AI Overviews: The Clicks Search Console Doesn’t Count

A Tuesday Morning, Links Without Clicks

A client calls me on a Tuesday morning. He has a screenshot in front of him. His pages hold three links in an AI Overview. The text is underlined. The site logo shows in the preview. Everything looks perfect.

He opens Search Console. Clicks? None. Impressions? None either. For twenty minutes, he compares dates. He exports. He runs the test again in a private window. Nothing.

The content holds up. The measurement doesn’t.

According to Search Engine Journal, not every link in an AI Overview leads to a web page. Some open a conversation in AI Mode. Search Console doesn’t count them as clicks. Or as impressions.

Yet Google announced more links on display. On August 25, 2025, Robby Stein, Vice President of Product at Google Search, announced more inline links and carousels on desktop. On January 27, 2026, he described the follow-up question that opens AI Mode. On February 17, 2026, he talked about link pop-ups on desktop. On May 6, 2026, Hema Budaraju listed five changes: article suggestions at the end of the answer, press subscription links, excerpts from public discussions, additional links along the text, and hover previews.

Result: more links. Fewer attributed clicks.

I’ve been seeing this gap with my clients for several months. One e-commerce site gets 47 visible links in AI Overviews over a week. Search Console records only 9 clicks. The gap isn’t a bug. It’s a different accounting method.

A media site tracks 18 keywords. Twelve show AI Overviews with its domain. Seven pages are cited. Search Console reports 2 clicks per week. The rest: links that re-run queries.

The first reaction is always the same: add a tracking tool. Then a second. Then a click-capture script.

Before building, understand the mechanism. Otherwise, you’re measuring noise.

Un même lien dans un AI Overview peut finir à deux endroits — et Search Console ne les traite pas de la même façon. Ce schéma montre où part réellement chaque clic.

Two Destinations, Two Accounting Methods in Search Console

A link in an AI Overview has two possible fates.

• It leads to an external page. Search Console records a click.
• It re-runs a Google query. Search Console records neither a click nor an impression.

This rule changes everything. The volume of links displayed doesn’t predict your clicks. An underlined link in front of 100,000 users can produce zero measurable clicks if Google sends it into AI Mode.

What Search Console actually counts:

Google describes links next to the answer text. Article suggestions at the end of the answer. Previews of public discussions. Follow-up questions that open AI Mode. Since January, on mobile, a follow-up question turns the AI Overview into a conversation.

Search Engine Journal reports a chart published by Peec AI on September 21. It measures the percentage of AI Overviews containing an external link in the body of the answer, from August 25 to September 20. The data confirms the presence of links. It says nothing about clicks.

The same day, Gagan Ghotra publishes a recording. He clicks on underlined text in an AI Overview. He lands in an AI Mode conversation. Not on a site.

Two observations, two different natures. The first speaks of visibility. The second speaks of real traffic.

If you add the two in a single dashboard, you think you’re measuring a trend. You’re measuring a mix of incompatible natures.

Technical reminder: in Search Console, a click to an external page counts as a click. The link that re-runs a Google query doesn’t count. It doesn’t appear in impressions either. It’s a structural blind spot, not a temporary bug.

Why does Google do this? Because the user hasn’t left Google. A new query in AI Mode isn’t a visit to your site. It triggers no server-side data.

This distinction changes how you read a report. I’ve seen peaks in links displayed without the slightest peak in clicks. Separate the two flows.

What Google Describes: A Five-Faced Link Architecture

Since August 2025, Google has changed the display of links in its AI answers five times. Each version adds a surface. And each surface leads somewhere different.

1. Inline links in the answer text: external site destination.
2. Desktop link carousels: external site destination, often.
3. The follow-up question that opens AI Mode: conversation destination, no external click.
4. Desktop hover link pop-ups: external site destination, with preview.
5. Article suggestions at the end of the answer, press subscription sources, previews of public discussions: destination sometimes a site, sometimes a community, sometimes a subscription.

Robby Stein, cited by Search Engine Journal, talks about « a smooth expérience with visible links to keep exploring. » The key word is « keep. » Not « visit your site. »

Keep exploring can mean staying inside the Google ecosystem. For you, it produces a Q&A session that leaves no trace on your servers.

Hema Budaraju, Vice President of Product Management, detailed five updates on May 6, 2026. « Where to go next » suggestions appear at the end of many AI answers. They point to articles or in-depth analyses. Press subscription links carry a label. Public discussion and social network perspectives show the community name.

Google also mentions additional links along the relevant text. On desktop, hovering over an inline link shows a preview with the site name or the page title.

What Google doesn’t specify: the share of these links that truly leads to the open web. The May post gives no count. It doesn’t segment by market. It doesn’t date the end of the rollout.

For an SEO, this gray zone speaks. You’re driving without a precise counter. You’re deciding with a partial picture.

A client shows me a report: 12 appearances in AI Overviews on Tuesday. On Wednesday, 8 appearances. Zero clicks. He thinks his pages aren’t converting. In fact, half the links pointed to AI Mode.

Visibility in AI doesn’t read like classic visibility. It reads by destination.

The Uncertainty Bias: Why Missing Reliable Data Drives Over-Instrumentation

When a data point is missing, the brain tries to fill the void. An old mechanism. Guillaume Attias teaches it at BMO Academy, in the DOSE framework. He calls it the uncertainty bias.

The absence of reliable data creates cognitive dissonance. You see links. You don’t see clicks. Your dashboard shows a zero. Your eye sees a presence. The two don’t meet.

The classic reaction: add a tool, then a script, event tracking, a daily export. You think you’re measuring more. Most of the time, you’re measuring noise more often.

It’s measurement dopamine. Each new indicator offers a micro-reward. The number moves. You feel active. The decision doesn’t improve.

I look at 15 sites per week. All have the same problem. Three superimposed tracking tools. Four different views. Zero separation between an external click and a query re-run.

Over-instrumentation has a cost: it slows down the site and complicates reading. Above all, it gives a false sense of control.

With AI Overviews, the worst mistake is consolidating different click natures. A click to your page is a visit. A query re-run is a search behavior. Adding the two is like comparing apples and conversations.

The uncertainty bias disappears when you accept a simple rule: not everything can be measured. Part of AI interactions will remain invisible in Search Console. Better to accept it and build tracking that separates the measurable from what isn’t.

Guillaume Attias insists on one point: facing uncertainty, the answer lies in better categorization, not more sensors. The bias is handled by structure, not accumulation.

Concretely, I propose a two-tier measurement. The first reads what Search Console counts. The second tracks indirect signals: a rise in brand queries, a rise in direct visits, a rise in page views per session. These signals don’t replace the click. They indicate that a passage happened, even without an attributed click.

Indirect Signals When the Click Doesn’t Exist

When the click doesn’t exist, traces remain. They are indirect. They don’t replace direct measurement: they complete it.

First trace: the brand query. A user sees your site cited in an AI Overview. They don’t click. They type your name into Google later. The brand query rises. It’s a brand awareness signal.

Second trace: the direct visit. The person remembers the domain name. They type the URL. No referrer, no organic click. Your server sees a direct visit.

Third trace: time spent on a page right after an AI search. The logs show an arrival on a product page five minutes after an unclicked AI query. It remains a clue.

An e-commerce client supported for a year saw its brand queries rise by 34% after strong presence in AI Overviews. Direct clicks only moved by 6%. The link existed mentally. It wasn’t yet passing through the click.

These percentages are orders of magnitude observed with my clients. They don’t come from a public study. They indicate a trend: part of the AI audience goes through the brand before clicking.

The lesson: your dashboard needs two columns. Column A: real measured clicks. Column B: observed indirect signals. Keep the columns separate. Compare the trends.

If column A stagnates but column B rises, you have an asset. You manage it as brand awareness, not as a billable click.

If column A rises, that’s better. You’re reaching qualified traffic. The link leads to your site.

The editorial decision changes by column. Content that generates external clicks deserves click-through rate optimization. Content that generates brand searches deserves memorization optimization. These are two distinct crafts.

Building a Two-Tier Measurement: Columns That Don’t Mix

The first tier is Search Console. Filter pages. Isolate clicks that came from AI Overviews. From the Performance tab, use the « search type » dimension. Cross it with appearance. You see the real external clicks.

The second tier is your server. Look at the logs. Several views of the same page within seconds, without a click in Search Console, can signal a passage through AI Mode. It remains a clue.

Another signal: brand queries. If an AI Overview cites your brand without a click, the user often types your name next. The rise in brand query then shows the AI is making you known.

I’ve been building semantic cocoons since 2016. The structure holds. The way to validate has changed. Before, a click validated a page. Today, it’s its footprint: direct click, brand query, post-AI consultation.

A two-tier dashboard can be set up in 45 minutes. It saves you weeks of false conclusions.

To build it, start with a single question: does this signal prove a visit to my site? Answer no, column B. Answer yes, column A.

Columns are read side by side, never one against the other. A rise in column B without a rise in column A means your brand awareness is moving. A rise in column A without a rise in column B means your direct link is working.

This separation avoids stacking five tools. One read in Search Console, one in the logs, one spreadsheet. That’s enough.

Dans un rapport mensuel, voir 1 000 clics ne signifie pas 1 000 provenances identiques. Voici comment se répartissent vraiment vos clics mesurés.

The Single Dashboard Trap: Why Adding Heterogeneous Clicks Distorts Your Decisions

Take a monthly report. You have 1,000 organic clicks. Of which 120 AI Overview clicks to your pages. And 80 AI Overview links that re-run a Google query, generating not a single click. Your dashboard shows 1,000. In reality: 1,120 passages happened. 80 never touched your server.

If you add those 80 to the 1,000, you believe traffic increased. You invest in the wrong lever. If you ignore them, you miss an audience that starts with an AI question before arriving.

Both mistakes come from the same reflex: putting everything into a single gauge.

Search Engine Journal reports that Google doesn’t specify the share of links that returns to the web. The May 2026 post gives neither count nor segmentation. I treat this blur as a constraint. Up to me to structure differently.

Your classic tracking tools don’t capture query re-runs. Google Analytics will see nothing. A click-tracking script won’t either. Only cross-reading logs, brand queries, and Search Console appearances gives a usable picture.

A frequent trap: confusing the displayed link and the counted click. A displayed inline link is not an impression. A link that opens AI Mode is not a click. Both can appear in your clients’ screenshots. Neither appears in the report.

Next time you see an AI Overview with your page, ask yourself: does this link lead to my site? That simple question separates the two realities.

What matters is readability, not volume. Two columns. One line per signal nature. One decision per footprint type.

With my clients, we stop dashboards that mix everything. We start from scratch. I trace the exact source of the click. I note whether the link was external or conversational. And we watch the trend over four weeks.

The result: fewer tools, more clarity. And decisions based on what actually circulates, not on what shines on screen.

Did your last Search Console report separate these two click natures?

Stéphane Jambu

SEO & AI Engineer

I build growth systems / AI / Neuroscience | 650+ clients · 80 LinkedIn testimonials · 30 years of expertise · 15 years of systems running without me.

Follow on LinkedIn