How to Optimize for Google AI Mode

Google AI Mode is a tab in Google Search that answers your question with a written response instead of a list of ten blue links. You find it next to All, Images and News, or by tapping AI Mode on the Google homepage and in the Google app.

Google introduced it on 5 March 2025 as an early experiment in Search Labs, built on a custom version of Gemini 2.0. Google said it was “particularly helpful for questions that need further exploration, comparisons and reasoning” (Google). On 20 May 2025 it rolled out to everyone in the US with no Labs sign-up, now on Gemini 2.5 (Google).

Three things make AI Mode feel different from normal Google.

It answers, then links. The response comes first. Links sit in a source panel, inline next to the sentence they support, and sometimes in a short list at the end.

It remembers the conversation. A follow-up such as “which of those has a pool?” keeps the earlier context. Google’s I/O 2026 post says that as you dig deeper, “the links and supporting articles get even more relevant” (Google).

It handles long questions. People type whole situations into it. That’s the part that matters most for site owners, and we’ll come back to it.

Let’s make this concrete with one running example. Picture a small, hypothetical boutique hotel on the Chao Phraya River in Bangkok. A parent in Sydney opens AI Mode and types: “Which boutique hotel near the river in Bangkok is good for a family with two young kids, has a pool and is close to the BTS or a boat pier?”

No page ranks for that exact sentence. AI Mode doesn’t need one. It builds the answer from many pages, and the hotel either shows up in it or it doesn’t. Getting it to show up, and described correctly, is what optimizing for AI Mode means.

Query fan-out is Google’s name for breaking one question into several smaller searches and answering from the combined results. Google’s I/O 2025 post describes it as breaking your question into “subtopics and issuing a multitude of queries simultaneously on your behalf” (Google).

Google’s own example is simple. A question about fixing a weedy lawn can spawn separate searches about herbicides, chemical-free removal and prevention (Google Search Central). Each of those returns its own results, and the model picks passages from all of them.

Back to our hotel. The parent’s long question about a family-friendly riverside hotel might fan out into searches like the ones below. Google doesn’t publish its real sub-queries, so treat this as an illustration of the pattern.

How many searches does AI Mode run?

More than you’d guess, and the number is rising. Seer Interactive ran 501 prompts through the Gemini 3 API with Google Search grounding switched on and logged the fan-out queries. Prompts produced 10.7 fan-out queries on average, anywhere from 3 to 28, up 78% from 6.01 with Gemini 2.5 (Seer Interactive).

Google’s own descriptions point the same way. With Gemini 3, Google said Search can “perform even more searches to uncover relevant web content” and find content it previously missed (Google). Its Deep Search feature, built on the same technique, can “issue hundreds of searches” for a single research task (Google).

What fan-out queries look like

Seer’s data holds three lessons that shape everything else in this guide.

They are long-tail. The sub-queries averaged 6.7 words, and 95% had zero measurable global search volume. Classic keyword research would never surface them.

They often include a year. 21.3% of the fan-out queries included a year. Fresh, clearly dated content has an edge.

They often name brands. 26.4% of the queries named a brand. If a fan-out search already contains your brand name, your chance of being cited goes up, which is why your reputation off your own site matters so much.

Mike King, who built a fan-out simulator called Qforia, summed it up for Digiday: an AI Mode answer is a synthesis, not a copy of what ranks for one query. His tool typically shows 20 to 30 related queries for a single search (Digiday).

Want to hear it from Google? Robby Stein, VP of Product for Google Search, walks through AI Mode and query fan-out in this Google for Developers interview. The fan-out chapter starts at about 4:17.

Google for Developers · YouTube

Google for Developers · Building a frontier AI search experience · watch on YouTube

AI Overviews and AI Mode are both Google’s generative search, but they’re built for different moments and they cite different pages. Treat them as two separate surfaces.

Google’s documentation says AI Overviews are meant to give a quick overview of a complex topic and only show when Google thinks they add something, while AI Mode suits questions that need deeper exploration, reasoning or comparison. It also says the two can use different models, so the links can differ (Google Search Central).

Ahrefs measured how different. Comparing AI Overviews and AI Mode responses for the same US queries in September 2025, across 540,000 query pairs, it found:

• Only 13.7% of cited URLs overlapped between the two, rising to 16.3% for the top three citations.
• The answers still meant much the same thing: 86% average semantic similarity.
• AI Mode answers were roughly four times longer.
• AI Mode named 3.3 people, organizations or brands per answer, against 1.3 for AI Overviews.
• Only 3% of AI Mode answers had no citations, against 11% of AI Overviews.

Ahrefs also found Wikipedia made up 28.9% of AI Mode citations against 18.1% for AI Overviews, and Quora was cited 3.5 times more in AI Mode (Ahrefs). One fair caveat from the study itself: 45% of AI Overview citations change between generations, so a single snapshot understates how much the two pools can overlap over time.

The line between them is also getting blurrier. Since January 2026, tapping “Show more” on an AI Overview on mobile and asking a follow-up sends you straight into AI Mode (Search Engine Roundtable), and at I/O 2026 Google made follow-ups from AI Overviews into AI Mode available worldwide on desktop and mobile (Google).

So what does this mean in practice? Win the short summary and you still might be missing from the longer conversation that follows. The foundations are shared, though, and our guide on ranking in Google AI Overviews covers the overview side in depth.

AI Mode cites a wider set of pages than Google’s top 10, and it leans on video, forums and Google’s own properties. Three studies make the pattern clear.

Ranking in the top 10 helps, but it isn’t enough

Semrush checked 5,000 keywords in July 2025 and compared each platform’s citations with Google’s top 10 organic results for the same query. AI Mode overlapped by about 54% at the domain level and 35% at the URL level. AI Overviews overlapped far more, at 86% and 67%, and Perplexity most of all, at 91% and 82% (Semrush).

seoClarity found a similar gap for buying-intent searches. Across 1,000 transactional US queries in September 2025 and 12,011 AI Mode citations, only 28% of queries had at least one AI Mode citation from the top 10 organic results. Widening to the top 20 raised that to 81% of queries, yet only 19% of all citations came from the top 20 (seoClarity).

That last number is the hopeful one. Remember fan-out: AI Mode isn’t citing pages for your keyword. It’s citing pages for the ten narrower questions behind it.

Who gets cited most

The newest data we found comes from Surfer, which recorded 1,545,892 AI Mode answers for 69,392 prompts in October 2026, about 5.19 million citations in all (Surfer). The prompts are ones Surfer’s customers track, so it’s a big sample, not a random one.

Two things stand out. First, YouTube, Reddit and google.com together took 55% of the top 50’s citations, and Google-owned sites alone took 36.4%. Video and community content are central to AI Mode.

Second, and more useful for most businesses: the top 50 domains earned only 18.6% of all AI Mode citations. The other 81.4% went to a long tail of other sites. AI Mode isn’t a closed club of giant publishers. A clear, specific page from a small brand has real room to win citations.

Semrush’s 2025 study adds a format detail: 92% of AI Mode answers had a sidebar of links averaging about seven unique domains, against three for AI Overviews, and 13.49% of AI Mode queries included a local pack (Semrush). More slots per answer means more chances to be one of them.

In 2026 Google kept changing how AI Mode shows links, and nearly every change gave cited sources more visibility. Here’s the sequence, because it tells you where the clicks that remain will come from.

January 2026. Follow-up questions asked from an AI Overview on mobile started opening directly in AI Mode (Search Engine Roundtable).

February 2026. Groups of links started appearing in a pop-up when you hover over them on desktop, with more descriptive link icons inside the response (Search Engine Roundtable). Robby Stein announced it on X.

6 May 2026. Hema Budaraju, VP of Product Management for Search, announced five link changes: suggested articles at the end of responses, “Subscribed” labels for your news subscriptions, previews of forum and social posts with the creator’s name or handle, inline links placed beside the text they support, and website previews on hover (Google).

27 May 2026. Preferred Sources came to AI Overviews and AI Mode. Links from sites a user has chosen carry a “Preferred” badge, and Google says people are twice as likely to click a link with that label. Any site that publishes fresh content can be chosen, at the root domain or subdomain level (PPC Land).

June to August 2026. Search Console gained a Generative AI performance report and a control to opt out of generative AI features, first tested in the UK and rolled out to all websites worldwide by 31 August 2026 (Search Engine Roundtable).

15 October 2026. The latest batch, and the one most relevant to anyone optimizing for AI Mode today. Inline links now appear “higher up in responses more frequently.” Google is testing a section at the top of some answers for the “one or two official places on the web” to complete a task, with context on why each link is useful. End-of-answer links got thumbnails and a more clickable design, and Google said Preferred Sources usage had grown more than seven times since the start of the year (9to5Google).

Two caveats keep this honest. Google hasn’t published click data for any of these changes, so “more visible” doesn’t yet mean “more traffic.” And Search Engine Journal points out that Search Console counts all the links inside an AI Overview as a single position, so an inline link moving higher there won’t show up as a better reported position (Search Engine Journal).

What should you take from the timeline? Inline links reward passages that answer one specific claim well. The official-sites test rewards being the clear, authoritative page for your own brand, product or service. And Preferred Sources rewards publishers who give readers a reason to choose them. Note one dependency: to appear as a Preferred Source in AI Mode, your site must be included in generative AI features in Search Console (Search Engine Journal).

To be used in AI Mode, a page must be indexed and eligible to show with a snippet in Google Search. Google is explicit: “There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary” (Google Search Central).

Since mid-2026 there’s one more condition. Google’s AI optimization guide says the site must also be included in generative AI features in Search Console (Google Search Central). That’s the default, but it’s worth checking, because the setting can be changed by anyone with access to the property.

According to Search Console Help, an exclusion usually takes effect within a few days, and it doesn’t affect your ranking in regular results. It also doesn’t control AI training, which is handled separately by Google-Extended (Search Console Help). Google adds that sites which opt out “will not receive traffic or impressions from our generative AI features” (Google).

Then run the usual technical checks, because they matter as much here as in classic SEO.

Our free AI Crawler Checker shows in a few seconds whether your robots.txt blocks Googlebot or the main AI crawlers. For our hotel, the classic trap is a booking engine that loads room types and rates by JavaScript from a third-party domain. Google might render it, or might not. Putting the room names, sizes, bed setups and “from” prices in plain HTML removes the doubt.

Find your fan-out questions by working backwards from the full questions your customers ask, then listing every sub-question an answer would need. You can’t see Google’s real sub-queries, but you can predict most of them.

Seer’s finding that 95% of fan-out queries have no measurable search volume is the reason keyword tools fall short here (Seer Interactive). You need other sources.

Your own inbox and chat logs. Emails, LINE chats, WhatsApp messages and call notes hold the exact long questions people ask. For our hotel: “Is the pool safe for a four-year-old?”, “How far is the boat pier with a stroller?”, “Can you add a baby cot to a deluxe room?”

AI Mode itself. Ask AI Mode your customers’ big questions and note the subtopics it covers in the answer and the follow-ups it suggests. Those headings are a good map of the fan-out.

People Also Ask and related searches. Still one of the best free views of the sub-questions Google connects to a topic.

Forums and reviews. Reddit threads, Pantip for Thai audiences, TripAdvisor and Google reviews show the details people care about and the words they use.

Both languages. In bilingual markets, run the exercise in English and in the local language. Thai questions about the same hotel often emphasize different details, such as parking or the distance to a specific mall.

Then group the sub-questions into the facets an answer always needs: what it is, cost, location and access, who it suits, how it compares, timing, and proof. Each facet becomes a section, a page, or an FAQ. Our guide to prompt research walks through building and grouping the full list.

Write so that any single section makes sense when it’s lifted out of the page on its own. That’s the test AI Mode applies when it picks a passage for an inline link.

Google says you don’t need to chunk content into tiny pieces or rewrite it in a special AI style, and that’s true (Google Search Central). Clear writing for people is what works. In practice, though, clear writing for busy people and quotable writing for AI turn out to be the same thing.

A handful of habits make most of the difference.

Answer first. Put the direct answer in the first one or two sentences under each heading, then explain. Olaf Kopp, quoted by Digiday, recommends passages of two to four sentences for the key answer (Digiday). That’s a useful target, not a rule.

Use question headings where they fit. “Is the pool suitable for young children?” matches the way people ask and the sub-searches fan-out produces. Google says to organize pages with clear sections and headings.

Swap adjectives for facts. “Stunning pool” can’t be cited. “15 m rooftop pool with a separate 0.6 m kids’ pool, open 7:00 to 21:00” can. Prices or price ranges, sizes, times, distances, who it suits and who it doesn’t suit are exactly what AI Mode answers are made of.

Name things. Ahrefs found AI Mode answers mention 3.3 entities on average (Ahrefs). Name your brand, your location, the nearest landmarks and the products you use, so the model can connect your passage to the entities in the question.

Use tables and lists for real comparisons and steps. King’s patent analysis notes the model may consume “a list, table, or semantic triple” instead of the whole page. A room comparison table with size, beds, max guests and view is far easier to use than three paragraphs of description.

Our free AEO Content Checker scores a page for answer-first structure in your browser, and our guide to answer-first content has more before-and-after examples.

AI Mode can already write the generic answer from a hundred similar pages, so it has little reason to cite one more. What it can’t get elsewhere is what you know first-hand.

Google’s 2026 guide puts this at the center. It contrasts “commodity” content, meaning generic common-knowledge tips, with non-commodity content that offers an expert or experienced perspective, and it says content with a unique point of view, such as first-hand experience, stands out more than summaries of existing material (Google Search Central).

The second version could only come from someone who has done it. That’s the kind of passage AI Mode is short of, and it’s also the kind of detail parents actually want.

This is E-E-A-T applied to AI search. Show who wrote the page and why they know. Add a visible updated date that reflects real changes. Use your own photos and short videos, which Google says can add visibility, and follow normal image and video SEO practice. Our guide on E-E-A-T and how to improve it covers the signals in detail.

A practical way to find non-commodity material: ask your front-line staff. Reception teams, sales reps, nurses and support agents answer the same specific questions every day. Each answer they give, written down with numbers and caveats, is a passage nobody else has.

Keep the facts AI Mode is most likely to repeat accurate and visibly current. Stale prices or old opening hours are the fastest way to get described wrongly, or dropped.

Freshness has a measurable pull here. Seer found 21.3% of Gemini 3’s fan-out queries included a year, mostly the current or previous one (Seer Interactive). A page that clearly says “2026 rates” or “updated November 2026” matches those sub-searches better than one with no date at all.

A simple routine works for most businesses.

Monthly. Check prices, hours, availability and anything seasonal on your key pages. Change the visible updated date only when you’ve changed something real.

Quarterly. Re-run your prompt list in AI Mode and compare how it describes you against reality. Fix the pages that feed wrong details, and correct outdated profiles on other sites.

Whenever something changes. New service, new price, closed branch: update your site, Google Business Profile and major listings on the same day.

Structured data belongs in this routine too. Google says structured data isn’t required for AI search and there’s no special AI schema, but it’s still worth using for rich results, and it must match the visible text on the page (Google Search Central). A Hotel or LocalBusiness block that lists a different phone number from your contact page sends mixed signals. Our guide to schema markup for AEO covers which types are worth adding.

For our hotel, the most repeated facts are room types, the kids’ pool details, the walking time to the pier and the price range. Those four deserve a monthly check.

Measure AI Mode with three layers: Google’s own data in Search Console, your analytics, and repeated prompt checks. Each covers a gap in the others.

Search Console Performance report

AI Mode clicks and impressions are counted in the normal Performance report under the Web search type. Google’s rules: clicking an external link in AI Mode counts as a click, standard impression rules apply, position follows the same method as a results page, and a follow-up question in AI Mode counts as a new query (Search Engine Roundtable).

That last rule creates odd data. Suganthan Mohanadasan found AI Mode follow-ups such as “yes” and “yes go on” showing up as queries in his Search Console, with “yes” alone earning 110 impressions and six clicks over the period he studied (Search Engine Journal). If you see strange conversational queries, that’s likely why.

Generative AI performance report

Rolled out worldwide by September 2026, this report shows impressions, which pages appeared in AI features, countries, devices and trends over time. It does not include clicks or queries (Search Engine Roundtable), and it doesn’t separate AI Mode from AI Overviews. Still, it’s the only first-party view of which of your pages AI features actually use.

Repeated prompt checks

Neither report tells you whether your brand was named, how it was described, or which competitor was recommended instead. For that you need to run real prompts in AI Mode and record the answers. Because AI answers vary, one run isn’t enough. Ahrefs found 45% of AI Overview citations change between generations (Ahrefs). That’s why GrixTrack runs each prompt three times and records mentions and citations in Google AI Mode, ChatGPT and Gemini, and why we report rates over many prompts, never a single screenshot.

From those runs, track four numbers: mention rate (how often your brand is named), citation rate (how often your pages are linked), share of voice against named competitors, and accuracy (whether the description is right). Our guide on measuring AI visibility sets up the full system, and GrixTrack has a free plan with 10 keywords, 5 prompts, 1 project and 2 competitors if you want to start small.

Most sites see technical fixes register within days to weeks, page rewrites within a few weeks, and off-site reputation gains over several months. Here’s the order we’d work in.

Week 1: eligibility. Confirm your key pages are indexed and snippet-eligible, check the generative AI control is set to include, and fix any robots.txt, CDN or JavaScript issue hiding important content.

Week 2: baseline. Write 20 to 50 real customer questions, in every language you serve. Run each in AI Mode three times and record whether you’re named, cited, and how you’re described. Note which competitors and which third-party sites keep appearing.

Weeks 3 to 8: content. Map the fan-out for your top questions, then rewrite your 5 to 10 most important pages answer-first, with specific facts and first-hand detail. Fill the biggest gaps in your cluster.

Months 2 to 6: reputation. Work on the third-party sites your baseline showed AI Mode citing: YouTube, the forums in your niche, review platforms, lists and press.

Every month: measure and refresh. Re-run the prompt set, check the Search Console reports, and update the facts AI Mode repeats.

For our hotel, that might mean a rooms page rebuilt with a family-room table in week 3, a new “getting here with kids” page in week 4, a short pier-walk video on YouTube in month 2, and a refreshed TripAdvisor and Google Business Profile listing with current photos the same month.

None of this is exotic. It’s good SEO pointed at a different target: not one keyword, but every question behind it. If you want the same thinking applied across Gemini, ChatGPT and Perplexity, read how to optimize for Gemini next. And if you’d like a second pair of eyes, our AEO team can run your key questions through AI Mode and show you exactly where you stand.

1. Google Search Central: AI features and your website (updated 10 Dec 2025)
2. Google Search Central: Optimizing your website for generative AI features on Google Search (updated 10 Jul 2026)
3. Google: Expanding AI Overviews and introducing AI Mode (Robby Stein, 5 Mar 2025)
4. Google: AI in Search, going beyond information to intelligence (Elizabeth Reid, 20 May 2025)
5. Google: Google Search with Gemini 3, our most intelligent search yet (18 Nov 2025)
6. Google: A new era for AI Search, I/O 2026 updates (Elizabeth Reid, 19 May 2026)
7. Google: 5 new ways to explore the web with generative AI in Search (Hema Budaraju, 6 May 2026)
8. Google: New opportunities, control and insights for website owners (3 Jun 2026, updated 31 Aug 2026)
9. Search Console Help: Search generative AI control
10. Google Thailand blog: AI Mode available in Thai (8 Oct 2025)
11. 9to5Google: Google puts links higher in AI Mode and AI Overviews (15 Oct 2026)
12. Search Engine Journal: Google puts inline links higher in more AI search responses (15 Oct 2026)
13. PPC Land: Google brings Preferred Sources into AI Overviews and AI Mode (27 May 2026)
14. Ahrefs: AI Overviews vs AI Mode, 540,000 query pairs (15 Dec 2025)
15. Semrush: AI Mode comparison study, 5,000 keywords (21 Jul 2025)
16. Semrush: Google AI Mode SEO impact, 69M sessions (30 Jul 2025)
17. seoClarity: AI Mode citations vs organic rankings, 12,011 citations (Sep 2025 data)
18. Surfer: The 50 most-cited domains in Google AI Mode (Oct 2026 data)
19. Seer Interactive: Initial research on Gemini 3 query fan-outs (21 Nov 2025)
20. iPullRank: Early referral data on Google's AI Mode (26 Sep 2025)
21. Digiday: WTF is query fan-out in Google's AI Mode (19 Jun 2025)
22. PPC Land: SEO expert reveals Google AI Mode patent details (May 2025)
23. Search Engine Roundtable: Search Console now counting AI Mode clicks, impressions and positions (17 Jun 2025)
24. Search Engine Roundtable: Google AI Mode expands to 35 new languages and 40 countries (9 Oct 2025)
25. Search Engine Roundtable: Hover pop-up links in AI Overviews and AI Mode (18 Feb 2026)
26. Search Engine Roundtable: AI Overviews follow-up questions go to AI Mode (28 Jan 2026)
27. Search Engine Roundtable: Search Console AI performance report and AI control go global (1 Sep 2026)
28. Search Engine Roundtable: Google Search using Gemini 3.8 Flash in AI Mode (2 Sep 2026)
29. Search Engine Journal: The AI conversations leaking into your Search Console (17 Aug 2026)
30. Google for Developers on YouTube: Building a frontier AI search experience (Robby Stein, Jul 2025)
31. Rajan Patel (Google) on X: AI Mode expands to 36 new languages and 40+ countries (7 Oct 2025)
32. Robby Stein (Google) on X: hover pop-up links in AI Overviews and AI Mode (17 Feb 2026)

Keep reading

• AI platformsHow to Optimize for AI Search: A Step-by-Step Guide
• AI platformsHow to Rank in Google AI Overviews
• AI platformsHow to Optimize for Gemini
• Content & technicalPrompt Research: How to Find AI Search Keywords