More links in AI Mode and AI Overviews: what Google's update means for your clicks

On 15 October 2026, Google unveiled a set of changes to AI Mode and to AI Overviews, the generated summaries that sit at the top of its search results. The announcement, written by Hema Budaraju, Vice President of Product Management for Search, promises more links to websites, better placed and better explained. For publishers and brands, the question is whether those links will turn into visits, and Google has published nothing that would answer it.

The timing matters. Every study published since 2025 points to fewer clicks when Google shows a generated summary, and the AI report in Search Console still shows impressions only. In the same week, a glitch in ChatGPT was a reminder of how fragile the measurement of AI-driven traffic remains.

This article sets out what Google actually announced, weighs it against the click data that exists, and then describes a method for measuring the effect on your own pages.

Key takeaways

• Google is adding more links to its generated answers but has published no figures on the clicks they produce.
• Available studies show a much lower click-through rate when a generated summary appears, and very few clicks on the summary's own links.
• Being cited in the summary appears to soften the loss without offsetting it, and no causal link has been established.
• Search Console measures your impressions in AI features, not your clicks, so measuring the effect is up to you.
• A page-by-page comparison of the weeks before and after 15 October 2026 gives a first reading, to be confirmed over several weeks.

What Google announced on 15 October 2026

The announcement bundles five changes, three of which bear directly on where links sit in the answer. Two are still in testing, while the other three are rolling out.

The first change concerns inline links, meaning the words or phrases in an answer that point to a page. Google reports "continued and substantial increases" in how often they appear and says it is placing them higher in the answer more often. The second change, still a test, applies to tasks where only one or two official sites exist. For those, Google shows a section at the top of the answer that surfaces the links and briefly explains why each one is useful.

The other three changes are about presentation. Article carousels for timely topics are expanding to more countries and languages, and they highlight the user's preferred sources. Links at the end of an answer are getting thumbnails and a more clickable design. Finally, creator and publisher names under AI Mode video results are becoming clickable, leading to their profile in Google Search.

The only figure in the announcement relates to Preferred Sources, a feature that lets people choose the sites they want to see more often in news results. According to Google, usage has grown more than sevenfold since the start of 2026. In a post quoted by Search Engine Roundtable on 15 October 2026, Nick Fox, one of Google Search's senior executives, also said that more than 400 publications had joined its pilot programme for licensing news content for AI features.

Do people click the links in Google's summaries?

The independent measurements published so far suggest those links are seldom used. They all predate the October changes, however, which set out to change precisely that.

The most widely quoted study comes from the Pew Research Center. Published on 22 July 2025, it draws on the real browsing of 900 US adults, recorded with their consent in March 2025. When a generated summary appeared, participants clicked a traditional result in 8% of visits. Without a summary, that share reached 15%.

The most telling figure concerns the links inside the summary itself. According to the same study, participants clicked one of those links in just 1% of visits to pages that displayed one. The researchers also found that browsing sessions ended more often after a page with a summary, in 26% of cases. For pages without a summary, the share was 16%.

Two limitations are worth keeping in mind. The study is observational, so it establishes an association between summaries and fewer clicks without proving that one causes the other. It also covers US users and a version of the summaries roughly a year and a half older than the October 2026 announcements.

Studies from 2026 find fewer clicks, with a smaller drop for cited pages

Two studies published in 2026, both built on Search Console data, back up the trend Pew observed. They add a useful nuance: a page cited in the summary loses less than a page left out of it.

Ahrefs' study published on 4 February 2026, led by Ryan Law across 300,000 keywords, compares December 2023 with December 2025. It associates the presence of an AI Overview with a 58% lower average click-through rate for the page ranked first. The figure is an estimate, since it compares the observed rate with a projection of what it would have been without a summary. The analysis also covers desktop traffic only.

Seer Interactive's study published on 24 April 2026 adds a complementary angle. It tracks 53 brands between January 2025 and February 2026. Taken together, that represents more than 5.4 million distinct queries. For informational queries in 2025, the table below shows the average click-through rates it recorded.

Being cited does not bring a page back to the level of a search with no summary, but the gap with an ignored page remains wide. The authors themselves write that they cannot claim causation, because high-authority brands are also the ones Google is most inclined to cite. They also flag that a single account makes up almost half the impressions in the "not cited" segment, which weakens that last figure. We have explained elsewhere why being cited by an AI does not mean being recommended, a distinction that applies to clicks as well.

Why won't Search Console tell you whether these links bring visits?

The generative AI report in Search Console counts impressions and nothing else. It contains no click data, and Google has set no date for adding any.

Google introduced these reports on 3 June 2026 on the Search Central blog, then opened them to all sites on 31 August 2026. They show how often your pages appear in AI Mode, in AI Overviews and in generative features in Discover. The available filters cover page, country, device and date. The authors of the post say only that Google is working with site owners to understand which data would help them, "such as adding additional metrics over time".

Google's position did not move with the October announcement. Asked by Search Engine Roundtable, a company representative said Google was continuing to work with website owners to understand which insights would be most helpful. A site can therefore watch its AI impressions rise without knowing whether those appearances come with visits or replace them.

Measurement on the site side has weak spots of its own, as an episode in the same week showed. According to Search Engine Roundtable on 14 October 2026, ChatGPT stopped adding its tracking parameter to outbound links for several days and stopped passing on its referrer. Cedric Huesler of Adobe saw, across roughly 500 pages, a fall in visits attributed to ChatGPT that was largely matched by a rise in traffic classed as direct. He reported on 15 October that the cause was a bug and that the parameter was back. Our article on AI traffic hidden in Google Analytics covers how to recover that traffic.

Measuring the effect of these links on your own pages

Without click data from Google, the most practical method is to compare AI impressions and organic visits, page by page, before and after 15 October 2026. All it takes is two exports and a spreadsheet, with no extra tooling.

Start by choosing a four-week baseline, from 17 September to 14 October 2026, and the matching period after the announcement. In Search Console, export page-level impressions from the generative AI report for both periods. In Google Analytics, export sessions whose source is Google organic search, for the same landing pages and the same dates. Combining the two files lets you sort your pages into four groups.

This reading is only indicative, because Google Analytics cannot tell a click from an AI Overview apart from a click on a traditional result. It becomes more reliable if you also track whether your pages actually appear in answers, using a fixed set of questions asked every week. That is what our citation tracking does, recording which pages each engine picks for a given question. Our article on how to prove your GEO is working explains the baseline principle in more detail.

What we recommend for the coming weeks

The section reserved for official sites deserves close attention, even though it is only a test for now. It applies to task queries, such as a form to fill in, a customer account or an administrative procedure, where Google looks for "the right page" rather than a synthesis. Check that each of those pages has an explicit title, a description that accurately states what it lets people do, and content readable in the initial HTML. A login page with no text, or a procedure page buried under several menus, stands little chance of being chosen.

For content pages, the aim is still to be cited in the answer, since Seer's data associates citation with a smaller loss of clicks. The passages picked up most often tend to be those that answer a question directly, with a sourced figure or a clear definition, in a form the AI can lift without rewriting. An AI visibility audit shows which pages already meet those conditions and which need reworking.

News publishers have a lever of their own in Preferred Sources, now that Google features them in news carousels. Encouraging regular readers to add you as a preferred source, where the feature is available, becomes audience work comparable to growing a newsletter list. Brands that publish video, for their part, should check what their profile in Google Search shows, since the new links point there.

Finally, we advise against drawing conclusions from the first few days. Google is rolling these features out gradually and testing some of them, so the effects may vary by country and by query for several weeks. A first comparison at four weeks, then a second at eight, will give you a sounder reading than any general study, because it will be based on your own pages.