TL;DR
Overlap between AI Overview citations and traditional top-10 rankings dropped from 76% to 38% in a year, meaning half of cited pages now rank outside the top 100. Word count shows virtually no effect (correlation 0.04), and 53.4% of citations go to pages under 1,000 words. Adding citations, quotations, and statistics boosted visibility 30–40% in a controlled benchmark, while keyword stuffing hurt.
AI Overview coverage itself is volatile—peaked at 25% of queries in July 2025, then fell to under 16% by November. The practical verdict: no special markup or AI-specific formatting is required, but investing in authoritative sourcing and evidence density is the only tactic with measurable support—and you should verify any traffic-impact claims against your own Search Console data, not aggregate studies.
What determines whether Google AI Overviews cite your content
Google AI Overviews pull from the same index and ranking systems used for regular Search results — there's no separate "AI index" and no special markup required to qualify. A page just has to be indexed and eligible to appear with a snippet in normal Search results; Google's own documentation states there are "no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary." What happens after that — whether a specific page gets cited — is driven by two AI techniques layered on top of standard ranking, and increasingly, by independent data that shows the pattern shifting fast.
How AI Overviews actually work, according to Google
Direct answer: Google describes AI Overviews as running on two mechanisms, both documented in its Search Central guide to AI features and its guide to optimizing for generative AI search:
- Retrieval-augmented generation (RAG): the AI Overview retrieves candidate pages using Google's core Search ranking systems, then generates a summary grounded in what those pages say.
- Query fan-out: for a single search, Google's systems can issue multiple related sub-queries behind the scenes — covering subtopics, comparisons, or follow-up questions — and pull supporting links from across all of them, not just the original query.
Google is explicit about what this does not require. Per its own guidance, you don't need an llms.txt file ("Google Search itself doesn't use them"), you don't need to chunk your content into small blocks, you don't need an "AI-specific" writing style, and structured data is not a documented requirement for AI Overview eligibility. Google frames "AEO" and "GEO" as rebrands of the same work: "optimizing for generative AI search is optimizing for the search experience, and thus still SEO."
That's the official version. Independent research over the past year shows the practical reality is messier — and shifting quickly.
What independent studies have actually measured
Ranking position matters less than it used to. Ahrefs' first large-scale citation study, published in July 2025, analyzed 1.9 million citations across roughly 1 million AI Overviews and found 76% of cited URLs also ranked in the traditional top 10 for the same query (median cited rank: 3). A follow-up analysis covering 863,000 keywords and 4 million AI Overview URLs, reported by Search Engine Journal in early 2026, found that overlap had fallen to 38%, with citations now split almost evenly between pages ranked 11–100 and pages ranked below 100. Ahrefs attributes part of the drop to improved citation-detection methodology (making the two datasets not perfectly comparable) and part to query fan-out pulling in pages that cover a subtopic well even without ranking #1 for the head term.
Content length has almost no measurable effect. In the same July 2025 dataset (174,048 pages, ~1.6 million cited URLs), Ahrefs found the Spearman correlation between word count and citation position was 0.04 — essentially none. The average cited page ran 1,282 words; more than half of all citations (53.4%) went to pages under 1,000 words. Pages that landed in the top citation slot averaged slightly fewer words (1,270) than pages cited further down (1,690 average for positions 4–10). Word count is not a lever worth pulling on its own.
Citing sources, quoting, and adding statistics measurably help — in a controlled benchmark. The academic paper that coined the term "Generative Engine Optimization," published by researchers from Princeton, Georgia Tech, the Allen Institute for AI, and IIT Delhi and later presented at KDD 2024, tested nine content-optimization tactics across roughly 10,000 queries. The three tactics that performed best — adding citations, adding quotations, and adding statistics — produced a 30–40% relative improvement on the paper's visibility metric, and combining them with clearer writing produced gains up to 40%. Keyword stuffing was one of the weakest tactics tested and sometimes hurt visibility. The caveat matters: this was a synthetic benchmark run against 2023–2024-era models, not Google's live AI Overviews system, so the exact percentages don't transfer directly — but it's the best controlled evidence available that sourcing and evidence density help.
AI Overview coverage itself is volatile. A Semrush analysis of over 10 million keywords, covered by Search Engine Land, found AI Overview visibility peaked at just under 25% of tracked queries in July 2025, then fell to under 16% by November 2025 — while the mix of query types shifted from almost entirely informational (91% of triggers in January) toward more commercial and transactional queries (up to 18% and 14% respectively by October). Google is actively tuning when and where AI Overviews appear; a page's citation status today isn't a stable signal.
Whether AI Overviews cost you clicks is genuinely contested. Pew Research Center tracked the actual browsing behavior of 900 U.S. adults in March 2025 (68,879 queries, 12,593 with an AI summary) and found users clicked a traditional search result in only 8% of AI-summary searches, versus 15% without one — and clicked a link inside the summary itself in just 1% of cases. Users also abandoned their session more often after an AI-summary search (26% vs. 16%). Yet Semrush's much larger 2025 dataset found organic click-through rates on AI-Overview-triggering keywords rose over the same period. Both can be true at once — different methodologies, different query mixes, different points in a fast-moving rollout — but treat any single "AI Overviews will/won't kill your traffic" claim as unproven until you check it against your own Search Console data.
What this means in practice
| Signal | What the evidence shows | Confidence |
|---|---|---|
| Indexed + snippet-eligible | Baseline requirement per Google's own docs | High — stated directly by Google |
| Ranking in the traditional top 10 | Still correlated with citation, but the overlap dropped from ~76% to ~38% in under a year | Medium — real but weakening trend |
| Word count / content length | No meaningful correlation (Spearman 0.04) | High — large-sample data |
| Citing sources, quotes, statistics | Boosted visibility 30–40% in a controlled academic benchmark | Medium — real effect size, unproven transfer to live Google systems |
| Structured data / schema markup | Not a documented requirement for AI Overview eligibility | High — stated directly by Google |
llms.txt files | Explicitly not used by Google Search | High — stated directly by Google |
| Topical coverage across subtopics | Plausible driver given how query fan-out works, not independently measured in isolation | Low–medium — mechanistic, not directly tested |
| Net effect on your click-through rate | Contested; direction varies by study and query type | Low — conflicting data |
How to actually prepare a page
- Confirm the basics first. The page needs to be indexed, crawlable, and eligible for a normal Search snippet. If it isn't showing up in regular Search, it won't show up in an AI Overview either — Google draws from the same index.
- Answer the core question in the first sentence or two. This doesn't have a documented citation-boost number attached to it, but it matches how Google describes AI Overviews summarizing "the gist" of a query, and it's the pattern seen in most cited pages.
- Cover the query's likely sub-questions on the same page, not just the head term. Query fan-out means Google may be citing whichever page best answers a related sub-query, not necessarily the one ranking #1 for your target keyword.
- Back claims with real citations, direct quotes, and specific numbers. This is the one tactic with actual (if benchmark-limited) evidence behind it — a 30–40% visibility lift in the Princeton GEO study.
- Stop optimizing word count as a goal in itself. Write the length the topic needs. Both a 400-word direct answer and a 2,500-word deep dive can get cited; length alone predicts almost nothing.
- Skip the cargo-cult tactics. Google has explicitly said
llms.txt, forced content-chunking, "AI-flavored" rewriting, and inauthentic brand-mention campaigns don't move the needle. Spend that effort on the content itself. - Check Search Console, not vibes. Google's documentation says AI Overview appearances are reflected in the Performance report — track it over weeks, not days, given how much AI Overview coverage itself has swung in the last year.
Limitations and honest caveats
None of the studies above measure causation cleanly. Ahrefs itself has cautioned that its correlational findings — including a separate 75,000-brand analysis showing off-site brand mentions correlate more strongly with AI citation likelihood than backlinks do — show association, not proof that any single tactic causes a citation. The Princeton GEO paper is the closest thing to a controlled experiment in this space, and even its authors note the results came from a synthetic benchmark against older models, not live production systems. Google has not published the internal weighting of any ranking or citation signal, and its own documentation focuses on what you don't need rather than a scored checklist of what you do. Anyone who hands you a precise formula for "the AI Overview algorithm" is guessing.
How nqz.ai fits in
Direct answer: If you're trying to figure out whether your content is actually showing up in AI Overviews (and other AI answer engines) rather than guessing from anecdote, nqz.ai's AI search visibility tracking monitors citation and mention patterns across AI search surfaces over time, and its content-quality and E-E-A-T checks flag pages missing the sourcing, evidence, and directness signals covered above — without claiming to know Google's exact algorithm, because nobody outside Google does.
FAQ
How does Google decide what to show in AI Overviews?
It uses retrieval-augmented generation (RAG) to pull from the same index and ranking systems as regular Search, plus a "query fan-out" step that issues related sub-queries and can cite pages that answer those, not just the original search term. This is Google's own description, not third-party speculation.
Do I need schema markup or an llms.txt file to appear in AI Overviews?
No. Google's Search Central documentation states structured data isn't a requirement for AI Overview eligibility and that Google Search doesn't use llms.txt files at all.
Does content length affect whether I get cited?
Barely. A large Ahrefs analysis found essentially no correlation (Spearman 0.04) between word count and citation likelihood or position. Match the length to the query — don't pad or artificially trim to hit a target.
Is ranking #1 in Google still required to be cited in an AI Overview?
No, and it's becoming less predictive over time. The share of AI Overview citations coming from traditional top-10 pages dropped from about 76% (mid-2025) to about 38% (early 2026) in Ahrefs' data.
Do AI Overviews reduce website traffic?
The evidence is mixed and worth checking for your own site rather than assuming. Pew Research found meaningfully lower click-through rates on AI-summary searches; a much larger Semrush dataset found rising CTR on AI-Overview-triggering keywords over the same period. Different methodologies, different conclusions.
What's the actual difference between SEO, AEO, and GEO?
By Google's own account, none for its Search product: "optimizing for generative AI search is optimizing for the search experience, and thus still SEO." Other engines (ChatGPT, Perplexity) may weight signals differently, which is where AEO/GEO terminology is more genuinely distinct.
