How Google AI Overviews and AI Mode Pick What to Cite

How are AI Overviews and AI Mode actually different?

See the AI search glossary for short definitions of any term below that isn't familiar. Both sit inside Google Search, and Google's documentation treats them as a pair: "AI Overviews help people get to the gist of a complicated topic or question more quickly, and provide a jumping off point to explore links to learn more," while AI Mode is built for "further exploration, reasoning, or complex comparisons." In practice that means AI Overviews tend to be short, appearing above normal results only when Google's systems decide a direct AI summary adds value; AI Mode is a longer, more conversational answer a person actively chooses to use for a harder question.

The two aren't the same feature with a different wrapper. Ahrefs' analysis of 730,000 matched query pairs (September 2025 US data, via its Brand Radar tool) found AI Mode answers run about four times longer than AI Overviews on average and cite roughly 3.3 entities per response versus 1.3 for AI Overviews — AI Mode simply does more work per question, including issuing more sub-queries, and that shows up directly in how many sources and brands it ends up naming.

What does Google say actually decides inclusion?

Google's generative AI optimization guide is direct about this: the features "rely on our core Search ranking systems to retrieve relevant, up-to-date web pages," and "there are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary" beyond standard technical requirements — being indexable and eligible for a normal search snippet. The guide also closes off two common workarounds directly: no special content chunking is needed ("there's no requirement to break your content into tiny pieces"), and llms.txt files are not used — "Google Search ignores them."

What Google doesn't publish is the selection logic once a page is eligible — which of several qualifying pages gets the actual citation, and why. That part has to stay labeled as inference from observed behavior, not documented fact.

How does query fan-out change which pages get a chance to be retrieved at all?

Fan-out means Google's systems issue several related searches behind one question before writing an answer. Google's own documented example: a query about a lawn full of weeds can fan out into searches like "best herbicides for lawns" and "remove weeds without chemicals." Reporting based on this behavior puts the typical range at roughly five to eleven sub-queries for a single AI Mode question, though Google hasn't published an exact count or formula.

The practical effect: a page doesn't need to match the literal words in the box, it needs to match one of several sub-questions the system generates on its own. A page that only answers the surface-level question can miss every sub-query that actually drives the citations. We cover how to map and plan for this directly in what is query fan-out.

How is "included" different from "ranked," and why does it matter that the two features disagree so often?

On a results page, ranking is a single ordered list — being included and being ranked are the same event. In AI Overviews and AI Mode, a page can be retrieved and still never get cited in the final text, and — this is the part that trips people up — being cited in one Google AI surface says very little about the other.

Ahrefs' comparison of the two features found they cited the identical URL for the same query only 13.7% of the time, meaning roughly 86% of citations are different sources, even though the two answers reached a similar conclusion (an 86% average semantic-similarity score) 9 times out of 10. So the two systems usually agree on what to say while disagreeing on where they got it — a brand can be the cited source in AI Mode and completely absent from the AI Overview for the identical question, or the reverse.

That instability runs deeper when you compare either feature back to plain Google rankings. How much an AI Overview citation correlates with a page's position in Google's normal top 10 has been measured very differently depending on the study and when it ran:

Ahrefs' own update is explicit that part of the drop from 76% to 38% reflects improvements in its citation-detection tooling, not purely a change in how Google selects sources — the two waves aren't directly comparable. BrightEdge's 17% used a different dataset and method entirely. None of the three numbers is simply "wrong"; they're measuring a moving target with different nets. The one thing all three agree on directionally: a large and apparently growing share of AI Overview citations come from outside the top 10 — Ahrefs' March 2026 wave found 31% of citations now rank 11–100, and another 31% rank beyond page one entirely (outside the top 100). We go deeper on that pattern, across engines, in how AI decides which pages to cite.

A note on the evidence for this post. The original plan for this section was a side-by-side screenshot of the same prompt run in AI Overviews and AI Mode. No browser tool with live, logged-out access to Google's interface was available while drafting this post, so a screenshot pair wasn't captured — fabricating one wasn't an option. The Ahrefs figures above (730,000 real query pairs, a far larger sample than any single screenshot could show) are used here instead as the evidence for how differently the two features behave. A literal screenshot pair remains a reasonable thing to add later; see this post's entry in the production review for the exact capture instructions.

How can a brand check its own presence in each feature?

Ask the same real, unbranded question in a fresh, logged-out-style session in each surface separately — don't assume a result in one tells you about the other. Note, per answer: whether an AI Overview appeared at all (many queries don't trigger one), what was cited, and whether AI Mode's longer answer named different sources for the identical question. Repeat on more than one day, since Google's own guidance and third-party studies both treat these answers as dynamic, not fixed.

If you want a fast single check rather than a manual log across both features, the AI Citation Snapshot runs one real question through an engine and shows the live answer and its cited sources — one run, a first look, not an audit of Google's two surfaces specifically.

FAQ

Do AI Overviews and AI Mode cite the same sources for the same question? Rarely. Ahrefs found the two features cited the identical URL only about 13.7% of the time across 730,000 matched queries (September 2025 data) — most citations differ even when the two answers agree on substance.

Does ranking in Google's top 10 guarantee a citation in AI Overviews? No, and the correlation has weakened over time by most measurements. Estimates range from 76% (Ahrefs, July 2025) down to 38% (Ahrefs, March 2026) and 17% (BrightEdge, February 2026), depending on dataset and method — not a single stable number.

What is query fan-out, briefly? Google's systems splitting one question into several related searches before writing an answer, so coverage of the underlying sub-questions matters more than matching the literal query. Full detail in what is query fan-out.

Do I need an llms.txt file or chunked content to appear in AI Overviews? No. Google's own guidance states explicitly that llms.txt files are ignored by Google Search and that no special chunking is required.

Is AI Mode just a longer version of AI Overviews? Not functionally. It's a separate experience for harder, more exploratory questions, and it tends to cite more entities, longer source lists and different pages than an AI Overview answering the same base question.

What to do next

1. Read Google's own generative-AI optimization guide directly rather than a summary — it states plainly what isn't required (chunking, llms.txt) and what still is (standard indexing and SEO fundamentals).
2. Ask one real, unbranded buyer question in AI Overviews and separately in AI Mode, log what each cites, and don't assume either result predicts the other.
3. Map the likely sub-questions behind your top few buyer prompts using the fan-out method in what is query fan-out, since a page answering only the surface question misses most of the retrieval opportunity.

ContextOwl sells one-time AI visibility reports. The AI Citation Snapshot is a free, single-question version of the same check.