and where the field is heading.

From Chatbots to Agents

The idea of an “agent” isn’t new.  Since the earliest days of AI, researchers have used the term to describe software that acts on behalf of a user.  In the 1990s, simple reflex agents responded to environmental stimuli using hard‑coded rules, while goal‑based and utility‑based agents introduced planning and preference functions.  What’s different today is the underlying technology: modern agents are built on top of large language models (LLMs) and can draw on external tools, memory and self‑improvement loops.  They don’t just chat; they perceive, reason, plan and act, and they can improve through feedback.

That autonomy distinguishes AI agents from chatbots.  A chatbot’s only job is to generate conversational responses.  A good large language model can summarize documents or answer questions, but it remains a passive component.  In contrast, an AI agent is designed to achieve goals in a complex environment.  It will break a task into sub‑steps, select the right tools, execute those steps and update its own state based on the results.  The MarkTechPost guide notes that agentic systems combine decision‑making, memory, multi‑step planning and tool execution.  This extra “agency” means they can be left alone to complete tasks and only occasionally report back.

This distinction matters when we talk about generative AI.  Many people use the term “agent” loosely, applying it to any AI software with a user interface.  Deloitte cautions that chatbots and co‑pilots, though impressive, “lack the degree of agency and autonomy that agentic AI promises”.  True agents choose their own actions to accomplish the user’s goal.  They can plan and adjust the plan mid‑execution, and they work with a persistent internal state.  Where a co‑pilot generates suggestions, an agent decides, acts and verifies.

Key Components of Modern AI Agents

To build such autonomy, developers assemble modules.  MarkTechPost breaks down an agent into seven key components: perception, memory, planning, tool use, reasoning, feedback loops and a user interface.  Perception covers the input interfaces text prompts, API calls, sensors that allow the agent to observe the world.  Memory comprises both short‑term context and long‑term storage, often implemented with vector databases or relational stores.  Planning and decision‑making modules use search algorithms, reinforcement learning or tree‑of‑thought approaches to map a route from the current state to a desired goal.

Tool use is what allows a model to go beyond language generation.  Using function calling or external APIs, an agent can fetch data, perform computations, write files, send emails or invoke other services.  In 2025, widely used frameworks like LangChain, AutoGen and Microsoft’s Semantic Kernel provide standardised interfaces for connecting to these tools.  Reasoning and control logic tie the modules together; they evaluate the outputs of perception and memory modules, decide on the next action and route requests to the appropriate tool.  Finally, feedback and learning loops allow agents to assess how well they are doing and adjust accordingly.

If this sounds familiar, that’s because the structure echoes classical AI textbooks.  What’s new is the combination of LLMs for language understanding and synthesis with vector memory, search, planning and tool interfaces.  This modularity has given rise to a cottage industry of frameworks aimed at reducing the friction of building and deploying agents.

Frameworks and Tools: The Agentic Ecosystem

By mid‑2025 there are dozens of platforms and libraries for building agents.  MarkTechPost lists several, including LangChain, Microsoft AutoGen, Semantic Kernel, OpenAI Agents SDK, SuperAGI, CrewAI and IBM watsonx Orchestrate.  Each has its own philosophy and strengths.  LangChain dominates open‑source development and provides primitives for composing chains of prompts, memory stores and tool calls.  It integrates with numerous LLM providers (OpenAI, Anthropic), vector stores (FAISS, Weaviate), and external APIs.  AutoGen, by contrast, focuses on multi‑agent orchestration; it defines roles like Planner, Developer and Reviewer that can collaborate on code automation tasks.

Semantic Kernel offers an enterprise‑grade toolkit that abstracts LLMs behind “skills” and planners and is language‑agnostic.  OpenAI’s Agents SDK (often referred to as Swarm) provides light‑weight constructs for defining agent behaviors, tools and guardrails with integrated monitoring.  SuperAGI positions itself as an operating system for agents, offering persistent multi‑agent execution, memory handling and a marketplace for plug‑and‑play components.  CrewAI emphasises team‑style orchestration by letting developers assemble specialised roles like Planner, Coder and Critic that communicate and hand off tasks.  IBM’s Watsonx Orchestrate takes a no‑code approach: it is a SaaS platform that business users can drag‑and‑drop into workflows.

The ecosystem is evolving quickly.  There is also an emerging taxonomy for multi‑agent architectures.  InfoServices’ deep‑dive into LangChain’s multi‑agent system describes four core agent types: Planner, Executor, Communicator and Evaluator.  The Planner decomposes a high‑level goal into subtasks and sequences them.  Executors perform specific actions like retrieving documents, generating code or translating content.  Communicators facilitate handoffs between agents and ensure context is preserved.  Evaluators check outputs for correctness and may trigger retries or route tasks to different agents if needed.  Above these agents sits an orchestration layer that can design graph‑based workflows, route tool calls and manage memory.  Features such as dynamic tool routing, context sharing, asynchronous execution and error recovery flows make these multi‑agent systems robust and adaptable.

Use Cases and Real‑World Impact

What can agents actually do?  Practical use cases span a remarkable range of industries.  The MarkTechPost article highlights several categories.  In enterprise IT, AI agents act as helpdesk assistants, triaging tickets and diagnosing issues.  Tools like IBM’s AskIT reportedly reduce support calls by 70%.  Customer support and sales agents integrate with CRMs and knowledge bases to handle common inquiries, manage returns and recommend products; some e‑commerce bots have cut support costs by 65% and increased lead volume by 50%.  Contract and document analysis agents extract and summarise legal and financial documents, cutting review time by up to 75%.

In e‑commerce, agents monitor inventory, predict demand and help shoppers find products using visual search.  Logistics companies like UPS save hundreds of millions of dollars annually by using AI route optimisation systems.  Human resources and back‑office workflows are another fertile ground: digital HR agents handle routine queries, process payroll and manage invoices.  Meanwhile, research agents summarise reports, fetch relevant insights and build dashboards, and AI assistants in developer tooling accelerate code generation and testing.

One high‑profile example is Devin, an autonomous software engineer introduced by Cognition in March 2024.  Deloitte describes Devin as an agent capable of reasoning, planning and executing complex programming tasks.  Devin turned natural language descriptions into working code, tested and fixed bugs and even trained models.  Early benchmarks showed Devin could resolve about 14% of real‑world GitHub issues, outperforming standard LLM chatbots by a factor of two.  While far from replacing human developers, such systems demonstrate the potential for agentic AI to automate multi‑step workflows.

Adoption, Hype and Reality

With new frameworks and successful pilots, it’s easy to get swept up in excitement.  Deloitte’s 2025 prediction that 25% of gen‑AI‑using companies would experiment with agentic AI, and that this figure would double by 2027, underscores the momentum.  Investors have poured more than $2 billion into agentic AI startups over the past two years.  Companies are not only building new tools but also acquiring startups and licensing technology to accelerate their programs.

However, there are reasons to temper expectations.  Traditional chatbots and co‑pilots already struggle with hallucinations and unpredictable behaviour.  Agents layer planning and tool execution on top of those same language models.  As Deloitte notes, the “autonomous” part of agentic AI may take time for wide adoption.  In fields like software engineering, early agents like Devin show promise but still make too many mistakes to be trusted with complex tasks without human oversight.  Agents must also cope with edge cases, ambiguous instructions and dynamic environments.  As MarkTechPost points out, emergent benchmarks such as AARBench, AgentEval and HELM are being developed to measure how well agents handle tool use, long‑term memory and holistic task execution.

Another challenge lies in security and reliability.  Agents that can call external APIs or execute code may inadvertently produce harmful outputs if not properly sandboxed.  The MarkTechPost guide emphasises the need for improved planning algorithms, multi‑agent coordination, self‑correction mechanisms and secure tool sandboxes.  Enterprises must also consider issues around data privacy, compliance and auditability when deploying agents that read contracts or access sensitive systems.  Additionally, there is the human factor: knowledge workers may be sceptical about delegating tasks to autonomous systems, especially when mistakes could have significant consequences.

On the cultural side, agentic AI has sparked debate about the role of automation in knowledge work.  Some view agents as productivity multipliers that free humans to focus on creativity and complex decision‑making.  Others worry about job displacement and deskilling.  The truth likely lies somewhere in between: as with previous waves of automation, agentic AI will change the nature of some jobs while creating new roles (for example, prompt engineers, agent orchestrators and AI ethicists).  For now, most organisations treat agents as assistive tools rather than autonomous employees.  Humans remain in the loop, setting goals, reviewing results and handling exceptions.

Building with Agents: Best Practices

If you are exploring agents as a developer or architect, there are a few principles worth keeping in mind:

Start with clear objectives and scope.  Agents excel when their goals are well defined.  Before integrating them into your workflow, articulate what success looks like and how you will measure it.  Use existing benchmarks and metrics like task completion rate, cost per task and error rate to compare different designs.

Choose the right framework for your needs.  LangChain is a good starting point if you want flexibility and a large ecosystem.  AutoGen or CrewAI might be better if you need multi‑agent collaboration.  For enterprise settings, Semantic Kernel or IBM Orchestrate offers stronger governance and integration options.

Design for observability and safety.  Instrument your agents to collect logs, track tool calls and capture state transitions.  Implement guardrails such as timeouts, cost limits and input sanitisation.  Where possible, keep humans in the loop for critical decisions.

Invest in memory management.  Decide what information should persist between sessions and how to store and retrieve it efficiently.  Vector stores like FAISS or Weaviate are popular choices, but relational databases can be used for structured data.

Anticipate failure modes.  Build mechanisms for error recovery retry strategies, fallback prompts and alternate agents as suggested by the InfoServices article.  Evaluate your agents on edge cases and adversarial inputs.

The Road Ahead

Agentic AI is still in its infancy.  Yet the trajectory is clear: as models improve, memory grows cheaper and planning algorithms become more sophisticated, agents will take on more complex tasks.  MarkTechPost lists research directions like graph‑of‑thoughts planning, multi‑agent coordination, persistent memory, role guardrails and self‑correction strategies.  We can already see these ideas manifest in frameworks like LangGraph, which uses directed graphs to model agent interactions, and evaluation suites like AgentBench and HELM.

Another interesting development is the move toward specialised agents.  Rather than building a monolithic generalist, developers are creating swarms of narrow agents that collaborate.  For example, a reading assistant might comprise one agent for summarisation, another for fact checking and a third for question answering.  The InfoServices architecture emphasises decoupled agents with specialised roles.  This modularity aligns with the microservices paradigm in software engineering and makes systems easier to scale and maintain.

Finally, the business impact will depend on how organisations manage the socio‑technical transition.  Companies that successfully deploy agents will treat them as part of a broader digital transformation strategy rather than a magic bullet.  They’ll invest in training, change management and robust governance.  And they’ll recognise that agentic AI is not about replacing humans but about augmenting them freeing us from repetitive tasks so we can tackle the problems that still require creativity, empathy and domain expertise.

AI agents represent a compelling evolution in software design, combining the generative power of large language models with the autonomy of classical AI.  They are not just glorified chatbots; they perceive, plan, act and learn.  Frameworks like LangChain, AutoGen and Semantic Kernel make it easier than ever to build such systems, while multi‑agent architectures add robustness and scalability.  The use cases are expanding from IT support and sales to legal analysis, logistics and software development.  Early successes like Cognition’s Devin showcase both the promise and the limitations of today’s agents.

Yet the road to widespread adoption will be long.  As Deloitte points out, only a fraction of companies will run pilots in 2025, and reliability, security and governance challenges remain.  Researchers are working on better planning algorithms, evaluation benchmarks and self‑healing mechanisms.  Meanwhile, business leaders must navigate the human factors of trust, transparency and job design.

For those building the next generation of intelligent systems, the advice is simple: stay grounded in real‑world needs, embrace modular design, invest in safety and never lose sight of the human at the centre.  Agentic AI is an exciting frontier but like any frontier, it requires careful exploration.

If you’ve ever had a great idea for a startup, you know how exciting it can be. You start envisioning the product, imagining how it’s going to change the world—or at least a small corner of it. But here’s the thing I’ve learned from building several startups: it’s incredibly easy to get carried away and jump straight into building without first confirming whether your idea will actually solve a real problem for real people.

That’s where validation comes in. I’ve made this mistake before—jumping the gun without validating—and it almost always leads to wasted time, money, and effort. Over time, I’ve developed a simple but effective process to make sure my idea has legs before I commit to building it. Here’s how I do it.

1. Talk to Potential Customers (And Actually Listen)

This might sound obvious, but the first step I take is talking to potential customers. You’d be surprised how easy it is to skip this part or assume you already know what your customers want. In the past, I’d get so excited about my idea that I’d bypass these early conversations. Big mistake.

Now, I make sure to speak with people who would actually use my product before I even think about building. And the key here is listening, not selling. These conversations are not about convincing people that your idea is great—it’s about understanding their pain points and seeing if your solution is a good fit.

I usually start with friends or acquaintances in my target audience, asking open-ended questions like:

• “What’s the most frustrating part of [the problem area] for you?”
• “How do you currently solve this issue?”
• “If there was a tool that did X, would you use it?”

This process has saved me countless times. I remember once thinking I had a foolproof idea for a time management tool. After a few conversations with potential users, I realised my product wasn’t addressing their actual frustrations. They didn’t need another tool; they needed something simpler that integrated into tools they already used. Those early chats helped me pivot and come up with a version of the product that actually resonated with my audience.

2. Research the Market (Because You’re Probably Not the First)

It’s tempting to think your idea is one-of-a-kind, but more often than not, someone out there has already had a similar thought. Instead of letting that discourage me, I’ve learned to embrace it. If there are already similar products on the market, that’s a sign that there’s demand. What I focus on is finding my unique angle.

I start by searching for competitors. I use tools like Google Trends to see what’s popular and browse Product Hunt to discover products that are in the same space. This helps me figure out what the market looks like—who the key players are, what they’re doing well, and what gaps exist that my idea can fill.

I once had an idea for a niche social networking platform. After a bit of research, I discovered not only a few direct competitors but also several bigger players who had tried and failed in the space. Rather than being discouraged, I dug into the reasons they failed, and it helped me refine my concept to avoid making the same mistakes.

Even if the market seems crowded, there’s almost always room for improvement. Maybe your product can offer a better user experience, or perhaps you can cater to a niche that’s been overlooked. This research helps me zero in on what’s going to set my idea apart.

3. Build a Landing Page (Without Building the Product)

One of the best ways I’ve found to gauge interest in an idea is by creating a simple landing page. It’s a low-effort, high-reward tactic. I don’t need to build a full product—just a single page that clearly explains the value of the idea. If I can get people interested enough to sign up for early access or share their email address, I know I’m onto something.

Here’s how I do it: I create a basic landing page with a clear value proposition, a few visuals (mockups if I have them), and a strong call to action. The goal here isn’t to sell a product; it’s to see if people are intrigued enough to take the next step. Usually, I’ll include an email signup form for updates or early access.

Once the page is live, I’ll drive some traffic to it using social media, Reddit, or low-cost ads. I set a small budget—just enough to get a few hundred clicks—and see how many people convert. If I’m getting solid interest, it’s a green light to move forward.

I’ve done this with several projects, and each time it helps me get a feel for how my message is resonating. I remember one time when I thought I had nailed the concept for a new app, but after running the landing page, the signup rate was abysmal. I quickly realised that the way I was framing the problem didn’t connect with users, so I reworked the messaging and tried again. Huge difference.

4. Start with an MVP (Not the Full Product)

In the early days, I used to fall into the trap of trying to build a full-fledged product right out of the gate. Over time, I learned that it’s much better to start with a minimum viable product (MVP)—the simplest version of your product that still solves the core problem.

The MVP approach has been a game-changer for me. Rather than spending months perfecting something, I now aim to get a rough version of the product out there quickly. This way, I can collect feedback from real users and iterate based on their needs.

For example, I once built an MVP for a project management tool with only a few core features—task tracking and basic collaboration. It wasn’t fancy, but it allowed me to test my assumptions. And guess what? Users didn’t care about half the features I thought were essential. They wanted better integration with existing tools and a cleaner UI, which shifted my focus when building the next version.

When building an MVP, I focus on solving the primary pain point as simply as possible. Whether it’s using no-code tools, wireframes, or basic prototypes, the goal is to test, learn, and iterate.

5. Crowdfunding or Pre-sales (The Ultimate Test)

One of the most effective ways I’ve found to validate an idea is by asking people to pay for it before it exists. Sounds scary, right? But it’s an incredible way to measure real demand.

Platforms like Kickstarter or Indiegogo have been instrumental for me when validating product ideas. By launching a crowdfunding campaign, I can present the concept to a broader audience and see if they’re willing to put their money where their mouth is.

I’ve also tested pre-sales on my own site by offering early-bird pricing for a product that’s still in development. This approach not only gives me validation but also helps fund the project before I’ve even built it.

I once launched a pre-sale for an app idea that I wasn’t entirely sure would take off. Within a few weeks, I had enough pre-orders to cover the initial development costs, which gave me both the financial backing and the confidence to move forward. Pre-sales or crowdfunding can be an incredible motivator because they show that people are truly invested in your solution.

6. Get Social Proof & Feedback from Communities

Whenever I’m testing a new idea, I turn to online communities for feedback. Whether it’s Reddit, Indie Hackers, or specialised forums related to the industry, sharing my concept with others always helps me refine it. It’s a great way to get social proof and see how people respond to the idea.

I usually frame my posts in a way that invites constructive feedback: “I’m working on this idea for a [product/solution], and I’d love to hear your thoughts on it.” The responses are often honest and unfiltered, which is exactly what I need to improve the concept.

What’s cool about sharing in these spaces is that you also start to build an early community around your idea. I’ve had people who were part of these discussions become some of my first users once the product was live. And the best part? These communities are usually filled with other startup founders, so they get what you’re going through and want to help.

Conclusion: Validate First, Build Smarter

Validating your startup idea before building doesn’t have to be a complex, time-consuming process. It’s about being intentional with how you approach your idea—by talking to potential customers, researching the market, creating a simple landing page, and testing with an MVP or pre-sales. These steps have saved me from building products that no one wanted and helped me focus on ideas that actually resonate with people.

Trust me, the validation process will get you even more excited about your idea. You’ll have real feedback, real interest, and perhaps even real customers before you’ve written a single line of code. So, take the time to validate—it’s worth every second. And once you’ve got that validation, you’ll feel confident diving into the next phase of your startup journey. Happy building!

In the sprawling digital landscape, the importance of a professional photo – capturing not just your features, but your essence – cannot be overstated. Traditionally, this would involve studio lights, a professional photographer, and a rather hefty fee. However, the digital domain is abuzz with the launch of Magic Lens, a web application that'll improve your profile photos in minutes.

At its core, Magic Lens bridges the chasm between selfies' casual nature and professional headshots' polished elegance. Using a harmonious blend of advanced technologies, it provides users with the unparalleled convenience of obtaining high-end headshots and all this from the comfort of their homes.

The tech behind the magic:

• NextJS: Driving user experience to unparalleled heights, ensuring a smooth and responsive interaction every step of the way.
• LeapAI: The mastermind, expertly refining and transforming your selfies into captivating headshots.
• Supabase: Ensuring the platform's wheels turn without a hitch, managing data flows seamlessly.
• Stripe: Keeping your transactions streamlined and secure.

Simplicity in Action:

Start with your selfies. Just remember, clarity is key. Four good-quality, front-facing selfies without accessories like hats or glasses will do the trick. Once uploaded, the state-of-the-art LeapAI steps into the scene, meticulously crafting your professional headshots. You don't need to hover around either; a notification email will let you know once your photos are primed and ready. Finally, marvel at the transformation and download your brand-new headshots.

And for those curious about the results? Just glimpse at the sample to gauge the prowess of Magic Lens.

As the digital realm becomes more intertwined with our daily lives, the line between the physical and virtual continues to blur. In this dynamic environment, Magic Lens stands out as a beacon for those seeking to elevate their online presence. It's not just a tool; it's a statement – one that says you value quality, sophistication, and the magic of innovation.

As the app's launch gathers momentum, early adopters are already raving about its results. So, if you've been contemplating that perfect headshot but have been deterred by the logistics, it might be time to give Magic Lens a whirl. After all, why settle for ordinary when magic is just a click away?

As someone who has always been fascinated by artificial intelligence and natural language processing, I was excited to hear that OpenAI had given developers the ability to build plugins on top of ChatGPT. It's a simple ChatGPT plugin for managing a to-do list. Check it out on GitHub.

As artificial intelligence and natural language processing continue to advance, ChatGPT has emerged as a powerful tool for generating human-like responses to text inputs. With ChatGPT plugins, developers can integrate this technology into various platforms and applications, enhancing user experiences and increasing accessibility. In this blog post, we will explore what ChatGPT plugins are and how to create one.

What are ChatGPT plugins?

ChatGPT plugins are third-party integrations that allow developers to incorporate ChatGPT's capabilities into their applications. These plugins can be used to enhance chatbots, language learning platforms, productivity tools, and more. With a ChatGPT plugin, developers can provide users with a more engaging and personalised experience by leveraging the language generation capabilities of ChatGPT.

How to create a ChatGPT plugin?

Creating a ChatGPT plugin requires some technical knowledge, but it is an exciting and rewarding process. Here are the steps to create a ChatGPT plugin:

Step 1: Set up a ChatGPT API account

To get started with a ChatGPT plugin, you need to set up an account with the ChatGPT API. This will give you access to the tools and resources you need to integrate ChatGPT into your application. Once you have created an account, you will be given an API key that you can use to authenticate your requests to the ChatGPT API.

Step 2: Choose your development platform

ChatGPT plugins can be developed on a variety of platforms, including web applications, mobile apps, and desktop applications. Choose the platform that is best suited for your needs and ensure that you have the necessary tools and resources to develop your plugin.

Step 3: Integrate ChatGPT into your application

Integrating ChatGPT into your application involves sending requests to the ChatGPT API and receiving responses that can be displayed to users. To do this, you will need to use a programming language that can send HTTP requests and handle JSON data. Python, JavaScript, and Ruby are popular programming languages for developing ChatGPT plugins.

Step 4: Test and refine your plugin

Once you have integrated ChatGPT into your application, you should test your plugin to ensure that it is functioning as intended. You may need to make adjustments to your code and settings based on your testing results. Refining your plugin is an iterative process that involves making small changes and testing them until you achieve the desired functionality.

Step 5: Deploy your plugin

Once you have tested and refined your plugin, you can deploy it to your application. Depending on your platform, this may involve uploading your plugin to an app store or hosting it on a web server.

So, what are the benefits of using ChatGPT plugins?

Here are just a few:

1. Enhanced customer service: By incorporating ChatGPT into customer service chatbots, businesses can provide more personalized and efficient support to their customers. ChatGPT can quickly analyze a customer's inquiry and provide a relevant response, saving both time and resources.
2. Improved language learning: With ChatGPT plugins, language learners can practice their skills in a more interactive and engaging way. ChatGPT can provide feedback on grammar, vocabulary, and sentence structure, helping learners to improve their language proficiency.
3. Increased accessibility: ChatGPT can be integrated into various platforms, making it more accessible to a wider range of users. This includes people with disabilities who may have difficulty typing or using traditional input methods.
4. Better productivity: ChatGPT can assist with a variety of tasks, from scheduling appointments to generating reports. This can help individuals and businesses save time and increase productivity.

Overall, ChatGPT plugins offer a range of benefits and can be used in a variety of ways. I am excited to continue exploring this technology and developing new plugins that can help people in even more ways.

Thank you for joining me on this journey, and I look forward to sharing more updates with you soon.

P.S. This post was generated with ChatGPT 😉

I've been learning web3 programming via Alchemy University and got my introduction to Solidity. Here's a post to explain the language to beginners.

Solidity is a programming language that is specifically designed for the development of smart contracts on the Ethereum blockchain. It is similar to other programming languages such as JavaScript and C++ and allows developers to write self-executing contracts with certain pre-defined rules and regulations.

One of the main advantages of Solidity is its ability to enforce the rules of a contract automatically, without the need for a third-party intermediary. This makes it a popular choice for creating decentralised applications (dapps) and other blockchain-based projects.

Before getting started with Solidity, it's important to have a basic understanding of the Ethereum blockchain and the concept of smart contracts. A smart contract is a program that runs on the blockchain and automatically executes when certain conditions are met. They can be used for a variety of purposes such as creating digital assets, managing supply chains, and automating financial transactions.

When writing a Solidity contract, it's important to keep in mind that the code will be executed on the blockchain and therefore, any errors can have severe consequences. This is why proper testing and error handling are crucial when developing smart contracts.

One of the first things to know about Solidity is that it is a strongly typed language, which means that variables need to be declared with a specific data type such as uint (unsigned integer) or address. Additionally, Solidity has its own set of data structures such as mappings and arrays.

A contract in Solidity is composed of several elements:

• State Variables: These are the variables that hold the current state of the contract. They can be of any data type and can be accessed from anywhere within the contract.
• Functions: These are the methods that can be called on the contract. They can be used to modify the state of the contract or perform other actions.
• Events: These are similar to functions, but they are used to notify external parties of certain actions that have occurred within the contract.
• Modifiers: These are special keywords that can be used to add additional functionality to functions. For example, the onlyOwner modifier can be used to restrict the ability to call a function to the contract's owner.
• Constructor: This is a special function that is executed when the contract is deployed to the blockchain. It is used to initialise the state of the contract and set any necessary values.
• Inheritance: This feature allows a contract to inherit properties and functionality from another contract, so developers can use pre-existing code and avoid duplicating effort.

It's also important to keep in mind that Solidity code is executed on the Ethereum Virtual Machine (EVM) and that each action on the contract will cost a certain amount of gas. Gas is the internal pricing for running a transaction or contract on Ethereum. To optimise the cost, developers need to be mindful of the gas usage in their code.

Solidity is a powerful programming language that allows developers to create self-executing smart contracts on the Ethereum blockchain. It has its own set of data types, data structures and features that make it unique and useful for decentralised applications. However, as with any new technology, there are some complexities and considerations that developers must keep in mind when working with Solidity.

Companies like Facebook and recording artists like Billie Eilish have promised to make changes to bring their greenhouse-gas emissions closer to net zero. The race to zero is a vital step toward managing climate change. But what does net zero really mean and is achieving it even possible?

Humans burning fossil fuels has resulted in more carbon dioxide in the atmosphere which is warming the planet. To stop the warming the level of greenhouse gases in the atmosphere has to stop rising. The obvious way to do that is to stop emitting them. But that is easier said than done. For some industries such as aviation and manufacturing eliminating emissions is really hard. In the years leading up to the Copenhagen climate conference in 2009, scientists realised something. It wasn’t possible to cut emissions fast and thoroughly enough to meet the temperature targets that policymakers wanted. What was needed was to actively remove greenhouse gasses from the atmosphere too. People began to talk about a world in which greenhouse-gas emissions and greenhouse-gas removals balanced each other out. So that the overall effect was net zero.

The countries that signed up to the Paris Agreement pledged to turn this idea into reality by agreeing to balance their emissions and removal of greenhouse gases in the second half of the century.

Meeting the net zero target requires two things. The first thing is to cut our greenhouse gas emissions and the second is to remove emissions from the atmosphere. The removal part is the hardest and sometimes more obscure aspect of reaching net zero.

At the moment we have different ways to remove CO2 from the air. Trees capture carbon from the air naturally but to achieve our ambitious targets, we need man-made solutions.

With the new carbon capture technologies, we can capture carbon dioxide and store it underground. But we don't know if this can work at the scale required to meet our targets. The world is counting on these innovations and that's a risky bet.

Who's responsible for each molecule of greenhouse gas?

A lot of countries and companies don’t want to own up to their carbon footprint. For example, carbon-intensive countries like India, China or other emerging markets are producing enormous amounts of emissions today. They point out that the goods they produce, for example, may be consumed by Americans and Europeans. So they should do the negative emissions to they may say rich countries got rich putting carbon dioxide into the air. Now it’s our turn to lift our people out of poverty so you pay for the negative emissions.

As yet there is no universal policy for accounting for and attributing emissions. Today net-zero governmental pledges cover over two-thirds of the global economy. America and the EU are working towards a target of net zero by 2050. President Xi of China, the world’s largest emitter has pledged to achieve carbon neutrality before 2060.

Intro
Companies are driven by money, so you'd correct in thinking that putting a price on carbon emissions should encourage businesses to stop polluting, right? That's exactly what carbon markets were designed to do. Reduce emissions by charging polluters but so far they haven't achieved that aim.

Birth of carbon markets
Towards the end of the 1980's America had a problem. For years its power stations had been emitting large amounts of sulphur dioxide - which falls back to earth as acid rain; causing harm to plants, aquatic animals and infrastructure. But there were no incentives for the power plants to stop emitting sulphur dioxide. So in 1990 the American Government launched an experiment, passing a law to force polluters to pay for their emissions by establishing a new kind of market governed by a system called "cap-and-trade". Eight years later, acid-rain levels over large regions in Eastern America had fallen by 20%. And a new way of cutting emissions was born.

How a carbon market works
In 1997 the international climate-change treaty known as the Kyoto protocol suggested applying the concept of cap-and-trade to carbon. In the years that followed different countries and regions set up their own carbon markets. Many of these used cap-and-trade and this is how it works.

A government sets a cap on the amount of CO2 that can be emitted by an industry, it splits the cap into permits and either gives or sells these permits to firms. If a company doesn't use up all of its allowance it can sell what it doesn't need. If it needs more permits, it can buy them from those with spares. Each year the cap gets stricter and the shrinking pool of permits gets more expensive.

Market failure
While regulation can introduce a new industry standard, it doesn't give firms an incentive to cut emissions below a certain level. But a carbon market creates a race in which companies are motivated to cut emissions as fast as they can. The more they cut emissions, the fewer permits they have to buy and the more excess that have to sell. So in theory in a cap-and-trade market carbon dioxide emissions should fall but in reality they've continued to rise because incentives only work if they're big enough.

According to the economists Joseph Stiglitz and Nicholas Stern, in order to meet the Paris Agreement goal of limiting global warming to 2 degrees above pre-industrial levels the price of carbon worldwide needs to be between $50 and $100 per tonne by 2030.
However the majority of carbon prices still remain far below that figure. What's more, even if carbon is priced appropriately the fines for exceeding permitted levels are sometimes ineffectively low. In the EU a fine can be as low as €100 per excess tonne. Considering that's not that much more than the price of a permit it's hardly a deterrent. And that's if firms even get caught.

Carbon leakage
Carbon markets are regulations are very challenging. There are measurement problems, direct emissions versus indirect emissions. Regulations enforcement and punishments are lax in some markets. In regions where the markets adopt effective deterrents, the ones next door may not.
So for a company based in multiple regions, divisions of that company making the same product in the same way could face a patchwork of regulations that differ. Worse than that you have leakage among these markets as well.
The direct result of this patchwork of systems is known as "carbon leakage". A business or an industry relocates from an area with high environmental regulations to somewhere where the rules are more relaxed. Which means it can avoid having to pay its carbon emissions. But these problems have solutions.

Solutions
If Governments limited the number of permits, it would drive their price up. Setting a minimum price that rises over time would mean the price would never fall too low and Governments could enforce more stringent deterrents for potential rule-breakers.

Another solution is to have more integrated carbon markets. But as that's unlikely to happen in the short term, one solution to help prevent carbon leakage is a border tax.
The EU has recently proposed to tax the carbon emitted in goods produced outside its carbon market. Importers will have to pay the same amount as if the goods have been made in the EU. Meaning it wouldn't be any cheaper to source goods from a region with less regulation.

As climate change has moved up the political agenda, Governments are starting to improve how their carbon markets work. Since 2019, the EU has been taking steps to reduce the number of permits it hands out. Partly as a result, carbon prices in the EU are now hitting record highs of over €60 per tonne. Carbon prices in other markets are also rising as regulators look at ways to make them more effective.
If prices stay high enough helped by commitments from Governments and regulators then greener industrial processes would become more attractive. And carbon markets could start to achieve their original goal of helping to decarbonise the world.

Credit:  The Economist

1 - Turn kid cast-offs into breakfast

If you’re a parent of a small child, you are likely very familiar with the apple with two bites out of it and the half-eaten banana. Rather than toss these into the compost, you might want to use a paring knife to cut off the bitten-off portion and then store the rest into the freezer for future smoothies—win-win!

2 - Think like a restaurant owner

Restaurants tend to practice a FIFO “First In First Out” system. You probably don’t have a walk-in cooler. However, the trick is basically to remember that when you’re unpacking your groceries, you consciously put your newly purchased foods behind the older ones.

3 - Upcycle those beautiful glass jars

A lot of people will buy new glass storage containers when they focus on reducing plastic at home, but you can upcycle glass jars. Soak your containers in hot water, scrape with a glass scraper, and you’ve got yourself a free and waste-free place to store food.

4 - Try composting your food waste

Start by storing your fruit and veg scraps in a bag in the freezer for a week and bring them to your local composting program (often at community gardens). If you have the space, you might want to consider graduating to a home compost bin.

5 - Buy your produce “naked”

Try to buy most of your produce without any packaging, because it’s not just grocery and shopping bags that are the problem: it’s all the bags, including produce bags. You’re going to wash everything at home anyway, so you shouldn’t be worried about placing your produce straight into your basket.

6 - Learn a back-pocket recipe for lonely vegetable

Making stir-fry and soups are a great way to use up random vegetables that are too small to make a side dish on their own. Put together a half head of broccoli, a lone carrot, half a bell pepper, and a leek and you’ve got yourself a stir-fry—and almost any root or shoot can get tossed into a soup pot.

7 - Cook Down The Pantry

Make a conscious effort to work my way through the non-perishables in your pantry so that nothing goes to waste. A couple of kid-friendly ideas to use them up are clean-out-the-pantry granola and multi-grain pancakes. You can easily substitute half the flour in a waffle or pancake recipe (think Mark Bittman’s basic pancake recipe) for the random whole grain you want to use up.

8 - Try Low-Waste Cleaning

DIYing your cleaning products is a noble endeavour (Cameron has tons of recipes in her book Simply Sustainable). A less wasteful solution has been to seek out concentrated cleansers that you add water to at home, reducing the packaging waste to almost zero.

9 - Cut Back On Paper Towels

Two things that really helped: Swedish-style dishcloths, which can be used to clean up spills; and old T-shirts cut into rags, but you don’t have to go cold turkey: To start, just place your roll of paper towels in a less convenient location, like under the sink.

10 - Save Every Little Leftover

Most people throw out their leftovers if they have less than a serving left, but you can save everything small leftovers and assemble them into a curated selection just like you would at a salad bar. Picture a small scoop of roast beets, a dollop of hummus, a piece of whole-wheat toast, and a handful of salad greens tossed with a little dressing. Teeny portions are great lunch items for both kids and adults.

Here are some easy tips you can use to keep sustainable when working from home.

Unplug devices

Even if an electronic device is not actively being used, if it is left plugged in, then the appliance continues to use energy. This draws energy from the grid, which puts unnecessary strain on the environment.

“A simple way to avoid the unnecessary strain on the environment is to unplug all electronics which you aren’t actively using, including chargers, printers and even your television,” Thomas Goodman advises. “It is estimated that the average home has 40 idle products which are constantly plugged in, making up as much as 10% of household energy use. By unplugging these appliances, not only can you decrease your household energy use by that 10%, but you could save up to £30 a year in electricity bills.”

Adjust the heating

As summer is ending, many of us may find ourselves starting to reach for the heating. In lockdown, 56% of home workers said they heated their homes for longer during the day, with home energy use increasing up to a third during the middle of the day. “By being more mindful of the temperature we set our thermostats, we can save both energy and money on our bills,” Goodman explains. “In fact, it is estimated that lowering your thermostat by just 1 degree can reduce your energy bill by up to £80, as well as reducing your overall energy consumption by 13%.”

To keep your home feeling warm after turning the temperature down, make sure your home is properly insulated. If you can’t completely insulate your home, then a short-term solution is to cover up any areas which allow draughts to come in.

Use sustainable work materials

Firstly, try and avoid buying office supplies and stationery when you still have perfectly usable materials to use up. By purchasing goods that aren’t necessary, even you are purchasing them to replace your less sustainable materials, you are unnecessarily causing more waste. “When opting for sustainable work materials, such as pens and paper, look for goods which are used from recycled materials and products that can then be easily recycled after use,” Goodman explains. “Many office supplies contain plastic, which takes years and years to biodegrade. Opt for pens which are refillable, or even better, look for specialist bamboo pens which can be recycled.”

If home working becomes a permanent routine, and you are considering renovating a room to be your home office, then try and find second-hand furniture. “Shopping vintage can not only create a nice look, but many vintage items are really high quality and will last longer than a cheaper alternative. Second-hand furniture also tends to be cheaper than buying new.”

Reduce home waste

It is estimated that UK households waste over 4.5 million tonnes of food every year, and with working from home, more of us are eating every meal at home than we ever have before. While this might be good for our purse strings, as we are less likely to head out for lunch when we can make our own lunch at home, this can have a negative effect on our food and packaging waste.

Take note of your council’s recycling. Many councils offer kerbside recycling, so ensure you dispose of your waste correctly to avoid goods finding their way into landfills. Also, try and use up all foods that may otherwise be thrown away, such as eating dinner leftovers at lunchtime or planning meals in advance to avoid making food that will just go to waste.

Natural light

Think back to your office. The lights always seemed to stay on, which is an unnecessary use of energy, in fact, it is estimated that 15% of office building’s energy comes just from lighting the office space, yet office workers don’t tend to have a say on whether the lights are left on. Working from home is great, as you are in control of the environment in which you work, however it is still important to be mindful of your habits.

“Replace inefficient lighting with energy-saving or LED bulbs to reduce your energy use,” Goodman advises. “Or make the most of natural light and try and find an area to work which allows for natural light. Not only does this eliminate the need for turning lights on unnecessarily, but studies have shown that natural light can release endorphins and make you happier and healthier, as well as boosting productivity levels.”

You might have read the 6th report from the Intergovernmental Panel on Climate Change (IPCC) report that painted a grim picture of our climate future. Fear not, I believe that our collective actions for a sustainable world is growing and many more people are getting motivated to help reverse climate change. This week I want to share a guide to home composting that you can share with friends and family.

Home Composting

Home composting is an effective and efficient way to dramatically reduce your waste stream at home while doing your part to reduce your carbon footprint. Organic material sent to landfills creates methane, a powerful greenhouse gas that contributes to the negative impacts of our changing climate. By making compost, you are creating a valuable soil amendment that you can use to benefit your landscape, boost plant growth and sequester carbon.

Many types of food waste and yard waste can be composted at home, including grass clippings, tree and shrub trimmings, vegetable garden and fruit tree waste, lawn clippings, autumn leaves, coffee grounds, and fruit and vegetable scraps from the kitchen. Home composters should not attempt to compost meat, dairy or large amounts of baked goods.

Home composting can be done in an enclosed bin or tumbler, in an easily made bunker, or using a no-fuss pile. The key is to blend your feedstocks to achieve a balance of carbon and nitrogen, keep things damp but not saturated, and ensure adequate oxygen deep in the pile. The microbes will do the rest.

What’s the recipe for home composting?

The four basic ingredients for composting are Nitrogen, Carbon, Water, and Air. The easiest compost recipe calls for blending roughly equal parts of green or wet material (which is high in nitrogen) and brown or dry material (which is high in carbon). Simply layer or mix these materials in a pile or enclosure; chop or shred large pieces to 12" or shorter. Water and fluff the compost to add air. Then leave it to the microorganisms, which will break down the material over time.

Nitrogen

• Green materials such as grass clipping and landscape trimmings are ideal sources of nitrogen for composting. Vegetable and fruit trimmings and peels can also provide nitrogen for composting. Coffee grounds and tea bags may look brown but are actually potent nitrogen sources. To reduce the potential for pests or odours, avoid meat or dairy scraps and always bury food scraps deep within the compost pile. Avoid pet faeces due to concerns about pathogens. However, manure from chickens, turkeys, cows or horses is rich in nitrogen and can help your compost pile get to proper temperatures, and make very good compost.

Carbon

• Brown (dry) yard and garden material such as dry leaves, twigs, hay, or shredded paper can provide the carbon balance for a compost pile. Chop or shred large pieces to 12 inches or shorter (thick, woody branches should be chipped, ground up, or left out). Untreated wood chips and sawdust are powerful carbon sources that may be useful if the pile contains excess nitrogen.

Water

• One of the most common mistakes in composting is letting the pile get too dry. Your compost pile should be moist as a wrung-out sponge. A moisture content of 40 to 60 per cent is preferable. To test for adequate moisture, reach into your compost pile and grab a handful of material and squeeze it; if a few drops of water come out, it’s probably got enough moisture, if it doesn’t, add water. When you water, it is best to put a hose into the pile so that you aren’t just wetting the top. You can also water as you are turning the pile. During dry weather, you may have to add water regularly. During wet weather, you may need to cover your pile. A properly constructed compost pile will drain excess water and not become soggy.

Air

• The bacteria and fungus doing the hard work in your compost pile need oxygen to live. If your pile is too dense or becomes too wet, the air supply to the inside is cut off and the beneficial organisms will die. Decomposition will slow and an offensive odour may arise. To avoid this, turn and fluff the pile with a pitchfork often, perhaps weekly. You can also turn the pile by just re-piling it into a new pile. Forcing air deep into the pile using blowers, perforated pipes or other means of aeration is one way to reduce the work of turning a pile.

How do I compost?

A well-managed compost pile will produce finished compost in about three months. A less intensively managed pile may need six months to a year. You may want to stop adding to your compost pile after it gets to optimal size (about 1 cubic yard) and start a new pile. This will allow your first pile can finish decomposing.

How to compost food waste?

Food is generally a high-nitrogen feedstock that should be blended with plenty of dry leaves, sticks and twigs, wood chips, sawdust, dried/dead plants, shredded newspaper, or paper from a home shredder, and mixed yard waste. Bury food materials deep in your pile, and always cover fresh food with a thick layer of high-carbon brown materials to keep out rodents and other vermin. Adding a dusting of dirt or unscreened, mature compost will help. If the pile gets too wet or dense with food scraps, it will smell bad and composting will slow down or stop altogether. Avoid meat, dairy, fats and oils and large amounts of carbohydrates like bread and pasta; these can cause odours and are very attractive to pests.

Composting Techniques

Composting can be done with more effort and faster results–or can be done with less labour, which will take longer and may not kill all weed seeds.

• Hot composting: Compost piles that have the right blend of nitrogen (greens) and carbon (browns), and are kept moist and fluffed regularly, will heat up fast, stay hot, and destroy most weed seeds and pathogens. With faster decomposition, the compost may be ready in 2 to 3 months. Once the pile is fully built, new material is not added, so proper hot composting requires more than one pile.
• Slow composting: Compost will happen even if you just pile up organic waste, water sporadically, and wait. Since this type of pile won’t get too hot and is turned infrequently, breakdown will be slower and less even. Weed seeds and recalcitrant materials may survive. Worms and other insects, which cannot survive high heat decomposition may be able to live in these piles and help break down the material. Casual composting can take up to a year.
• Vermicomposting: Composting with the aid of worms.

Home Composting Bins

Composting can be practised in backyards in a homemade or manufactured composting bin or simply an open pile (some cities require enclosed bins).

While backyard composting can be done successfully in uncontained piles, a composting bin can provide benefits such as improving the look of your composting area, improving your ability to maintain temperatures when hot composting and preventing rodents from accessing your compost pile.

Some communities offer free or discounted bins to residents as an incentive to compost at home. A successful program to promote home composting can reduce a community’s cost of solid waste collection and disposal. In addition to bins, some communities offer technical assistance programs, such as workshops or a hotline service, to assist residents.

Bins vary in terms of cost, size, ease of use, and rodent resistance.

• Hoops and Square Bins: These are usually the least expensive type of bin. Because hoops can be rolled, they are also easier to ship. Some hoops have tops and lids, which help in making them more rodent resistant. Square wire or plastic bins are like hoops except that they usually have supports in four corners, making them a square rather than round shape.
• Cones and Boxes: These bins are typical of solid construction with lids and bottoms. Some have doors at the bottom for harvesting finished compost. Cone/box-type bins tend to be more expensive than hoops.
• Stackables: These are box-shaped with sections that come apart.
• Tumblers: These are self-contained barrels, drums, or balls that rotate for mixing the composting materials.

Typical homemade bins can be constructed out of scrap wood, chicken wire, snow fencing or even old garbage cans (with holes punched in the sides and bottom).

Manufactured bins include turning units, hoops, cones, and stacking bins. Take the time to consider your options and then select a bin that best fits your needs.

What Size?

This depends on the space available to you but ideally, the compost pile should be at least three feet wide by three feet deep by three feet tall (one cubic yard). This size provides enough insulation for the organisms to live. Smaller piles may struggle to achieve sufficiently high temperatures to kill bad organisms and weed seeds. Larger piles can achieve composting temperatures more easily but need more frequent turning to ensure proper oxygen supply.

Can I compost in cold temperatures?

Yes, as long as the compost pile can get hot enough at its core. The colder the outside temperatures, the more volume (bigger pile) you need to generate heat.

Compostable materials go a step beyond biodegradable materials by breaking them down into natural components and becoming a part of healthy soil. Home compostable materials do not require the high heat (over 136° F) of industrial compost facilities to break down.

They can biodegrade in the moderate heat (68-86° F) of home compost piles/bins. The concept of home compostability has come to carry extra weight as many commercial compost facilities are refusing to collect products designed for industrial compost.

How to Tell When you have Finished Compost?

Compost is finished when the original material has been transformed into a uniform, dark brown, crumbly product with a pleasant, earthy aroma. There may be a few chunks of woody material left; these can be screened out and put back into a new pile. You should not be able to recognise any of the original feedstocks. There should be no foul odours.

Troubleshooting a Home Composting Bin or Pile

Symptom: The pile smells bad

Problem(s):

• Not enough air
• Too much moisture

Solution(s):

• Turn the pile if not enough air
• Add dry materials if too moist

—————————————————————————

Symptom: The pile will not heat up

Problem(s): could be one or all of the following

• Not enough moisture
• The pile size is too small
• Lack of nitrogen-rich material
• The particle size is too big

Solution(s):

• Add water if dry
• Build pile to at least 3’ x 3’ x 3’
• Mix in grass clippings or fruit/vegetable scraps
• Chip or grind materials

—————————————————————————

Symptom: The pile attracts flies, rodents, or pets

Problem(s): Pile contains bones, meat, fatty or starchy foods, or animal manure

Solution(s): Alter materials added to a pile; bury fruit/vegetable scraps in the middle of the pile, or under 8" to 10" inches of soil, or compost them in a worm bin.

—————————————————————————

Symptom: Pile has slugs in it (and so does the garden)

Problem(s): Pile is easily accessible and provides daytime hiding place and breeding ground for slugs

Solution(s): Remove slugs and slug eggs from a pile (eggs look like very small clusters of pearls). Locate compost pile far from vegetable gardens and/or create barriers around pile/garden (for example, traps and copper flashing).