Your shopping cart is empty!
AI Overviews got more links, but some of them open AI Mode instead of a site
Publication date: 26.09.2026
Quick take
- Google has been adding links to AI Overviews all year, but some of them open AI Mode rather than a website.
- Search Console counts a click only on an external page; query refinements count as neither clicks nor impressions.
- Peec AI charted the share of external links in answer text rising from near zero to 26.2% in three weeks.
No, not every link in an AI Overview leads to a website. Matt Southern at Search Engine Journal pulled together what Google has said about links over the past year and how Search Console counts them. The result is a picture where it's easy to draw the wrong conclusion about traffic.
A year of small tweaks to the answer box
In August 2025, Robby Stein, Google's VP of Product for Search, said the goal was "to show more inline links", and link carousels appeared on desktop. On January 27, 2026 he announced that AI Overviews let users jump into a conversation with AI Mode, on mobile worldwide. In February Google tested link groups in pop-ups.
On May 6, Hema Budaraju, VP of Product Management for Search, listed five updates: article suggestions at the end of a response, labeled subscription links, previews of public discussions, extra inline links beside relevant text and hover previews with site names. On June 3, Google said it had increased the number of inline links.
It sounds generous. The catch is that these links don't all behave the same way.
What Search Console actually sees
The rules in Google's documentation are simple. Clicking a link to an external page in an AI Overview counts as a click. A link that runs a new Google query counts as neither a click nor an impression. A question asked in AI Mode as a follow-up is treated as a new query, and all resulting data goes to that query.
The generative AI report shows impressions only, no clicks, and aggregates them at property level: three links to your site in one answer equal one impression.
| What the user did | What Search Console records |
|---|---|
| Clicked a link to an external page | A click |
| Clicked a link that runs a new query | No click, no impression |
| Asked a follow-up in AI Mode | A new query with its own data |
| Three links to your site in one answer | One property-level impression |
26.2%was the share of external links inside AI Overview answer text on September 20 in a Peec AI chart; until September 10 it sat near zero.
What the Peec AI chart shows and what it leaves out
On September 21, Peec AI published a chart: between August 25 and September 10 there were almost no external links inside the answer text, a sharp jump came on September 11, and by September 20 they made up 26.2%. It's an eye-catching number, but Peec AI doesn't specify the sample, markets or devices. I'd read it as a sign that something changed, not as a measure of what your customers see. Peec AI also sells AI visibility analytics, so it has a natural interest in changes like this.
The same day, Gagan Ghotra shared a screen recording in which underlined text inside an answer opened AI Mode instead of an external page. It's a single case and Google hasn't commented. It does touch the most interesting question, though: whether that jump will count as a query refinement, meaning no click and no impression. The documentation doesn't say.
Picture a simple case. In Search Console, your page's AI Overview impressions rise over a week, but clicks don't. Either people read the answer and leave, or part of the taps went into AI Mode and never reached your stats. You can't tell those two apart right now. In that situation I'd avoid conclusions from one week and compare the trend against what the guide to SGE and AI Overviews describes.
Ukrainian sites will notice this more, since AI Overviews appear unevenly across most queries here, and visibility in AI answers has to be measured from several sources at once.
I wonder whether Google will ever say where a click on an underlined word actually goes. For now even that is a guess.
