Google AI Mode: What It Is and How It Picks Sources

Google AI Mode is a tab inside Google Search that answers a question with a generated response rather than a ranked list of links. It can use query fan-out to run related searches across subtopics, then surface supporting links alongside the generated response.

AI Mode passed one billion monthly users in its first year. Queries have more than doubled every quarter since launch, according to Google's I/O announcement of 19 May 2026.

Pew Research Center measured the other side. Users who met an AI summary clicked a result in 8% of visits, against 15% with no summary, in its March 2025 browsing study of 900 US adults.

That second figure describes AI Overviews, not AI Mode. No clickstream study of comparable scale has been published for AI Mode. This article marks that line every time it matters, and every behavioural claim below was checked against Google's own documentation on 20 August 2026.

Key Takeaways

• AI Mode is a surface, not a product: it is a tab in Search running a custom Gemini model against Google's index, ranking systems, and Knowledge Graph, so being indexed and eligible to appear in Google Search with a snippet is the technical entry requirement.
• Query fan-out is documented: Google states that AI Mode splits a question into subtopics and searches each one at the same time, which is why a page can be cited for a query it never targeted.
• Contrarian: almost everything sold as AI Mode optimisation is unfalsifiable. Google publishes no additional technical requirement beyond ordinary Search eligibility, so the honest lever list is short, dull, and already on your roadmap.
• Measurement arrives as impressions first: Search Console's generative AI report, launched 3 June 2026, reports impressions with no clicks, no query dimension, and no split between AI Overviews and AI Mode.
• Two identical searches can diverge: personalisation, available model choices, and the way a question is framed can change the response, which makes single-session screenshots worthless as evidence.

What Is Google AI Mode?

The short answer to what is AI Mode is a tab inside Google Search. Google describes it as its most powerful AI search experience, reached at google.com/ai, from the AI Mode tab, or from the Google app.

It runs on a custom Gemini model wired into Search itself. Google's May 2025 explainer puts AI Mode on the same ranking and safety systems as classic Search. It draws on live web results, Knowledge Graph facts, and shopping data that the same document sized at more than 50 billion products. Gemini 3.5 Flash became the default model globally on 19 May 2026, with a Pro model selectable by some users.

The practical consequence for a publisher is narrow. Eligibility is inherited from ordinary Search.

Search Central states the requirement plainly. A page must be indexed and eligible to appear with a snippet. There are no additional technical requirements.

AI Mode vs AI Overviews vs Classic Search

Google AI Mode is the third surface a query can land on. The three behave differently enough to budget for separately.

The distinction that matters sits under the interface: which system decides what the reader sees, and where your link lands when it does.

Search-share and click figures are Pew's March 2025 browsing data. Position and opt-out behaviour are Google's own documentation, checked 20 August 2026.

What Triggers Each Surface

AI Overviews are conditional. Google says they appear only where its systems judge a generated answer additive, and where confidence in the quality is high. They often do not trigger at all. Pew measured them on 18% of Google searches in March 2025.

AI Mode is elective, then sticky. A user chooses the tab, and since 19 May 2026 can also flow into it by asking a follow-up straight from an AI Overview. Google's documentation notes that AI Mode will fall back to plain web links when confidence in the generated answer is low.

What Each One Shows

The two AI surfaces are not the same engine wearing different clothes. Google states that they may use different models and techniques. The same query will likely return a different set of responses and links on each.

Where the Links Sit

Position reporting differs, and that changes what a rank number means. An AI Overview occupies a single position, and every link inside it inherits that number.

AI Mode links are scored with the standard page methodology instead. A position from AI Mode therefore reads closer to a classic one.

How AI Mode Selects Its Sources

Google documents query fan-out first, followed by retrieval of supporting web content as the response is generated. That order explains most of what teams find confusing, including citations for queries a page was never built to rank on.

Query Fan-Out

Fan-out is the mechanism worth understanding. Google's help documentation states that AI Mode divides a question into subtopics. It searches each one at the same time, across multiple data sources, then brings the results back into one response.

One user question becomes many machine queries. A page can enter through a subtopic search it happens to answer well. That is closer to how AI surfaces choose what to show than to classic keyword ranking.

Source Selection

Grounding comes from ranked web results. Google's May 2025 explainer says AI Mode shows information supported by high-quality web content. The model is trained through agentic reinforcement learning to prefer claims it can back with retrieved inputs and Knowledge Graph facts.

Google also says its models decide when a hyperlink helps and when a visual answers better. That means retrieval and link selection both affect what the user ultimately sees, while the underlying decision process is not exposed publicly.

Why Results Vary by Session

Three documented factors move the output, and none of them is your page.

• Personal Intelligence: on 19 May 2026 Google expanded it in AI Mode to nearly 200 countries and 98 languages with no subscription, drawing on Search history and, with consent, Gmail and Google Photos.
• Model selection: users can pick a Pro model instead of the default, and Google states the two behave differently.
• Query framing: Google's own documentation contrasts "Why is an iguana a good pet?" with "Why isn't an iguana a good pet?" The two return different responses.

Treat any single screenshot as an anecdote. A prompt run once, signed in, on one device, is not evidence of anything you can report to a board.

What Is Documented and What Is Only Observed

What This Means for Publisher Traffic

The traffic story is easier to state for AI Overviews than for AI Mode. Public clickstream evidence shows fewer clicks when AI summaries appear, while comparable AI Mode clickstream data is not yet available.

Impressions Without Clicks

Pew's browsing study is the clearest public evidence, and it covers AI Overviews.

Users who met an AI summary clicked a search result in 8% of visits, against 15% where no summary appeared. Clicks on a link inside the summary itself ran at 1% of visits.

Session endings moved too. Pew recorded sessions ending on 26% of pages carrying an AI summary, against 16% of pages with only classic results. For a publisher, that gap is the whole argument.

What Publishers Have Measured

Google's position is that the clicks that do arrive are better. Search Central says clicks from result pages with AI Overviews are higher quality, and that users spend longer on the site. Google publishes no figure beside that claim.

Both things can be true, and neither is a plan. A publisher can keep engaged readers and still have to explain the shortfall. That is why AI referrals now behave as a distinct traffic source in most reporting stacks.

Which Queries Carry the Most Exposure

Pew's data gives a publisher one usable predictor. Query shape drives whether an AI summary appears at all, and the gradient is steep.

• Length: 8% of one-word and two-word searches produced a summary. That rose to 53% at ten words or more.
• Question words: 60% of searches opening with who, what, when, or why produced one.
• Full sentences: 36% of searches carrying both a noun and a verb produced one.

Map that onto your own query mix before you forecast anything. A reference archive built on long explanatory questions sits in the exposed band. A masthead living on short brand and navigational searches sits mostly outside it, at least for now.

We ran a small check of our own on this article's topic. In Ahrefs Keywords Explorer, United States, on 20 August 2026, we submitted the twelve queries below and read the stored SERP features for each. Ahrefs returns metrics for nine and holds a SERP snapshot for seven, captured between 8 and 20 August 2026.

• Seven with a stored snapshot, and all seven carry an AI Overview plus AI Overview sitelinks: ai mode, google ai mode, ai overviews, ai overviews tracking, generative engine optimization, answer engine optimization, how to track ai overviews.
• Five with no stored snapshot: what is google ai mode, how does google ai mode work, how does ai mode choose sources, how to appear in google ai mode, does ai mode reduce website traffic. Three of those five return no Ahrefs metrics at all.

Read the gap as carefully as the hit rate. Seven of seven is a small sample, and a missing snapshot is not evidence that no AI Overview appeared. What the split shows is that, in this small Ahrefs sample, every query with a stored SERP snapshot displayed an AI Overview. The long question tail is where third-party SERP data runs out.

How to Check If You Appear

Teams asking how to track AI Overviews and AI Mode in one place now have a first-party answer. It is narrower than most expect.

Google launched Search Generative AI performance reports in Search Console on 3 June 2026. The report shows impressions from generative AI features on Search, covering both AI Overviews and AI Mode. It is rolling out to a subset of properties rather than to everyone.

What the Report Gives You

• Impressions: how often links to your site were shown inside a generative AI feature.
• Pages: which URLs appeared, assigned to the canonical URL.
• Countries and devices: where the search originated, and on what.
• Dates: hourly, daily, weekly, and monthly granularity, in Pacific Time.

What It Does Not Give You

No clicks, no CTR, no position, and no query dimension. Google's help documentation lists impressions as the metric and says additional metrics may come later.

No split between the two surfaces. AI Overviews and AI Mode are reported together. A change in the line tells you the family moved, not which member. Search Labs data is excluded, and the usual 1,000-row limit applies.

How to Run the Manual Check

Manual checking still has a place. Treat the output as a sample, never as a measurement.

1. Fix the prompt list first. Twenty to forty buyer questions, written once, changed only by agreement. A list that drifts cannot show a trend.
2. Control the session. Signed out, history off, one country, one device class. Personalisation is the loudest confounder Google documents.
3. Record the conditions alongside the result. Date, model shown, country, and the exact wording. A screenshot without those four fields proves nothing later.
4. Repeat on a schedule. Monthly is enough for most teams. Weekly only if you are actively publishing into the gap.

How to Improve Your Odds of Inclusion

Google's published guidance for AI features is deliberately unremarkable, and pretending otherwise is how agencies sell work that cannot be checked. There is no AI Mode markup, no submission route, and no eligibility separate from Search.

The documented list, from Search Central's AI features guidance as at 10 December 2025, is this:

• Crawl access: allow crawling in robots.txt and at the CDN or host layer.
• Findability: make content reachable through internal links on your own site.
• Text availability: keep important content in textual form rather than locked in images.
• Structured data honesty: ensure markup matches the visible text on the page.
• Business data currency: keep Merchant Center and Business Profile records up to date.

Fan-out is where editorial judgement earns its keep. One question becomes many subtopic searches. Pages that answer a subtopic cleanly have more ways in, which is the practical case for structuring pages so a machine can lift the answer.

The rest of the work sits off your domain, where the corroborating sources live. That half is the same discipline covered in how models choose what to cite. It does not change because the surface is Google's.

What Does Not Move the Needle

Three things get sold as AI Mode levers with no documentation behind them.

• A dedicated schema type. None exists. Google asks only that markup match visible text.
• Rewriting every page into question-and-answer format. Google publishes no format preference for AI features, and fan-out already reads ordinary prose.
• A submission or feed route. There is no way to hand pages to AI Mode directly. Indexing is the queue.

The negative claims here are the safest ones in this article. They cost you nothing if a future Google update proves them wrong, and they save a quarter of misdirected work if it does not.

Tracking AI Mode Visibility

AI Overviews tracking and AI Mode tracking are one job inside Google's own reporting. Everywhere else they are two. Pick the route by the question you have to answer.

Set the baseline before you change anything. Impressions in the generative AI report are the only Google-supplied number tied to this surface. A month of it before a content push beats a year of it afterwards.

Do not let one report stand for the whole question. Impression movement without click data answers half of a board question. Teams pair it with prompt-level tracking across models on the engines Google does not report.

How Klarivo Tracks AI Mode Alongside the Other Engines

Klarivo Monitor does not query Google AI Mode, and no honest vendor claims to. I would rather write that plainly here than let a reader discover it in month two. Monitor tracks five answer engines, named on its own page: ChatGPT, Claude, Perplexity, Gemini, and Grok. Google Gemini is a model. Google AI Mode is a Search surface. The two are not interchangeable.

That boundary is the useful part. Google's generative AI report is now the source of truth for impressions on its own surface, and it stops there. Klarivo covers the half Google will never report on. That is which domains the engines drew on when they answered, tracked per provider, on a schedule the client controls. Read together, you get the Google surface from Google and the source stack from Klarivo.

The wrong fit is a team that only needs a Google number, because Search Console now supplies it free.

Book a Klarivo discovery call if the question is which sources the engines read before they wrote you out. Fifteen minutes, a slot you choose, and instant confirmation, with no form to fill in first. The tracking side runs in Klarivo Monitor.

Frequently Asked Questions

Can you opt out of AI Mode without losing classic search visibility?

Yes, and Google says so in writing. The Search generative AI control sits in Search Console settings. It lets an owner exclude a property from AI Overviews, AI Mode, and generative features in Discover. Google states the control is not used as a ranking or inclusion signal elsewhere in Search. Three caveats matter. It is rolling out to a subset of owners. The three features cannot be excluded separately, and exclusion means no impressions and no traffic from any of them. Changes generally take a few days to propagate. Google says content is typically excluded within 1–2 days after the control goes live, though some content may take longer because of caching and propagation.

Do AI Mode citations pass any measurable referral traffic?

Clicking an external link in AI Mode counts as a click in Search Console, so the traffic exists in the total. Isolating it is the problem. The generative AI report carries impressions only. The main performance report folds AI clicks into the Web search type, with no filter to pull them back out. The nearest public read on click behaviour is Pew's 1% click rate on links inside AI summaries, and that measures AI Overviews. For AI Mode specifically, nobody outside Google can size it.

Does AI Mode behave differently by country?

Availability differs, and so does the feature set. Google's own announcements track the expansion:

• May 2025: AI Mode opens to everyone in the United States, with no Labs sign-up.
• 21 August 2025: already live in India and the United Kingdom, it reaches over 180 more countries and territories in English.
• 7 October 2025: more than 35 new languages and over 40 more countries and territories.
• 19 May 2026: Personal Intelligence in AI Mode reaches nearly 200 countries and territories across 98 languages.

The Pro model runs in English only, on a shorter country list published on Google's AI Mode help page. Availability, model, and personalisation all vary by market, so a UK result and a Brazilian result are not comparable evidence.

How stable are AI Mode results between two identical searches?

Less stable than a ranking, and unstable by design. Google documents three sources of variation: personalisation, the model the user selected, and query framing. Two similar phrasings can return different responses. Follow-up questions carry context forward, so the second question in a session is not the first one repeated. Sample across sessions and dates. Never report a single run as a position.

Does structured data change your odds of being selected?

Not as an AI-specific lever. Google's guidance lists structured data under ordinary SEO fundamentals. It asks only that markup match the visible text on the page. No documentation claims that a schema type improves AI Mode inclusion. Keep doing it for rich results and machine legibility, which is a different argument from the one usually made for it.