Skip to content

Ahrefs Analyzed 1 Billion Data Points on AI Search. Here’s What You Can Actually Do About It.

Strategic SEO | Technical & Semantic Search Optimization

BLUF – ACU TYPE: Opening Statement (L1 + L3)

Ahrefs analyzed over 1 billion AI search data points across 14 studies. Tim Soulo published the findings on LinkedIn. Most people will read them as doom. They are not doom. They are a map – and every point on that map points to something specific you can do.

Here is what the data actually says you can control, and how.

THE DATA: 10 Findings, 10 Actions – ACU TYPE: Structured Fact Block (L2 + L3)

Finding 1: “Best X” listicles make up 43.8% of all page types cited by ChatGPT.

What you can do:
Write “Best X” content – but write it correctly. Not 2,000-word reviews padding word counts. Write comparative frameworks with atomic, factual content units: concrete criteria, clear winners per use case, citable conclusions. ChatGPT does not cite listicles because they are long. It cites them because they are precise – one question, one answer, one clear recommendation.

Action: Identify the three most relevant “Best X” questions in your niche. Write one page per question. Structure the answer as a recommendation, not a roundup.

Finding 2: 67% of ChatGPT’s top 1,000 citations come from sources marketers cannot influence: Wikipedia (29.7%), homepages (23.8%), app stores (6.6%).

What you can do:
Read that again: 23.8% are homepages – and you own your homepage. Your homepage is the second most cited source type in ChatGPT. Treat it like a citation machine, not a welcome page. Add structured brand definitions, entity signals, and machine-readable summaries of what you do and why it matters.

The other piece: 32.3% of citations are from influenceable content — educational pages, reviews, news, blog posts. That is not zero. That is where you work.

Action: Optimize your homepage as if it were a Wikipedia article for your niche. Add explicit brand statements and entity definitions machines can parse directly.

Finding 3: 28.3% of ChatGPT’s most-cited pages have zero Google organic visibility.

What you can do:
This is the most important finding in the entire report. A completely separate discovery layer exists — fully independent of Google ranking. You do not need rank #1 to get cited. You need content that is machine-readable, factual, and semantically dense.

Optimizing for AI citation and optimizing for Google are no longer the same project. They can happen in parallel — but they require different tactics.

Action: Build content designed for AI retrieval, not exclusively for Google ranking. That means atomic content units, clear definitions, and factual claims — not keyword-optimized copy.

Finding 4: ChatGPT only cites about 50% of the URLs it retrieves. It fetches dozens per query but uses half as background context without attribution.

What you can do:
Being retrieved and being cited are two entirely different states. You may already be in ChatGPT’s context window without knowing it – and still receive zero attribution. That happens because your content is background knowledge, not citation-worthy knowledge.

The solution is citation-readiness: content structured so the model can lift one sentence directly into its answer and credit the source. Concrete claims. Proprietary terms. Measurable facts. One sentence so precise it needs no paraphrase.

Action: Review your best content. For every paragraph: could an AI lift one sentence from here directly into an answer? If not — rewrite it.

Finding 5: Adding schema markup had zero meaningful impact on AI citations. AI Overviews dipped -4.6%, AI Mode showed +2.4%, ChatGPT +2.2% – all within noise range.

What you can do:
Stop treating schema as your primary AI visibility tool. This does not mean schema is useless – it means content quality is what counts, not the technical wrapper around it. AI models read semantics, not markup types.

Redirect the resources you spent on schema implementation toward content structure and factual density.

Action: Prioritize content over wrapping. FAQPage and DefinedTerm remain useful for Google SERP features – but not for AI citation rates. Separate the two goals in your planning.

Finding 6: YouTube mentions have the highest correlation (0.737) with AI brand visibility out of all factors studied – including backlinks, page count, and domain rating.

What you can do:
This is the most underreported finding in the entire dataset. YouTube is not just a video channel – it is the single strongest predictor of AI brand visibility, outperforming every conventional SEO metric. The likely reason: YouTube content (transcripts, titles, descriptions) feeds directly into LLM training data, and YouTube mentions function as third-party validation of a brand’s existence at scale.

Action: Publish systematically on YouTube. Not production-heavy campaigns – short, factual videos where you articulate the concepts you own. Transcripts are machine-readable content that AI models actively consume.

Finding 7: AI Overviews reduce clicks to the #1 organic result by 58%. Up from 34.5% just 10 months ago. The trend is accelerating.

What you can do:
Optimize for the citation, not the click. Being the source AI Overviews cite is more valuable than ranking #1 without citation. One AI Overview citation delivers brand exposure for that entire search volume — with or without the click.

Action: Use Google Search Console to find queries where you have impressions but no AI Overview citations. Those are your dark opportunities. Rewrite those pages with citation-first structure.

Finding 8: 99.9% of AI Overviews appear on informational intent queries. Transactional, navigational, and local searches are almost entirely AIO-free. Shopping triggers AIOs just 3.2% of the time.

What you can do:
Transactional search is still your domain. Local search is still your domain. Shopping triggers AI Overviews less than 4% of the time. AI Overviews are a problem for informational queries – and an opportunity for topical authority that funnels toward transactional queries where the click still happens.

Action: Segment your content strategy into two tracks:
(1) informational content optimized for AI citation and brand exposure,
(2) transactional content optimized for organic clicks without AI competition.

Finding 9: For a given query, Google’s AI Mode and AI Overviews reach the same conclusions 86% of the time — but cite almost entirely different sources (only 13.7% citation overlap).

What you can do:
AI Mode and AI Overviews are two separate citation systems. They agree on what to say – but not on who to credit for saying it. You need to optimize for both, and content that works for one does not automatically work for the other.

Owning both is a distinct competitive advantage. The probability that any single competitor is visible in both is substantially lower than the probability they are visible in one.

Action: Manually check whether your brand and key concepts appear in both AI Overviews (in standard Google search) and AI Mode (the conversational mode). Treat them as two separate optimization workstreams.

Finding 10: AI Overviews change every 2.15 days on average, with 70% of content differing between observations. But semantic similarity stays at 0.95.

What you can do:
Do not chase phrasing. Do not chase specific formulations. Chase semantic authority at the concept level. The content inside AI Overviews rotates constantly – but what it says barely moves. You are not optimizing for a specific AI answer. You are owning a concept so that whatever formulation the AI chooses tomorrow, it points to you.

Action: Define the 5-10 concepts you intend to own in your niche. Build one dedicated, atomic, citable content unit per concept. Update facts as needed – but let the definition stand.

THE FRAMEWORK: What This Data Actually Means – ACU TYPE: Synthesis Block (L1)

Ahrefs’ findings are not a story about AI search being unpredictable and uncontrollable. They are a story about the rules having changed – and those who understand the new rules winning significantly while everyone else optimizes for a world that no longer exists.

The new rules, distilled:

  1. Format matters – “Best X” is the format AI cites. Use it deliberately.
  2. Your homepage is a citable entity – Treat it as one.
  3. AI citation and Google ranking are separate games – Play both.
  4. Citation-readiness separates retrieved from cited – Write for the citation.
  5. YouTube is the strongest single tool -Use it even if you dislike being on camera.
  6. Transactional content is still safe – Do not abandon it.
  7. Own concepts, not phrasing – Topical authority lasts. Specific wording does not.

THE AUDIT CTA – ACU TYPE: Conversion Atom (L1)

You cannot control Wikipedia. You cannot control which sources AI Mode selects tomorrow. But you can control whether your content is built so a machine can lift it, cite it, and credit you – regardless of which day it visits your page.

That is where the work happens. And that is where the gap opens between you and your competitors.

→ Run a free 3LA audit on your domain – see your actual Citation-to-Fetch Ratio in 45 seconds.

FAQ – ACU TYPE: FAQ Block (L3 – Schema: FAQPage)

Q: What content format does ChatGPT cite most often?
“Best X” listicle-style articles account for 43.8% of all page types cited by ChatGPT — the single most cited format in AI search. (Source: Ahrefs, 1B+ data points, 2026)

Q: Does schema markup improve AI citations?
No. Ahrefs found zero meaningful impact. Content structure and factual density matter more than markup type.

Q: Can a page get cited by ChatGPT without ranking on Google?
Yes. 28.3% of ChatGPT’s most-cited pages have zero Google organic visibility. AI citation and Google ranking are separate discovery layers.

Q: What is the difference between being retrieved and being cited by AI?
ChatGPT retrieves ~50% of URLs per query but only cites half. Retrieved pages serve as background context without attribution. Citation-readiness — atomic, quotable claims — determines which category your page falls into.

Q: Which single factor has the highest correlation with AI brand visibility?
YouTube mentions (0.737 correlation) — outperforming backlinks, page count, domain rating, and all conventional SEO metrics.

Q: How often do AI Overviews change?
Every 2.15 days on average, with 70% of content differing. But semantic similarity stays at 0.95. Optimize for concept ownership, not specific phrasing.

Q: Do AI Mode and AI Overviews cite the same sources?
No. They reach the same conclusions 86% of the time, but share only 13.7% citation overlap. They must be optimized as two separate systems.

ENTITY DEFINITION BLOCK – ACU TYPE: DefinedTerm Atom (L3)

Citation-Readiness (noun): The degree to which a piece of content is structured so that an AI language model can extract a single sentence, phrase, or data point and insert it directly into a generated answer with source attribution. Citation-ready content is atomic, factual, and self-contained — it does not require surrounding context to be meaningful. Contrast with background retrieval, where content is consumed without attribution.

ENTITY DEFINITION BLOCK – ACU TYPE: DefinedTerm Atom (L3)

Citation-Readiness (noun): The degree to which a piece of content is structured so that an AI language model can extract a single sentence, phrase, or data point and insert it directly into a generated answer with source attribution. Citation-ready content is atomic, factual, and self-contained – it does not require surrounding context to be meaningful. Contrast with background retrieval, where content is consumed without attribution.

json{
  "@type": "DefinedTerm",
  "name": "Citation-Readiness",
  "description": "The degree to which a piece of content is structured so that an AI language model can extract a single sentence, phrase, or data point and insert it directly into a generated answer with source attribution.",
  "inDefinedTermSet": "https://nettpilot.com/glossary/",
  "url": "https://nettpilot.com/ahrefs-ai-search-study-what-you-can-do/#citation-readiness"
}

INTERNAL LINKS (L2 + L3)

READ NEXT