Updated
Updated What changed on
Fact-checked Every model, launch date and control re-checked on Google's own pages, and every Ahrefs and BrightEdge figure on the original studies, on 14 October 2026, along with the FAQ and How-to rich result status and Google's schema guidance for AI features. The 18% to 48% prevalence range was last checked in August 2026 and is dated as such.
- Google-Extended does not control either surface, so I corrected that and added the Search Console opt-out that does, live for all sites since 31 August 2026.
- AI Mode now defaults to Gemini 3.5 Flash (since 19 May 2026), while AI Overviews stays on Gemini 3.
- Removed the unsourced 47%, 53% and 71% UK trigger rates and the 5 to 7 sub-query count, neither of which I could trace to a primary source.
- Corrected the Ahrefs figures: rank correlation is 0.347, not 0.65, and AI Overviews name 1.3 entities per answer, not 1.1.
- Added the new Search Console report that shows AI Overviews and AI Mode impressions together.
- Added that a follow-up under an AI Overview now carries on in AI Mode, on desktop and mobile worldwide.
- Corrected the Ahrefs citation figure: 37.9% of AI Overview citations sat in the first 10 results blocks, not the top 10 rankings, and 31.0% were beyond the top 100.
- Removed the advice to add FAQPage and HowTo schema. Google stopped showing FAQ rich results on 7 May 2026, retired How-to rich results in September 2023, and says no special schema is needed for its AI features.
AI Overviews vs AI Mode: what each one actually does.
The 30-second verdict
Both are AI answers inside Google Search, but they no longer run on the same model. AI Overviews defaults to Gemini 3 and AI Mode to Gemini 3.5 Flash. AI Overviews appear above organic results without the user asking, though how often is contested: published estimates, last checked in August 2026, run from 18% to 48% of searches. AI Mode only opens when someone chooses it, from its tab or by asking a follow-up under an AI Overview. Optimise for AI Overviews first because of reach. Much of the groundwork is shared with AI Mode, but the sources they cite mostly differ, so check both. Google-Extended controls neither. The opt-out is a Search Console setting.
What is measured about AI Overviews
of Google searches trigger an AI Overview, depending entirely on whose panel you read.
How it was measured: Pew measured 18% on real-user browsing data. Ahrefs measured 48% of its keyword set in March 2026. Similarweb reported 43%, up from 15% a year earlier, via TechCrunch on 27 July 2026. The panels sample different query mixes, so the spread is a methodology artefact. Anyone quoting a single figure as the number is overstating what is known.
Sources disagree on this one. The range is the honest answer.
of URLs cited in Google AI Overviews appeared in the first 10 results blocks for that query. Ahrefs counts blocks, not organic results, so this is not the same as ranking in the top 10.
How it was measured: 863,000 keyword SERPs and 4 million AI Overview URLs. Ahrefs measures the top 10 as blocks, counting ads, featured snippets and video packs as separate blocks.
was Ahrefs' July 2025 figure. It counted only the top 3 citations per AI Overview, and Ahrefs has since changed its parsing, so the two years are not like for like.
How it was measured: 1.9 million citations from 1 million AI Overviews, analysing only the top 3 most visible citations in each one. Ahrefs notes it improved its parsing between this study and the 2026 update. Both differences mean the fall to 37.9% cannot be read as a like for like change.
Side-by-side comparison
 AI Overviews |  AI Mode | |
|---|---|---|
| Position in Google Search | Above traditional organic results, shown when Google judges it adds something | Its own tab, or a follow-up from an AI Overview (desktop and mobile since 19 May 2026) |
| Default model (checked Oct 2026) | Gemini 3 (since 27 Jan 2026) | Gemini 3.5 Flash (since 19 May 2026). Google AI Pro and Ultra subscribers can also use Gemini 4 Argon (since 13 Oct 2026) |
| How often it shows | Contested. Published estimates, last checked in August 2026, run from 18% to 48% of searches | Only when the user chooses AI Mode |
| Monthly users (Google's figures, 2026) | Over 2.5 billion | Over 1 billion |
| UK launch | Rolled out from 15 August 2024 | 28 July 2025 |
| Query handling | May use query fan-out. Google gives no sub-query count | Query fan-out: Google describes "a multitude of queries" run at once. No count given |
| Output format | Concise paragraph + linked citations | Conversational, about 4x longer (Ahrefs), follow-up enabled |
| Agentic actions | None. Follow-ups hand over to AI Mode | UK restaurant search since April 2026: finds tables, then links you to partners such as TheFork, SevenRooms and OpenTable to finish the booking |
| Rank correlation (Ahrefs) | Spearman 0.347 between top-10 rank and a top-3 citation (July 2025). Even #1 pages made the top three citations only about 50% of the time | Not covered by that study |
| Opt-out control | Search Console > Settings > Search generative AI (all sites since 31 Aug 2026). Google-Extended does not apply | Same setting. Google-Extended does not apply |
| Search Console reporting | Generative AI performance report: impressions only, combined with AI Mode | Same report, combined with AI Overviews |
| Optimisation priority for UK B2B | 1st (reach) | 2nd (depth + agentic future) |
Where they overlap
Both answer from Google's search index, and Google's AI features guide says there are no extra requirements to appear in either. Neither is controlled by Google-Extended: Google's crawler documentation says it "does not impact a site's inclusion in Google Search". It covers Gemini training instead. The opt-out for both surfaces is one Search Console setting, Search generative AI, live for all sites since 31 August 2026. Both reward content depth on the query, named-author credibility signals and clean extractable structure. Ranking helps but guarantees little: Ahrefs found only 37.9% of AI Overview citations sat in the first 10 results blocks (March 2026), and 31.0% were beyond the top 100. Schema such as Article is hygiene that quality sites carry anyway, because it helps rich-result eligibility. Google stopped showing FAQ rich results on 7 May 2026, retired How-to rich results in September 2023, and says no special schema is needed for its AI features. Expect ranking and depth to do the citation work (per Ahrefs' May 2026 controlled test).
Where they differ
Google says the two "may use different models and techniques", and since May 2026 they run on different default models. Both may use query fan-out. Google describes AI Mode as breaking a question into subtopics and running "a multitude of queries" at once, but gives no count for either surface. My view: a page that covers a parent topic and its related sub-questions on one URL gives AI Mode more chances to cite you for one prompt.
They also lean on different sources. In BrightEdge's August 2026 US data on consumer product and finance prompts, google.com itself made up 44.9% of AI Mode citations, against 9.9% for AI Overviews.
The other difference is action. Since April 2026 AI Mode in the UK can find restaurant tables and link you to a partner such as TheFork or OpenTable to finish the booking. AI Overviews has nothing like it. For travel, hospitality, restaurants and any sector where agentic flows matter, AI Mode is the surface to prepare for. For information-only sectors (consultancy, professional services, B2B technical), AI Overviews reach matters more in the short term.
Practical pick order
For 9 out of 10 UK brands the order is: AI Overviews first (reach), AI Mode second (depth + future agentic surface). The work is largely shared so cover both unless you genuinely have nothing to expose to an agent.
What 730,000 Ahrefs response pairs revealed
Ahrefs' December 2025 study compared AI Overviews and AI Mode answers to the same queries: 540,000 query pairs for citations and 730,000 for content similarity. The data is from the US in September 2025, so read it as direction for UK sites, not a UK measurement. Five findings that matter for optimisation:
- URL overlap was only 13.7% - of the sources cited by AI Overviews and AI Mode for the same query, fewer than one in seven matched. In Ahrefs' words, 87% of the time the two pull from completely different sources. Treating them as one surface misses most of the citation picture.
- Word overlap was just 16% - the two surfaces phrase answers very differently from the same source material. Different sentences get extracted, so writing for one specific extraction pattern under-serves the other.
- Responses averaged 86% semantic similarity, and 89.7% of pairs scored above 0.8 - the underlying answer is largely the same; presentation, citation set and length differ sharply.
- AI Mode responses are about 4x longer and name more people and brands (3.3 entities per answer against 1.3) - which means AI Mode is the bigger surface for brand-mention visibility, not just citation links.
- 59% of AI Overviews contain no brands or entities at all - AI Overviews often answer with pure information; entity inclusion is much more common in AI Mode. For brand-led SEO the AI Mode surface is the higher-yield work.
The practical read: optimise for AIO first because of reach, but treat AI Mode separately for brand mentions, because that is where far more brands get named. Ahrefs also found that a brand mentioned in an AI Overview has a 61% chance of appearing in AI Mode too.
You can now see both in Search Console's generative AI performance report, live for all sites since 31 August 2026. It counts impressions from AI Overviews and AI Mode together, and its help page describes no split between the two and no clicks. That traffic also still sits inside the normal Performance report totals.
More on AIO vs AI Mode.
Is AI Mode replacing AI Overviews?
Not for now. AI Overviews stay above traditional search results, and AI Mode only opens when you choose it. The two are now joined: since 19 May 2026 you can ask a follow-up under an AI Overview and carry on in AI Mode, on desktop and mobile worldwide. They no longer share a default model. AI Overviews defaults to Gemini 3 and AI Mode to Gemini 3.5 Flash.
Which should I optimise for first?
AI Overviews. They appear without the user doing anything, while AI Mode only shows when someone chooses it. How often AI Overviews trigger is contested: published estimates, last checked in August 2026, run from 18% to 48% of searches. Much of the groundwork is shared, so covering AI Mode as well costs little extra. Check both, though, because the sources they cite mostly differ.
Do they use the same retrieval mechanism?
Both answer from Google's search index. Google says both may use query fan-out, running several related searches across subtopics to build an answer, and that the two may use different models and techniques. Google gives no sub-query count for either. My view: content that covers a topic and its related sub-questions on one URL gives AI Mode more to cite.
Does Google-Extended control both?
No. Google's crawler documentation says Google-Extended does not affect whether a site is included in Google Search. It covers training of Gemini models and grounding in Gemini Apps and Vertex AI. To keep your site out of AI Overviews and AI Mode, use the Search generative AI setting in Search Console, live for all sites since 31 August 2026. Google says that setting is not a ranking signal elsewhere in Search, but you get no traffic or impressions from those features.
When did AI Mode launch in the UK?
Google started rolling AI Mode out in the UK on 28 July 2025. Since April 2026 AI Mode in the UK can find restaurant tables and link you to partners such as TheFork, SevenRooms and OpenTable to finish the booking. AI Overviews reached the UK earlier, rolling out from 15 August 2024.
Jason Burns
Independent UK SEO, GEO and AI consultant. In search since 2007. Portfolio includes 3M, BlackRock, Unilever and E.ON. Owner of SEO Moves Ltd since 2014.
More about Jason →See your pages against both surfaces together.
Send three priority queries. I run each through AI Overviews and AI Mode separately and send back which surface cites you on which query and where the gap is.
Test both surfaces