Gemini SEO: winning Google's three AI surfaces

Almost every article on this subject collapses these into a single thing called “Google AI”. They are not one thing. Same underlying Gemini models, three separate products, three separate sourcing architectures.

Notice the far-right column. The Yext study of 6.8 million citations found 52.15% of Gemini app citations came from brand-owned websites, the highest share of any major engine.

That is genuinely unusual. On most platforms your own site is a minor part of the evidence. On the Gemini app it is the majority of it.

If your content is actually good, this is where that shows up fastest.

AI Overviews is the most conservative of the three. Google runs the search, ranks the results as it always has, then synthesizes a summary from what it already retrieved. The organic index is the candidate pool. What changed is the selection step: Google picks based on how well a specific passage answers the implied question, not on overall page authority.

AI Mode is architecturally distinct. It decomposes your question into multiple parallel sub-searches, up to sixteen of them, retrieves and evaluates sources for each, then reasons across all of it to compose one answer. Deeper, more conversational, more citations, and a noticeably wider source pool that includes Reddit threads, niche forums and smaller publications that AI Overviews rarely touches.

The Gemini app behaves like ChatGPT or Claude. It answers from training knowledge by default and reaches for search grounding when the question demands it. Being present here is less about any single page and more about your standing across the web as an entity.

Same buyer, potentially all three surfaces in one afternoon. They see the Overview while searching, switch to AI Mode to research properly, then open Gemini to compare the shortlist.

Three separate chances to be present or absent.

This is the single most important mechanic on this page.

Your prospect types one question. AI Mode does not search for it. It invents a set of related sub-questions and searches for all of them in parallel.

Ask “which CRM is best for a small real estate team” and the system may separately retrieve for real-estate-specific CRMs, small-team pricing, ease of setup, integrations, migration effort and support quality. Then it composes one answer from the best passage it found for each thread.

Three consequences follow, and they should change your content plan.

• You are not competing on the prompt. You are competing on sub-queries you never see and cannot fully predict.
• Depth beats precision. A page that answers the headline question and four adjacent ones has five chances to be pulled in. A page that answers only the headline question has one.
• Passages get retrieved, not pages. Fan-out lifts sections, not URLs. A brilliant page with no extractable sections loses to a competent page with clean ones.

The practical structure this implies is not new, it is just newly essential. One comprehensive resource per topic, segmented so that every sub-question has its own heading and its own self-contained answer. Each H2 phrased as a question a buyer would genuinely ask, with the direct answer in the first 40 words beneath it, evidence after.

That format does double duty: the page ranks for more long-tail queries and hands Gemini a clean passage to lift for each one.

If you take one number away from this page, take this one.

In mid-2025, top-10 organic rankers accounted for roughly 76% of AI Overview citations. By early 2026 that share had fallen to around 38%.

Rankings still help. Google rank acts as a strong upstream filter, and pages ranking well are far more likely to enter the candidate pool. But among retrieved candidates, ranking position is no longer what decides who gets quoted. Passage quality is.

Across the wider body of research, the overlap between AI Overview citations and the organic top 10 now ranges anywhere from about 17% to 76% depending on the study and the query type. That spread is itself informative: it means the relationship is inconsistent and category-dependent, not a rule you can plan around.

Here is the uncomfortable operational implication. Ranking third on a query that triggers an AI Overview is worth materially less than ranking seventh and being the cited source inside it.

Most SEO reporting still treats position three as the win. On AI-heavy queries it is not.

You would reasonably expect two Google products, built on the same model family and feeding off the same index, to cite the same sources.

They do not.

Ahrefs analyzed 540,000 query pairs and found AI Mode and AI Overviews reached semantically similar conclusions 86% of the time while citing the same URLs just 13.7% of the time. A separate June 2026 analysis put the overlap at 14%. Two independent studies, same conclusion.

Same model, same index, same answer in substance, almost entirely different sources credited.

The reason is architectural rather than mysterious. AI Overviews draws from a curated, authority-weighted candidate set. AI Mode’s fan-out pulls from a much broader pool of sub-query candidates, which is why Reddit threads and niche publications show up there and rarely in Overviews.

So when someone tells you they have “optimized for Google AI”, ask which surface. If the answer is that there is only one, they have not looked at the data.

We should deal with the thing everyone is actually worried about.

Every serious study has reached the same conclusion. The magnitude varies, the direction does not.

• Seer Interactive, across 3,119 informational queries and 25.1 million organic impressions, measured organic CTR falling 61% when an AI Overview appeared, from 1.76% to 0.61%. Paid CTR fell 68%.
• Ahrefs analyzed 300,000 keywords and found a 58% CTR reduction for the top-ranking result on queries containing an AI Overview.

And do not count on your category being spared. Around 88% of healthcare queries now trigger an AI Overview, which puts paid to the assumption that sensitive or YMYL topics stay clear of them.

Two honest readings of this.

The pessimistic one: a meaningful share of informational traffic is not coming back, regardless of what anyone does.

The useful one: if fewer people click, then being named inside the answer is worth proportionally more than it used to be. The brand that gets cited shapes the decision even when nobody clicks through. The brand that ranks fourth and gets no citation increasingly gets nothing at all.

This is why measuring Google AI visibility only through sessions in GA4 will systematically understate what is happening.

The influence is real and the click is optional.

A point of persistent confusion worth settling.

Google-Extended controls whether your content is used to train Gemini models and improve grounded responses. It does not control whether you appear in AI Overviews.

There is no opt-out from AI Overviews that keeps you in Google Search. Appearing in the organic index and appearing in AI Overviews are the same decision. You can technically suppress snippet usage with nosnippet, max-snippet or data-nosnippet, but doing so damages your normal search presence as well.

For practical purposes: if you want to be in Google Search, you are in AI Overviews. Plan accordingly rather than looking for a door that does not exist.

One thing you do control that most sites get wrong: Googlebot access is the AI access. Gemini’s surfaces draw on Google’s standard index, so there is no separate AI crawler to allow. Crawl problems that hurt your rankings hurt your AI visibility identically. Fix them once, both channels benefit. That efficiency is the genuine advantage of Google-side AI work over optimizing separately for ChatGPT or Perplexity.

Google publishes more about this than any other AI vendor, and it still publishes no ranking factors. What follows comes from Google’s own guidance plus the citation research.

Track citation share on your top 50 priority queries weekly. Full methodology in how we measure AI answers.

The Google surfaces are where AI optimization and traditional SEO overlap most heavily. That makes them the efficient starting point for most businesses: the work compounds across three AI surfaces and your classic rankings at once.

It also means Google alone is not a strategy. Citation studies consistently find 80 to 91% of cited URLs appear on only one engine. Dominating AI Overviews tells you very little about whether ChatGPT names you, which is why we run all six through one AI optimization program.