What Google AI Overviews are, and what they do to your site traffic

What you get from this article

• AI Overviews passed 2.5 billion users per month, according to Google itself in 2026.
• Ahrefs measured a 34.5% drop in first-position clicks when the block is present.
• The drop bottomed out and partly recovered: Seer saw clicks go from 1.31% back to 2.36% between December 2025 and February 2026.
• Being cited inside the block is worth 120% more clicks per impression than being left out.
• There is no button to opt out of AI Overviews alone: the available tags also remove you from featured snippets.
• Top-of-funnel content is the hardest hit, and decision-stage content the least.

What it is, exactly

An AI Overview is the block of text Google assembles at the top of the results page, above the blue links, answering the question in its own words and citing a few pages as sources in side or inline links. It is not an ad, it is not a passage copied from a single site, and it is not chosen by bidding: it is generated on the spot.

The scale explains why this became compulsory reading. In July 2025 Google said the feature had more than 2 billion users a month, in more than 200 countries and more than 40 languages. In 2026 the company updated that to more than 2.5 billion monthly users. It is not a test, it is the default format for an enormous share of searches.

It is worth separating it from two neighbours people confuse it with. The featured snippet is older and works by cutting: Google takes a literal passage from a page and shows it in quotes, with the link. The AI Overview writes new text from several pages. And AI Mode is a separate tab, where the whole search becomes a conversation; the AI Overview appears inside ordinary search.

For a site owner, the practical distinction is one thing: in a featured snippet, one page wins. In an AI Overview, several pages are used as raw material and one of them may receive no click at all, even having been the source.

How Google assembles that answer

The question you typed is not the only search that happens. The system breaks your question into several sub-questions and runs them all at once, each one bringing back a set of results. Then it reads what came back and writes a single text that answers the original question.

That mechanic has a direct and counterintuitive consequence for anyone producing content: you do not need to rank first for the main question. It is enough to be the best answer to one of the sub-questions the system generated on its own. That is why very specific pages, which would never win the contest for the generic term, show up cited in blocks about broad subjects.

It is also why the format of the text matters so much. The system has to find, inside your page, a passage that answers the sub-question self-sufficiently. If your best explanation depends on three previous paragraphs of context, it is not usable.

And there is an effect the reports reveal before the revenue does: impressions go up and clicks do not follow. You appear more often, for more question variations, and receive proportionally fewer people. Anyone watching only average position thinks things are improving.

Impressions rising with clicks flat is not a sign things are improving. It is the portrait of somebody who became raw material for somebody else answer.

How much traffic this really costs

Here it is worth swapping opinion for measurement, because the subject attracts a lot of exaggeration in both directions.

Ahrefs compared 300,000 search terms between March 2024 and March 2025, half with the block present and half without. For the first position, the presence of an AI Overview cut the click-through rate by 34.5%. That is not the end of organic traffic, and it is not small either: it is a third of the audience of the best possible position.

Seer Interactive, tracking 53 brands, 5.47 million terms and 2.43 billion impressions, shows the full curve, and it is more interesting than the isolated number. The click-through rate on searches with the block fell from 3.19% in January 2025 to 1.31% in December of the same year. And then it rose again, to 2.36% in February 2026, an 80% recovery in two months.

The correct reading of that curve is not "the problem has passed". It is that the whole market was pushed down while the format stabilized, and then settled at a new level: better than the bottom, worse than before. Anyone who made a structural decision looking only at the December 2025 trough made a decision on the worst point in the series.

The number that matters more than the drop

The same Seer study contains the figure that changes the strategy: being cited inside an AI Overview delivers 120% more clicks per impression than not being cited. More than double.

That repositions the fight. If the block is going to appear anyway, and it is, fighting against it is arguing with the rain. The fight that pays off is being inside it. It stopped being a question of position in the list and became a question of being a source.

But the same study carries the honest counterweight, and it matters for not selling illusions: even when cited, the result still performs around 38% below what it would perform in a search where the block does not appear. In other words, the best possible situation inside the new format still yields fewer visits than the old situation.

The practical conclusion is hard and simple. For the same volume of business, the site needs more impressions, better conversion per visit, or sources of demand that do not depend on organic search. Companies that depended on a single channel are discovering this now, and that is usually where a paid media conversation starts that would not have made sense before.

When the block appears and when it does not

The distribution is not uniform, and understanding that avoids unnecessary panic for anyone not in the risk group.

The block appears most often on informational questions: what is, how does it work, what is the difference, how long does it last, why does it happen. That is exactly the kind of content many companies used as top of funnel, and it is where the loss is biggest.

It appears less, or not at all, on brand searches (somebody looking for you by name), clear transactional searches (buy, hire, price, quote), local searches with immediate intent, and sensitive topics such as health and personal finance, where Google is more conservative.

The strategic consequence is direct: top-of-funnel content lost return and decision-stage content held. If all your content investment was in explaining the basics of your sector, the arithmetic got worse. If it was in service pages, comparisons, use cases and pricing, little changed.

• Hardest hit: "what is", "how does it work", "what is the difference between", glossaries, introductory content.
• Barely hit: searches for your company name, "best X for Y", "how much does it cost", "near me".
• Unstable: health, money and law, where Google alternates between showing and hiding the block.

Can you opt out of AI Overviews?

That is the most frequent question and the answer is unwelcome. There is currently no control that removes your page from AI Overviews alone. What exists are the old snippet-control tags, and they apply to both places at once.

In practice: the nosnippet tag stops Google showing any passage of yours, which takes you out of AI Overviews and also out of featured snippets in ordinary search. max-snippet limits the size of the passage. And the data-nosnippet attribute lets you mark only part of a page as not cuttable, which is the finest instrument in the set. There is also Google-Extended, but that controls use of your content to train the Gemini models, not the assembly of the block in Search.

Google itself said publicly that it is studying updates to those controls to allow a specific opt-out from generative features in Search. Until that ships, the choice is between being in both formats or in neither.

Before using any of those tags, redo the arithmetic with the numbers from the previous section. Leaving the block does not give you back the old click: it gives you back a position in the list below a block that will still be there, assembled from your competitors content. In almost every scenario, that is worse than being inside.

What to do to be cited inside

Because the system breaks the question into sub-questions, the work is having, on your site, a self-sufficient answer to each sub-question that matters in your field. That is quite different from having one long, good page about the general topic.

In practice: each section of the text starts with a heading that is a real question, and the first paragraph below it answers that question on its own, without depending on what came before. Definitions in two sentences. Numbers with source and date written into the text itself. A table when there is a comparison, a list when there are steps.

Add to that what is infrastructure: structured data for the article and for frequently asked questions, authorship with name and role, publication and revision dates, and a page that does not depend on JavaScript to exist. It is the same set that serves for being cited by AI assistants, detailed in how to get your company cited by ChatGPT.

And review what already exists before writing anything new. Many companies have good material buried in badly formatted pages that answer everything but answer nothing in a cuttable way. Rewriting the structure of ten existing pages usually pays off faster than publishing ten new ones.

What changes in how you report results

An SEO report that only shows average position and traffic volume now hides more than it reveals. An average position that improves while traffic falls is exactly the pattern this format produces, and it looks like success on the spreadsheet.

Tracking that makes sense now has three layers. The first is the gap between impressions and clicks per term, in Search Console: terms with impressions rising and clicks falling indicate the block is present. The second is a periodic manual check of the terms that matter most, looking at whether the block exists and whether you are cited in it.

The third is the only one that pays the bill: how many qualified enquiries the organic channel generated. That number is not affected by a change in result format, and it is the only one that answers whether the investment is still worth it.

If you want to know on which of your terms the block already appears, who is being cited in your place and how much of your current content is cuttable, that survey is the first stage of an AI consulting engagement at ROO3. It answers, with data from your own site, whether the problem is content, structure or channel.

Frequently asked questions

Are AI Overviews and featured snippets the same thing?

No. A featured snippet cuts a literal passage from a single page and shows it with the link. An AI Overview writes new text from several pages and cites some as sources. Both can appear on the same search, and the snippet-control tags apply to both.

How do I stop my page appearing in AI Overviews?

Today there is no exclusive control for that. The nosnippet, max-snippet and data-nosnippet tags limit the passage shown, but they also apply to featured snippets in ordinary search. Google said it is studying a specific control for generative features, but it does not exist yet.

Do AI Overviews appear on every search?

No. They are far more frequent on informational questions, of the what-is and how-does-it-work kind, and much less so on brand searches, clear transactional searches and local searches with immediate intent. On health and finance topics Google is more conservative and alternates.

I lost traffic because of AI Overviews. Can I get it back?

Partly, by changing the target. The recovery comes from being cited inside the block, which according to Seer Interactive yields 120% more clicks per impression than being left out, and from shifting effort from introductory content to decision-stage content, which is less affected. The old level, however, does not return.

Does being cited in an AI Overview make up for the lost clicks?

Partly. In the Seer study, being cited yields more than double the clicks per impression compared with not being cited, but it still sits around 38% below what the same search would yield if the block did not exist. It is the best position inside the new format, not a return to the old one.

My content is used without me getting a visit. Is that fair?

The debate is legitimate and is being handled by regulators in several countries, including public consultations on the controls Google offers websites. From the practical point of view of a site owner today, though, the available decision is between taking part in the format or leaving search snippets entirely.

Sources

• Ahrefs: AI Overviews reduce clicks on the first position
• Seer Interactive: AIO Impact on Google CTR, 2026 update
• Google Search Central: AI features and your website
• Google: our approach to site controls in AI features in Search

Rodrigo Fávaro

Founder of ROO3, a marketing and technology agency in São José do Rio Preto, Brazil. Builds AI products running in production (Tobia, gerar.app, Pense Mercado) and maintains the AI Benchmark, a public ranking of AI models. See ROO3 AI consulting.

X @rodmf LinkedIn rodrigofavaro