Google AI Overviews and AI Mode: what they are and how your brand appears

AI Overviews are a summary written by AI that answers the search above the usual ten links, with the pages that back it up shown alongside. The reader gets the answer before seeing any organic result, so the question for a brand changes: being on the first page isn’t enough; what matters is whether the summary names you, how it describes you and which site it sends people to. In Latin America Google calls them «Visión general creada por IA», and in Spain «Vista creada con IA». In May 2026 Google reported more than 2.5 billion monthly active users.

They don’t appear on every search. According to Google’s documentation, they show only when Google’s systems decide they add something to classic search, so on many queries they simply don’t trigger. This matters when you measure: “there was no AI Overview” and “there was an AI Overview and it didn’t name you” are two different results.

AI Mode is Google’s most complete AI search experience: it answers with generated text, takes follow-up questions and links the sites it used. It is built for exploring and comparing. Since May 2026 it runs on Gemini 3.5 Flash by default, and in July 2026 Google reported that it has passed 1 billion monthly active users.

Google’s documentation says the two features can use different models and techniques, so they don’t give the same answers or cite the same sites. Since May 2026 Google presents them as one experience: you can move from an AI Overview into AI Mode with a follow-up question.

The practical conclusion: this is not a test you can wait out. In Mexico AI Overviews have been running for more than two years, and your customers are already reading them.

Neither feature searches just once. Both can use a technique Google calls query fan-out: they split your question into subtopics, run several related searches in parallel across different data sources, and then write an answer from what they found. Google describes it in its guide to generative AI features.

An example: asked “what’s the best health insurance plan for a family in Buenos Aires?”, AI Mode may search separately for coverage, provider networks by neighborhood, this year’s prices and reviews. Each sub-question is a separate chance to be cited. That is why a brand can be missing from the answer even when it ranks well for the full phrase, and can appear thanks to a page that answers one specific part well.

One consequence worth keeping in mind: the answer is built from what the web ranks for your language and your country. If your site doesn’t say where you sell, in what currency and with what coverage, the search engine has nowhere to get it from.

Another: the engines don’t agree with each other, so one engine’s answer tells you little about another’s. In Refractia’s Índice de Marcas en la IA (August 2026; 62 categories and 4,200 answers from ChatGPT, Gemini and Claude with web search off), in 46 of 62 categories at least one brand is named by one model in at least 60 points more of its answers than by another. Banco Macro, for example, appears in 44 of 48 ChatGPT answers about banks, 22 of 48 from Gemini and 7 of 48 from Claude. That index measured chat models, not AI Overviews, but it is the reason to measure each Google surface on its own as well.

The official advice is surprisingly simple. According to Google’s guide, updated in July 2026:

• There are no additional requirements: the page has to be indexed and eligible to be shown with a snippet.
• Original, useful content with first-hand experience, not a repeat of what everyone already says.
• Clear structure with headings, and good-quality images and video.
• A crawlable site, with internal links, a good page experience and the important content in text.
• Structured data that matches what is visible on the page.
• For retail and local businesses: Merchant Center feeds and an up-to-date Business Profile.

And what Google says you don’t need:

• llms.txt files, “AI text files” or special markup: Google doesn’t use them for this.
• Rewriting or “chunking” your content for AI.
• A page for every variant of a question: Google warns that this can violate its policy against scaled content abuse.

Transparency: Refractia publishes an llms.txt because other engines read it, not because Google uses it.

On top of that official foundation, this is what moves the needle most in our experience measuring brands:

1. 01

   Confirm your pages are indexed

   A page that Search Console reports as “Crawled, currently not indexed” doesn’t exist for AI Overviews or AI Mode. Check product, plan, coverage and FAQ pages, and every language version: it is common for one to drop out because of a canonical tag pointing to another page.

2. 02

   Answer the sub-questions of your category in your market

   Prices in local currency, coverage by country or city, honest comparisons with alternatives and concrete conditions, in text on a crawlable page. Those are the pieces query fan-out goes looking for.

3. 03

   Name your country and your cities

   Answers change from one market to another. If your content doesn’t say where you operate, you are competing for a generic answer that isn’t the one your customers see.

4. 04

   Earn real mentions in the sources AI cites

   Media, comparison pages and third-party reviews weigh on the answers. In Refractia’s data, a brand’s own website is between 0.5% and 2.8% of the sources AI engines cite; the rest are other sites. Google asks you not to chase artificial mentions: this is earned coverage, not bought.

5. 05

   Keep your structured data consistent

   Organization, products and prices marked up the same way they appear on the visible page.

6. 06

   Check your snippet directives

   nosnippet, data-nosnippet and max-snippet limit what Google shows from your page, in AI features too; noindex removes it from Search. If one was left in a template by mistake, every page on that template is out.

The wider discipline behind this, for every AI engine and not only Google, is in the GEO guide, and the item-by-item review of your site is in the GEO checklist.

Search Console. Clicks and impressions from AI Mode and AI Overviews are added to the Performance report, Web search type, and can’t be separated there. Since 31 August 2026 there is also a generative AI performance report: it shows impressions by page, country, device and date, but no clicks or queries, and it doesn’t tell AI Mode apart from AI Overviews.

What Search Console doesn’t tell you. What the answer says about your brand, whom it recommends instead of you, which sources it cites, and whether the answer changes in another country. For that you have to ask the engines what your customers ask, in their language and from their country, and read the answers.

Doing it by hand has two traps. The first is location: the answer you see from London may look nothing like the one someone gets in Mexico City or Madrid. The second is the sample: a single search says nothing, and on many queries Google doesn’t generate a summary at all.

This is how Refractia handles it:

• We query Google’s AI Overviews with each country’s location, and treat Google as a search engine, separate from the chat models.
• Questions are written in the market’s language and name the country in full.
• Every rate comes with its sample, “k of n”. If Google didn’t generate an AI Overview for a question, that question doesn’t count as “didn’t name you”: not measured is not zero.
• We report AI search (what the web ranks and the engine paraphrases) separately from chat (what the model knows): ChatGPT, Claude, Gemini and Muse, the model behind Meta AI. Mixing them hides the real picture, especially for growing brands.
• We record which domains each answer cites, separating your site from media and third parties, in Influence Sources.

• nosnippet, data-nosnippet, max-snippet: limit what Google shows from your pages, in AI features too. noindex removes the page from Search entirely.
• The generative AI control in Search Console (available to every site since 31 August 2026): excludes your site from AI Overviews, AI Mode and Discover’s AI features, and you stop getting their traffic. According to Google it doesn’t affect your ranking in the rest of Search. Google help.
• Google-Extended (in robots.txt) controls whether your content is used to train Gemini and for answers in the Gemini apps. It does not remove you from AI Overviews or AI Mode, and doesn’t affect your inclusion in Search. Documentation.