Google AI Overviews vs AI Mode: Differences & Similarities

Google AI Overviews and AI Mode are two separate AI features inside Google Search. An AI Overview is the summary Google generates automatically above the classic results; AI Mode is a conversational surface in its own tab. The two agree on the answer and disagree on the source: across 540,000 query pairs, Ahrefs found the same URL cited by both just 13.7% of the time, while their answers matched in meaning 86% of the time.

This page works through where each surface appears, how much of the page an AI Overview takes, who can use each, which Gemini model drives each, what AI Mode can do that an AI Overview cannot, and what the 730,000-response study found. It also carries measurements I ran myself in August 2026 from Bangladesh: how often an AI Overview fires across twenty queries, how many URLs each surface cites, which source types actually appear inside one niche, and how AI Mode resolves a reference across a multi-turn chain. Where my figures disagree with the published ones, both are shown.

For site owners the citation gap is also a traffic question, covered on zero-click search. Published estimates of the click cost range widely, from Amsive’s 15.49% to Seer’s 70% depending on query type.

All model names, availability figures and citation statistics reflect Google Search as of 15th September.

What’s the difference between AI Overviews and AI Mode?

AI Overviews are automatic AI summaries that appear inline above standard Google search results, while AI Mode is a separate, opt-in conversational search tab powered by Gemini. AI Overviews and AI Mode are not two names for the same thing, but two distinct features Google deploys inside Google Search. An AI Overview is a short synthesized answer generated above the results for a single query; AI Mode runs a conversational session across multiple turns where a user asks follow-up questions. If you build pages for featured snippets, the change in mechanism is the thing to read next: both features replace the single-source featured snippet with multi-source synthesis. AI Overview sits above the results, and AI Mode sits behind its own tab.

Where do AI Overviews and AI Mode appear on the SERP?

An AI Overview usually appears as an inline block at the top of the classic results, but placement is not guaranteed. The two features do not occupy the same slot. An AI Overview loads on the main results page. On commercial queries a sponsored listing can render above it, and Ahrefs measured AI Overviews sitting outside position #1 in 8.64% of cases in a study published on 18 July 2025. AI Mode does not share the page at all: it replaces the classic results in its own tab. For what that tab contains and how it behaves, see AI Mode. So the reader now has both surfaces mapped.

How much of the page does an AI Overview take?

An AI Overview pushed organic result #1 to 937 px from the top of the page: 1.04 screen heights of scrolling before the first classic result became visible. That measurement comes from the query “what is google ai overviews” on a 1400 × 900 desktop viewport in Chrome, logged out, Bangladesh locale, on 24 August 2026. The block was measured in its default collapsed state, without expanding it.

On “buy organic honey online“, measured the same day on a 1400 × 900 desktop viewport, the AI Overview did not open the page. AI Overview usually appears at the top. Two organic listings rendered first, and the AI Overview only third. The same domain was both cited and mentioned in the AI Overview and held the top organic position.

Which triggers first, AI Overviews or AI Mode?

An AI Overview triggers first because Google generates it automatically on eligible queries, while AI Mode triggers only when the user opts in. Both surfaces appear for the same query at the same time, since the AI Overview loads inline automatically and AI Mode remains a separate entry point the user chooses. The trigger mechanism is automatic (no user action) for AI Overview and opt-in (user-initiated) for AI Mode. AI Mode handles longer, complex, multi-part queries while the AI Overview handles shorter informational ones.

On four of the twenty queries in my trigger test no organic result title rendered on the initial page load. That finding belongs with the traffic question, covered on zero-click search.

Who can use each feature, and where?

Both AI Overviews and AI Mode now reach roughly 200 countries, and the language gap between them has closed. AI Mode launched narrower than AI Overviews and caught up across 2025 and 2026; treating it as the limited-release option is out of date.

Google reported AI Mode passing one billion monthly users in May 2026, with query volume more than doubling each quarter since launch. AI Mode is no longer a surface to deprioritize on reach grounds.

With placement, trigger order and access set, the mechanism each feature uses to build its answer marks the next distinction.

How does each feature build its answer?

Each feature builds its answer by grounding in a Gemini model, yet retrieves and reasons differently: AI Overview builds a single-pass summary while AI Mode fans a query into sub-questions and reasons across them. AI Mode does more retrieval and reasoning work before answering. AI Mode runs query fan-out (breaking one query into multiple sub-queries), so it gathers more candidates. AI Mode taps live sources at answer time, while AI Overview leans on pre-indexed web results. Both features retrieve at the passage level (scoring individual passages, not whole pages) from the same index but score and select candidates differently through grounding (tying the generated answer to retrieved sources) and RAG retrieval (retrieval-augmented generation). How those signals are weighted per surface is covered on AI ranking signals. AI Mode’s fan-out pulls a wider candidate pool.

AI Mode and AI Overviews may use different models and techniques.

Source: AI features in Search, Google Search Central

Which Gemini model powers each feature?

AI Overviews run on Gemini 3 and AI Mode runs on Gemini 3.5 Flash, so the two surfaces have not shared a model since January 2026. Google changed the AI Overview default first and the AI Mode default four months later; each change moved one surface without moving the other.

Google AI Pro and Ultra subscribers can select Gemini 3.7 Flash from the model menu in the AI Mode “Ask anything” bar. Free users stay on Gemini 3.5 Flash. Different models on the two surfaces is the mechanical reason the citation sets diverge, and it is why a page that wins one does not automatically win the other.

How does an AI Overview build its answer?

An AI Overview scans the pre-indexed results for a query, identifies the themes across them, and synthesizes one answer in a single pass. It runs on Gemini 3 and retrieves at the passage level, so it lifts a paragraph or a sentence rather than summarizing a page as a whole. That is why a page is often cited for a section its author considered secondary.

Google applies query fan-out here as well as in AI Mode, issuing related searches across subtopics before composing the answer.

Issuing multiple related searches across subtopics and data sources.

Source: AI features in Search, Google Search Central

An AI Overview does not appear for every query. Google generates one only when its systems judge it adds something beyond the classic results.

Why does query fan-out change what AI Mode cites?

Query fan-out changes what AI Mode cites because AI Mode expands one query into many sub-queries, so it retrieves and cites a document set that the narrower AI Overview never touches. AI Mode fans out a query into more sub-queries, so it pulls more documents than AI Overview reads from the same index. AI Mode leans harder on encyclopedic and forum sources than AI Overview does. Per Ahrefs’ analysis of 730,000 response pairs, Wikipedia appears in 28.9% of AI Mode citations versus 18.1% of AI Overview citations; Wikipedia appears in 10% more AI Mode citations, and Quora appears 3.5x more often in AI Mode; Reddit follows the same forum-heavy pattern. Both features fan out, but AI Mode fans out further. Depth per feature is covered on query fan-out.

Across five AI Overview Optimization queries tested in late August 2026 from a clean profile in Bangladesh, the AI Overview cited 42 URLs in total and AI Mode cited 65 on the same queries: 1.55 times as many.

Counted at domain level. Ahrefs’ 13.7% overlap figure is URL-level; the two are not directly comparable.

This wider retrieval is why the two features’ cited URLs diverge so sharply.

Which sources does each feature actually cite?

Across the five queries, not one of 107 citations came from an encyclopedic source. No Wikipedia, no Britannica, on either surface, on any of the five queries. Ahrefs’ cross-niche dataset puts Wikipedia at 18.1% of AI Overview citations and 28.9% of AI Mode’s; inside this niche the figure was zero.

Published guidance often puts the AI Overview at two to five sources. In five queries measured here it ranged from 6 to 10, with a median of 9.

The forum and UGC lean Ahrefs describes did appear, but not in the form the general data implies. Reddit, Quora and Stack Exchange were absent. The user-generated citations were YouTube videos present in four of the five queries alongside LinkedIn Pulse articles and Medium posts. That is practitioner-published material carrying a named byline, not anonymous community discussion. For a professional niche, the working read is that these systems reach for identifiable people publishing under their own name rather than for threads.

That lines up with the cross-niche picture in one respect. Ahrefs found YouTube holding the top position among AI Overview citations, cited more often than encyclopedic sources, and AI Overviews citing videos and core pages roughly twice as often as AI Mode. My five queries showed the same direction inside a professional niche: video present, forums absent.

For a page owner the read is a format instruction. Video and a strong core page serve the AI Overview side; depth and named-author publishing serve both.

On the one query where both lists were captured URL by URL, does schema markup help ai search, every one of the AI Overview’s six URLs appeared in AI Mode’s list of fourteen, holding positions 1, 2, 5, 6, 12 and 14, in the same relative order. One query is not a rule, but the direction is worth recording: it points at a single retrieval pass being truncated for the AI Overview rather than two retrievals diverging.

That reading also reframes the share gap. If AI Mode retains the AI Overview’s sources and appends more, a source type can gain share without anything being dropped. The composition shifts because the denominator grows, not because the retrieval target changed.

What can AI Mode do that AI Overviews can’t?

AI Mode can query live data sources and run multi-step agentic tasks; an AI Overview can do neither. Both surfaces now take follow-up questions, so conversation is no longer the dividing line. What separates them is what each can reach and how far each can act.

Both surfaces render answers in more than one shape:

• paragraph summaries
• bulleted lists
• tables
• product cards
• link cards

Live data

AI Mode queries live sources at answer time. Google names five:

• Shopping Graph: current price and stock for a listed product
• Stock market data: a quote as of the moment of the query
• Restaurant availability: a table at a named time
• Maps: travel time between two named places
• Knowledge Graph: entity attributes Google already holds

An AI Overview draws on the pre-indexed crawl, so it cannot answer a question whose value changed this morning.

Multimodal input

AI Mode accepts voice and image input during an active session and continues a search started in Google Lens. An AI Overview is text-triggered only.

Agentic, multi-step tasks

Agentic search means the system completes a sequence of steps for the searcher rather than describing them. AI Mode runs these; AI Overviews return text only.

Deep Search, Google’s longer research workflow inside AI Mode, sits behind a Google AI Pro or Ultra subscription, as do the agentic workflows. The free tier of AI Mode does not include them, and Search Labs enrollment is where the newer surfaces appear first. That gating matters before you treat agentic search as a mass-market surface.

AI Mode uses product and link cards more often, because its longer answers have room for them.

Which feature supports follow-up questions?

Both AI Overviews and AI Mode support follow-up questions, and have done since 27 January 2026. Google added a follow-up field directly beneath the AI Overview on the results page and made Gemini 3 the default AI Overview model on the same day. Before that date, only AI Mode held context across turns.

What remains different is where the conversation runs. A follow-up typed under an AI Overview moves the session into AI Mode, carrying the original query and its context. AI Mode holds the thread from the first question; the AI Overview hands it over on the second. Google confirmed at I/O on 19 May 2026 that the hand-off is live on desktop and mobile worldwide.

Because the hand-off carries context but re-runs retrieval, a page cited in the AI Overview is not automatically carried into the AI Mode turn that follows.

How does AI Mode resolve a follow-up reference?

An ordinal in a follow-up binds to the nearest enumeration, not to the first list in the chain. Across four multi-turn chains run in AI Mode on 26 August 2026 from a clean profile in Bangladesh, thirteen follow-up turns produced no request for clarification and not one case of the model resolving a reference to the wrong entity. Whether context carries is not the useful question. What the reference binds to is.

Ordinal references bind to the most recent list, not the first. In one chain the opening answer set out five ranking factors; the second turn reorganized them into three groups; the third turn asked “how would I measure the second one?” and AI Mode measured the second item of the reorganized list, not the second of the original five. Asked identically a second time, it resolved the same way. In another chain, where no intervening list had been produced, “how do i confirm the third reason?” resolved correctly to the third item of the opening answer. The rule that fits both: the ordinal attaches to the nearest enumeration, and any answer that produces a numbered list silently becomes the new anchor.

Ambiguous possessives were neither resolved nor queried. After a two-tool comparison, “earlier you mentioned its editor, how does that work” returned an answer covering both tools rather than picking one or asking which was meant, and the following turn behaved the same way. Answering for every candidate is a defensible default, but it means an ambiguous pronoun costs the reader a longer answer instead of producing a correction.

Long-range recall held. A turn-five question asking which factors the model had earlier said could not be controlled reached back past the intervening turn and returned both correctly, unprompted and unrestated.

Repeating a question verbatim did not repeat the answer. Asked “if that shows nothing what is next” a second time, AI Mode read the repeat as a report that its earlier suggestions had also come up empty and escalated to a deeper set of causes rather than restating the first.

One turn produced nothing at all. “going back to the first one you listed, does it have an api” returned a generation failure on two consecutive attempts and ended that chain at turn four. One case is not enough to say whether long-range ordinal reference was implicated.

Google’s developer documentation for the underlying search technology calls this a stateful, multi-turn conversation, under Vertex AI Search.

Can an AI Overview expand into an AI Mode session?

Yes, typing into the “Ask anything” field under an AI Overview opens an AI Mode session with the original query already in context. Google shipped this as the replacement for the earlier “Dive deeper in AI Mode” link, which passed the query across without a conversational field. The searcher retypes nothing.

Each turn after the hand-off runs its own selection, so a page can be cited on turn one and dropped on turn two.

What does the 730K-response study show?

The 730K-response study shows that AI Overviews and AI Mode produce near-identical answers from almost entirely different sources. Ahrefs analyzed 730,000 response pairs and 540,000 query pairs comparing AI Overviews against AI Mode. The dataset came from Brand Radar, using September 2025 United States data. Ahrefs measured Jaccard similarity (word-level overlap), cosine similarity (semantic alignment), and entity overlap analysis (shared brands and people). Four findings define the gap between AI Overviews and AI Mode:

• 13.7% cited-URL overlap
• 16% word-level overlap
• 86% average semantic similarity
• AI Mode responses roughly 4x longer than AI Overviews

Why do AI Overviews and AI Mode reach 86% semantic similarity while citing matching URLs only 13.7% of the time?

Why do the two features cite matching URLs only 13.7% of the time?

AI Mode and AI Overview cite matching URLs only 13.7% of the time because query fan-out and longer answers push each feature toward a different source set. Ahrefs put citation overlap at 13.7% across 540,000 query pairs analyzed, rising to 16.3% across the top 3 citations from each system, with the two systems pulling from completely different sources 87% of the time on the same query. Ranking in one feature does not secure a citation in the other, because AI Overview and AI Mode maintain separate domain preferences. Each feature requires its own citation tracker; for the tracker setup itself, see separate tracking, which records citations per surface rather than a shared list.

AI Mode cites sources more consistently than AI Overview: 97% of AI Mode responses carry citations against 89% of AI Overview responses. AI Mode links out to a wider source set per answer than AI Overview.

Ahrefs offers three reasons. Responses four times as long need more grounding. Users in a research interface expect source transparency in a way that users skimming a results page do not. And AI Overviews surface for a wider range of edge cases that warrant no citation at all.

Those edge cases are identifiable: simple calculations, sensitive or adult subject matter, queries that redirect to a help center, and queries in unsupported languages. If a query returns no citations, check it against that list before concluding anything about your page.

Across five queries captured in late August 2026 in a single session, AI Mode cited a median of 13 URLs per response against 9 for AI Overview on the same queries: AI Mode ranging 12 to 14, AI Overview 6 to 10. AI Mode’s count held a narrow band regardless of query; the AI Overview’s did not.

On does schema markup help ai search, all 6 of the AI Overview’s cited URLs reappeared in AI Mode’s set of 14: complete overlap, against the 13.7% Ahrefs measured across 540,000 pairs. That is one query, not a distribution, and it sits far enough outside the published figure to be worth naming rather than generalizing from.

Ahrefs notes that 45% of AI Overview citations change between generations, so the citation pools behind the two surfaces may overlap more than a single snapshot shows. That caveat matters here: on the one query captured URL by URL, the overlap was complete.

Query fan-out, described above, widens the source set, and answer length compounds the gap.

How much longer is an AI Mode response than an AI Overview?

An AI Mode response runs approximately four times longer than an AI Overview. Ahrefs put the length gap at 4x. AI Mode responses average 3.3 entity mentions (named brands, people, and products counted in a response) against 1.3 for AI Overviews. AI Mode carries about 2.5 times more entities. Greater length gives more entity slots, but 4x length does not guarantee brand inclusion, because entity share determines visibility. AI Mode’s higher entity density rewards recognized brand entities, which raises the case for entity density tracking per surface.

Which feature surfaces more brands and entities?

AI Mode surfaces more brands and entities than AI Overview, naming about 2.5 times more entities per response. Per Ahrefs, AI Mode averages 3.3 entities per response while AI Overview averages 1.3. For the query “cloud storage alternatives”, the AI Overview named 7 brands while the AI Mode response carried 23 brand mentions.

A majority of AI Overviews name no brand or person at all. Per Ahrefs, 59.41% of AI Overview responses contain no brands or entities, 34.66% of AI Mode responses contain none, and 32.8% of all responses in the dataset mention no brands or people. These are usually queries where no brand is expected. Ahrefs cites “Nov 21 zodiac”, “revenue cycle” and “meditation before bed”. If your topic looks like one of those, brand mention is not the metric to chase.

AI Mode is the surface where recognized brand entities appear, because AI Overview leaves brands and people unnamed in most of its responses. AI Mode’s higher entity density rewards recognized brand entities more than AI Overview does. For the tracking method behind brand mentions per surface, see brand entity recognition.

More entities per answer raises a second question: if AI Mode says more, is it saying something different?

Are the two answers semantically the same despite different wording?

AI Mode and AI Overview answers are semantically the same, scoring 86% average semantic similarity despite sharing only 16% of their words. Per Ahrefs, 89.7% of response pairs scored above the 0.8 cosine similarity threshold. Ahrefs measured 730,000 response pairs. Semantic similarity (how close the meaning is, measured by cosine similarity) differs from word-level overlap (how many identical words appear, measured by Jaccard similarity): near-identical meaning, different wording. AI Mode and AI Overview reach matching conclusions through different vocabulary.

Agreeing on the answer is not the same as being right about it.

Is AI Mode more accurate than AI Overviews?

No study measures accuracy directly on either surface, so the honest claim is that AI Mode is better grounded, not proven more accurate. AI Mode responses run approximately 4x longer than AI Overviews and carry about 2.5 times more entities (average 3.3 entities in AI Mode versus 1.3 in AI Overviews). Grounding (tying answers to cited sources) supports the gap: 97% of AI Mode responses carry citations against 89% of AI Overview responses. Longer, more grounded AI Mode responses reduce the error rate, but neither surface is error-free, so verify against the primary source for the claim: the company’s own documentation, the study itself, the regulator’s page.

An AI Overview stated that the SEO professional Lily Ray was nine years old, a factual error about a real person produced with full confidence. Brand misrepresentation is a monitoring problem, not only a visibility problem.

AI Mode’s higher entity density rewards recognized brand entities, which points to entity density work, and AI Mode’s preference for encyclopedic sources raises the bar on topical authority. If the two surfaces agree in meaning yet share only 16% of their wording, which one earns the optimization budget?

Which feature should you optimize for if word overlap is only 16%?

Optimize for both surfaces with one page, because the 16% word overlap means neither is matching your phrasing anyway. You do not need two separate content strategies, you need one entity-dense asset and two retrieval targets. One strategy governs the content asset, and two targets govern retrieval.

The word-level overlap between AI Overview and AI Mode sits at 16% Jaccard word overlap (word-for-word wording match), expressed also as a 0.16 Jaccard similarity score, with 2.51% of pairs sharing an identical first sentence and 0.51% of pairs fully identical. Brand carry-over runs higher. A brand present in AI Overview also appears in AI Mode 61% of the time, and in that 61% AI Mode includes every AI Overview entity and adds more.

In practice: an AI Overview names Mayo Clinic as the health authority for a query, and the AI Mode answer keeps Mayo Clinic but adds Cleveland Clinic and WebMD. The original source survives; it shares the answer with more competitors.

Wording rarely transfers across the two surfaces. Meaning and entities largely do. So the optimization instruction stands: build the content asset for entity coverage and semantic completeness rather than for exact phrasing. Ahrefs measured 1.3 named entities per AI Overview response and 3.3 per AI Mode response. A page that names fewer entities than the answer it hopes to appear in is unlikely to supply one. The verdict: one content asset, an entity-dense answer-first page, aimed at two retrieval targets, Google AI Overview and AI Mode.

Sub-question coverage starts on the SERP itself. List the questions a searcher asks before and after yours. Google exposes these directly: the People also ask block and the follow-up prompts under an AI Overview. Read them off the SERP for your target query rather than guessing. Give each its own heading and answer it in the first sentence beneath.

Referral behavior differs between the two. A SimilarWeb study of more than 100,000 AI Mode users found visitors arriving from AI Mode spent longer on site and viewed more pages per session. The same shift changes how listings disappear from view. AI Mode suppresses classic organic listings more completely than AI Overview does, and the traffic side of that is handled on Zero-Click Search.

How does AI Overview Optimization differ from AI Mode optimization?

AI Overview Optimization differs from AI Mode optimization because each surface cites a different set of URLs. The two systems share 86% semantic similarity but only 13.7% citation overlap. AI Overview Optimization targets passage clarity and explicit brand naming, because 59.41% of AI Overview responses contain no brands or entities. AI Mode optimization targets sub-query coverage through query fan-out (breaking a query into related sub-questions).

Four on-page changes map to specific surfaces.

Both surfaces use query fan-out; AI Mode simply fans out further and rewards it more. Explicit brand naming serves both surfaces, since a brand carries over 61% of the time.

AI Mode raises the topical-authority bar by preferring encyclopedic content, with Wikipedia appearing in 28.9% of AI Mode citations versus 18.1% of AI Overview citations. Each surface needs its own citation tracker because cited URL sets barely intersect, so use a separate citation tracker per surface.

Do AI Overviews and AI Mode share the same eligibility rules?

The eligibility rules are the same for both surfaces; the selection that happens afterwards is not. A page must be indexed and snippet-eligible before either surface will consider it. That gate is one step. Choosing which of the eligible pages to cite is a second step, and it is where the two surfaces part company.

The distinction matters because the two steps are routinely treated as one. Passing the gate is necessary and not sufficient: Ahrefs measured a 13.7% URL-level citation overlap on the same queries, meaning that most of the time both surfaces looked at the same eligible pool and picked differently.

My own measurement points at something narrower than “two different retrievals”. On the one query captured URL by URL, every AI Overview citation reappeared in AI Mode’s larger set in the same relative order — which looks less like divergent selection than like one retrieval pass truncated for the AI Overview. One query is not a distribution, but if that pattern holds, the work is not to win two selections but to survive one truncation.

Technical work buys eligibility on both surfaces at once. Everything after that is fought on rank within a single list. For the crawl and index side of the gate, the technical half is covered on AI Overview eligibility.

How much traditional ranking still drives AI citation is contested: Botify put 75% of AI Overview citations inside the top 12 organic results, while other analyses found far less. If you want that argument in full, it is on do rankings still matter.

How the measurements on this page were taken

Every first-party figure here was captured from a clean desktop profile in Rajshahi, Bangladesh, logged out, in late August 2026. Three separate tests:

The twenty queries were chosen to span intent types rather than to mirror search volume, so the trigger rate runs higher than a random sample would produce. Citation counts are domain-level; Ahrefs’ 13.7% is URL-level, and the two are not directly comparable. Viewport measurements were taken at 1400 × 900 in Chrome. Sample sizes are small and stated at each finding.

Everything above covers how the two surfaces differ and what that changes about a page you are writing today. The questions below sit one step outside that: they come up whenever the two are compared, but each belongs mainly to a different part of AI Overview Optimization, and each links to the page that covers it properly.

Why doesn’t an AI Overview appear for my query?

Google generates an AI Overview only when its systems judge one adds value beyond the classic results, so many queries never trigger one. Absence is a decision about the query, not a verdict on any single page.

AI Overviews are only shown when additive to classic Search

Source: AI features in Search, Google Search Central

My own trigger test lines up with that. Across twenty queries run from a clean desktop profile in Rajshahi, an AI Overview generated on nineteen. The exception, “dhaka to sylhet train time“, returned a specialized transit layout with a route and schedule table instead. The twenty split into seven short informational queries, seven long-tail questions and six commercial queries, and were chosen to span intent types rather than to mirror search volume, so the rate here runs higher than a random sample would produce.

When an AI Overview does appear, its position varies too: Ahrefs measured it sitting outside position #1 in 8.64% of cases. Checking a query once tells you little. The method is in how to check whether your page is cited.

How do I check whether my page is cited in both?

Run the same query twice in one session (once on the All tab and once on the AI Mode tab) and compare the two citation sets. Citations are stochastic: the same query in two sessions can return different sources, and Ahrefs found that 45% of AI Overview citations change between generations. A single check tells you almost nothing; the pattern across several queries and several sessions does.

1. Open an incognito window and set the location to the market you target.
2. Run the query on the All tab. If an AI Overview appears, expand the citation panel and record every cited domain.
3. Switch to the AI Mode tab, run the same query, record every cited domain again.
4. Compare: which domains appear in one, in both, in neither.
5. Repeat across five to ten queries, on two separate days.

Record whether you are counting domains or URLs and keep to one. The two are not comparable, and mixing them is the most common way this measurement goes wrong. Ahrefs’ 13.7% figure is URL-level.

Running this across five queries in my own niche produced the counts in which sources does each feature actually cite.

What do people get wrong about these two surfaces?

Four errors recur, and all four come from treating the two surfaces as one product at different lengths.

• Assuming a citation in one means a citation in the other. URL-level overlap is 13.7%. Track them separately.
• Checking one query once. Citations vary between sessions and between generations.
• Reading a missing AI Overview as a failure of your page. Google often decides one would add nothing for that query.
• Applying cross-niche citation averages to your own niche. In my five-query test not one of 107 citations came from an encyclopedic source, against Ahrefs’ cross-niche 18.1% and 28.9%.

How do Google’s AI search features compare to ChatGPT?

Google’s AI Overview and AI Mode ground answers in live web search results, while ChatGPT and Gemini answer from a trained model unless a browsing mode is enabled. AI Mode is a Google Search surface grounded in web results, distinct from the standalone Gemini app. ChatGPT blends several ranking signals through Reciprocal Rank Fusion, a method that merges separate ranked lists into one. Perplexity weighs freshness and embedding similarity more heavily than links. Google’s features are served by default inside search, which drives reach, while adoption of the opt-in AI Mode lags the assistants users already open directly. Default-served Google AI Overview summaries feed Zero-Click Search, so AI Overview Optimization targets those placements.

Early adoption was uneven: a SimilarWeb study of more than 100,000 users in mid-2025 found over half tried AI Mode once and did not return. That picture has since changed. Google reported AI Mode passing one billion monthly users in May 2026, with query volume more than doubling each quarter since launch. Google Search averages 5.6 searches per session against AI Mode’s 2 to 3.5 searches per session.

AI Overviews and AI Mode agree on the answer and disagree on the source, so one page has to survive two selections. Write the answer once in a form both surfaces can lift: a direct opening sentence under every heading, named entities rather than pronouns, figures carrying units and dates. Then measure both surfaces yourself rather than importing cross-niche averages — in my niche the encyclopedic share was zero against a published figure of 18.1%. The optimization checklist for the AI Overview side is on how to optimize for AI Overviews.