Why ranking well on Google does not mean AI Overviews read your page

Research papers, vendor studies and our own measurements all find that a large share of the websites AI Overviews cite are not in Google's top 10 for the same search, even though they disagree on how large. The work is to make sure one of your pages answers the questions behind the buyer's question, and that is a different job from ranking for one term.

This page is stage 1, Retrievable, of our AI search visibility funnel. If the engine cannot fetch or index your page at all, start with indexability instead.

Why does a competitor show up in the AI Overview when we rank higher?

A competitor shows up because the AI Overview is built from more searches than the one results page you rank on. Here is how it often looks for a software or fintech team. You rank first for "expense management software for startups". A buyer, preparing for your demo call, types a longer question: "what's the best expense tool for a 40-person startup that needs SOC 2 and NetSuite sync?" The AI Overview names three products. You are not one of them, and one of the cited sites is not on the first page of Google at all.

Nothing is broken, and nobody is penalising you. The engine answered a question you never ranked for, using pages it found through searches you could not see.

How does an AI answer engine decide what to read?

An answer engine searches, reads what it finds, and writes from the part it keeps. For Google, the company describes the pieces:

1. It turns your question into several searches. Google calls this "query fan-out": AI Overviews and AI Mode "may use a 'query fan-out' technique, issuing multiple related searches across subtopics and data sources" (Google Search Central). The buyer's one question might become searches about SOC 2 support, NetSuite integrations, pricing for small teams and alternatives to the market leader.
2. It collects candidate pages from those searches, using Google's ranking systems. Google says its generative AI features "are rooted in our core Search ranking and quality systems" and rely on them "to retrieve relevant, up-to-date web pages from our Search index" (Google Search Central). A Google explainer dated November 5, 2024, filed in the US search antitrust case as exhibit PXR0038, says AI Overviews are "built to only surface information that is backed up by top web results". Your ranking for the head term is one input among several. A page that ranks for none of the sub-searches may never enter the pool.
3. It reviews the retrieved pages and keeps what answers the question. Google says its systems "review the specific information from those retrieved pages" and that they "are able to understand the nuance of multiple topics on a page and show the relevant piece to users". A long product page that ranks well can still lose here if it never says plainly whether it syncs with NetSuite.
4. It writes the answer and links its sources. Google adds that fan-out lets it "display a wider and more diverse set of helpful links associated with the response than with a classic web search". The link list is meant to differ from the ranked list.

Two limits on this description. Google has not published how it scores or selects what it keeps, and the sub-searches it runs are not visible from outside. AI Mode and AI Overviews also differ: Google says they "may use different models and techniques, so the set of responses and links they show will vary". Our AI Overviews and AI Mode guide covers what Google has said in more detail, and the glossary defines query fan-out, passage and citation.

These two facts explain why one study calls the source selection "distinct from Google's ranking algorithm" while Google says the same features are rooted in its ranking systems. Both can hold: the cited pages are ranked, but for searches other than the one you see.

How much do Google rankings and AI Overview citations overlap?

They overlap far less than most teams assume. Independent researchers, using different queries and methods, land in the same place.

Kirsten and colleagues (arXiv 2510.11560, preprint) compared Google's organic results with five generative search systems across 4,706 queries, run in July to September 2025. On average, 53% of the domains AI Overviews consulted were not in the organic top 10 for that query, and 27% were not in the top 100. Running the same queries two months apart, only 18% of the web pages AI Overviews used were the same both times, against 45% for organic results.

Xu, Iqbal and Montgomery (ACM IMC 2026) issued 55,393 trending queries over 40 days in March and April 2026. Nearly 30% of the domains AI Overviews cited did not appear anywhere in the first page of results shown with them. They also found that 11.0% of the 98,020 individual claims in the answers were not supported by the pages cited for them.

Grossman and colleagues (ACM SIGIR 2026) ran about 11,500 queries through Google's organic results, AI Overviews and Gemini Flash 2.5. The sources the three returned were "substantially different", with an average Jaccard similarity below 0.2. Jaccard is the number of sources two lists share divided by the number of distinct sources across both: 1 means identical lists, 0 means nothing in common.

Li and Sinnamon (ASIS&T 2024) audited ChatGPT, Bing Chat and Perplexity on 48 queries across four public-interest topics over seven days. Of the 355 web domains that Bing Chat and Perplexity cited, only 26% were cited by both. It is an older, smaller study on other engines, and it says little about business buyer questions, but it shows the same pattern outside Google.

Read together: the source lists are largely separate from the ranked lists, they differ between engines, and they move over time. A rank report tells you about rankings. It does not tell you what an AI answer reads.

Why do the published overlap figures disagree so much?

The figures disagree mainly because they measure different things. Figures from about 17% to 94% circulate for "how much AI Overviews overlap with the top 10", and most of the gap comes from five choices that the headlines leave out.

The five choices:

1. Keywords or citations. "94% of keywords have at least one cited page in the top 20" and "44% of citations come from beyond the top 20" are from the same platform's data. One answer with one top-10 source and five others counts as a hit in the first measure and as one in six in the second.
2. Pages or websites. A URL match is stricter than a domain match: if your pricing page ranks and your blog post on the same site is cited, a domain count calls that overlap and a URL count does not. Domain-level shares will therefore look higher than page-level shares on the same data.
3. Which window. Top 10, top 20, top 100, or the first page as displayed. The third platform reports about 17% for the top 10 and about 53% for the top 100, and a headline saying that about 55% of citations now rank organically is the wider window.
4. When. The studies run from 2025 to 2026, and Google changes how AI Overviews retrieve and cite. A figure from one year is not a trend line against another study's figure from the next.
5. Whose queries. A vendor's tracked customer keywords, Google Trends queries, and non-branded category searches are different populations. Industry splits move the number a lot: the third platform reports about 75% for healthcare, 73% for education and 23% for e-commerce on its wider measure in September 2025.

The same applies to "how often does position 1 get cited": the first platform reports 43% for position-1 URLs in October 2025, the second about 25% in mid-2026. Different units, panels and dates, so neither contradicts the other.

The answer engines repeat the confusion. When we captured Google's AI Mode answer to "How much do AI Overview citations overlap with the organic top 10?" on October 6, 2026, it put a keyword-level "90% to 94%" next to per-citation figures of "17% to 38%" as one spectrum of overlap, and presented the third platform's healthcare and e-commerce figures as overlap with "top rankings", which in that platform's own report is the wider measure. Before you quote any overlap number, ask the five questions above.

What we see in our own measurements

Our own data points the same way. We track non-branded category searches in Google AI Overviews in the US, and we capture each search more than once, on different days, before we treat a result as real. Our unit is the website, counted per search. Our methodology page describes how.

"In our study of 150 non-branded Google searches (US, July 2026), about 57% of the websites cited in AI Overviews did not appear in that same search's top 10 organic results (95% interval 54% to 60%)."

142 of the 150 searches showed an AI Overview, and the share is over 1,051 answers from 8 captures. It describes what was cited; it is not a measure of cause. An independent study found a similar share: Kirsten and colleagues' 53% above, on different queries and dates. The second platform's 57% in the table is a different unit (cited pages across its tracked keywords), so the two figures match in number only.

That finding covers Google AI Overviews, US searches, and non-branded queries only. It says nothing about ChatGPT, Perplexity or searches for your brand name. Some of the sources from outside the top 10 are not web pages in the usual sense: "YouTube is one of the few social platforms that regularly shows up as a source; posts on X almost never do"

Four further measurements from the same program:

• "On 148 commercial SaaS searches such as 'best CRM for small business', Google showed an AI Overview on 143 of them, and each one cited a median of 9 different websites." That was one day, July 19, 2026, on software head terms only.
• "Google showed an AI Overview on 573 of 650 non-branded business searches we tracked. On 91% of the searches we added in July (95% interval 88% to 93%), at least one cited website was missing from the organic top 10." The 91% covers the 500 searches newly captured in July.
• "The website in the first organic position was cited by the AI Overview 56% of the time (95% interval 51% to 61%). Websites in positions 6 to 10 were cited 34% of the time (31% to 37%)." This counts websites per search on our panel, so it is not comparable with per-page position rates from vendor panels.
• "AI Overview source lists turn over fast: across 8 captures in one week, the typical website stayed in an answer's source list for one capture interval." A citation seen once is a reading, not a position you hold.

The position measurement matters for balance. Ranking first is associated with being cited more often, so rankings still count. They are just not the deciding step, and the association does not tell you what moving up a position would be worth.

If ranking is not the gate, what decides which passage gets cited?

The best controlled evidence says relevance to the specific question, and where a source sits among the candidates, matter most. Surface edits matter little.

Vishwakarma, Kumar and Jamidar (ACM SIGIR 2026) ran 252,000 trials across six language models. Each trial gave a model two sources that differed in exactly one of 18 content factors, with brand names hidden and source order swapped to cancel out position effects. Topical relevance and list position were the biggest drivers of being cited first. Explicit price information and a recent date also helped consistently. Formatting-only edits had little effect.

Puerto and colleagues (C-SEO Bench, NeurIPS Datasets and Benchmarks 2025) tested content-rewriting tactics sold as "conversational SEO" across several domains. Only 3 of 54 combinations of tactic and domain produced a statistically significant improvement, and some tactics made things worse. A document's starting position among the retrieved results had a much larger effect than any rewrite.

Google's guidance for its own AI features points the same way. Its July 2026 guide says there is no requirement to break content into small pieces, no need to rewrite content just for AI, and no special schema.org markup to add (Google Search Central). Google's own AI Overview for "Why does a competitor appear in Google's AI Overview when we rank higher?", captured the same day, told the reader that competitors win through structured data, Wikidata entries, "machine-readable schema" and FAQ sections anyway.

So the practical reading is this. Ranking still matters, but for the engine's sub-searches, not for your one head term. A page earns a place by plainly answering a question buyers ask, with the specifics (price, integrations, compliance, dates) that let the engine use it.

Is an AI Overview citation worth chasing if nobody clicks?

It is worth having, but as a brand impression during the buyer's research, not as a traffic source. Pew Research Center followed the browsing of 900 US adults across 68,879 Google searches in March 2025. When an AI summary appeared, people clicked a regular search result in 8% of visits, against 15% when no summary appeared, and clicked a link inside the summary itself in 1% of visits to pages that showed one.

How often an AI Overview appears at all depends on the queries: 18% of the searches in Pew's panel, 13.7% of Xu and colleagues' trending queries (64.7% of question-form ones), and 51.5% of Grossman and colleagues' real-user queries. Google says AI Overviews "often don't trigger". Measure whether you are named and cited on the questions your buyers ask, not sessions. Our page on zero-click AI answers covers what to measure instead.

What should we do this week?

Start with the questions your buyers ask, then make sure one of your pages answers each of them. Five steps, in order:

1. Write down the 15 to 20 questions buyers ask before a demo. Comparisons, "best X for Y", integration and compliance questions, pricing for a team your size. Check each one in AI Overviews on at least two different days and record which sites are cited. One check is not a measurement.
2. For each question, find the place on your site that answers it plainly. If there is none, add the answer on the page where it belongs (pricing, integrations, security, comparison pages) rather than a new blog post per question. Google warns that creating pages for every variation of a query mainly to influence AI answers counts as scaled content abuse. Include the specifics: price, named integrations, certifications, the date it was last checked.
3. Map the sub-searches, not just the head term. For each buyer question, list the narrower searches an engine is likely to run, and check whether any page of yours ranks for them. Gaps here are often the reason you are absent.
4. Confirm you are eligible. Google states that a page must be indexed and allowed to show a snippet, and that the site must be included in Search generative AI features in Search Console, which is the default. Check that the setting is on include, that your answers are not inside nosnippet or data-nosnippet markup, and that robots.txt lets Googlebot in: Google says robots.txt rules for Googlebot are the control for its AI features in Search. Google-Extended is a separate choice that Google describes as limiting AI training and grounding in some of its other systems; it is not how you leave or join AI Overviews. Grossman and colleagues found that sites blocking "Google's AI crawler" were less likely to be retrieved by AI Overviews, but that is an association between a site's choices and its citations, not a switch. Our indexability guide walks through the checks.
5. Report AI visibility separately from rankings. Per engine, per question, as presence and share of citations over repeated checks. For Google, Search Console's Generative AI performance report counts how often your links appeared in AI Overviews and AI Mode. A blended score or a rank tracker will hide the gap this page describes.

When a page of yours is being read but the engine still prefers other sources, that is the next stage of the funnel: being trusted as a source . When the engine reads you but the answer is hard to find on the page, see writing pages for AI answer engines .

Frequently asked questions

Work with us

If you want this work done for you, from mapping your buyers' questions to fixing the pages that answer them, our managed AEO services do it with your team: Book a call. If you would rather start by seeing where you stand, Lenz, our AI visibility platform, gives you a free analysis of how ChatGPT and Gemini answer your buyers' questions: Get your free AI visibility report.

How this page was made

Written by Soroush Mehri. Reviewed by Dr. Maziar Gomrokchi.

Sources

• Kirsten, Grosse Perdekamp, Wu, Upadhyay, Gummadi and Zafar. "Characterizing Web Search in The Age of Generative AI." Preprint, arXiv 2510.11560
• Xu, Iqbal and Montgomery. "Measuring Google AI Overviews: Activation, Source Quality, Claim Fidelity, and Publisher Impact." ACM Internet Measurement Conference (IMC) 2026. arXiv 2605.14021
• Grossman, Liu, Chen, Smith, Borcea and Chen. "How Generative AI Disrupts Search: An Empirical Study of Google Search, Gemini, and AI Overviews." ACM SIGIR 2026. DOI 10.1145/3805712.3809667. arXiv 2604.27790
• Li and Sinnamon. "Generative AI Search Engines as Arbiters of Public Knowledge: An Audit of Bias and Authority." Proceedings of the Association for Information Science and Technology, 2024. DOI 10.1002/pra2.1021. arXiv 2405.14034
• Vishwakarma, Kumar and Jamidar. "What Gets Cited: Competitive GEO in AI Answer Engines." ACM SIGIR 2026 (short paper). DOI 10.1145/3805712.3808445. arXiv 2605.25517
• Puerto, Gubri, Green, Oh and Yun. "C-SEO Bench: Does Conversational SEO Work?" NeurIPS Datasets and Benchmarks 2025. arXiv 2506.11097
• An SEO platform vendor's AI Overview tracking reports (2025 and 2026). Not linked: the publisher is on our no-link list.
• Chapekis and Lieb. "Google users are less likely to click on links when an AI summary appears in the results." Pew Research Center, July 22, 2025. Pew Research Center
• Google Search Central. "AI features and your website." developers.google.com
• Google Search Central. "Optimizing your website for generative AI features on Google Search", last updated July 10, 2026. developers.google.com
• Google Search Console Help. "Search generative AI control" and "Generative AI performance report (Search)". support.google.com, support.google.com

Written bySoroush MehriCo-Founder and Chief AI Officer

Ex-Microsoft. 10+ years of experience in AI/ML, MSc from Université de Montréal (Mila). US patent holder.

Reviewed byDr. Maziar "Maz" GomrokchiCo-Founder and CEO

AI/ML scientist with 20+ years of research & industry experience. Ex-Google DeepMind, MIT, Harvard, Mila. McGill-Mila PhD.