What this is about. Part of your audience now reads the answer without opening the page: as of August 2026 the US zero-click share sits at 68.01%, up from 58.5% in April 2025. That breaks some of the measurement everyone relies on and leaves the rest intact. This page separates the two, and sets out a measurement approach that survives the shift.
On this page
- What does Search Console actually count?
- Where does AI traffic actually show up?
- The zero-click problem, stated honestly
- A setup that survives the shift
- Tests worth running
Key takeaways
- Google folds AI Overviews performance into the standard Search Console numbers. It is not broken out, so you cannot isolate it there.
- Search Console keeps 16 months of data. If you want a longer baseline, you have to export it yourself, starting now.
- Traffic from ChatGPT and Perplexity does arrive with its own referrer, so that part is segmentable. Traffic from an AI Overview does not.
- Average position is an average of ranks over impressions. It moves when your query mix moves, with no ranking change at all.
- Citation without a click still has value, and there is no honest way to price it yet. One page tracked a 1,900% month-over-month rise in ChatGPT citations with almost no measurable business effect. Say that plainly rather than inventing a number.
What does Search Console actually count?
Search Console is Google’s own report on how your pages performed in Search, kept for 16 months. Four numbers do most of the work, and three of them are routinely misread.
| Metric | What it actually means | The usual misreading |
|---|---|---|
| Impressions | Your result was present for a query, whether or not it was scrolled into view | Treating it as “seen by a human” |
| Clicks | A click through to your site | Treating it as sessions; the analytics number will differ |
| CTR | Clicks divided by impressions | Comparing it across query types, where intent differs completely |
| Average position | Mean of your best rank per query, weighted by impressions | Reading a change as a ranking change when the query mix moved |
The one to internalise is average position. If you gain impressions on a large set of queries where you rank fourteenth, your average position gets worse while your visibility improves. Reporting that as a decline is a self-inflicted wound. The same trap sits under click-through rate: organic CTR on queries carrying an AI Overview fell to 1.3% in December 2025 and recovered to 2.4% by February 2026, an 85% rise that still leaves the number suppressed, and two months is not a trend.
On AI Overviews, Google’s stated position is that clicks and impressions from them are included in the standard Search Console totals rather than separated out. That is worth repeating to anyone asking for an “AI Overviews report”: the data is in there, mixed in, and no filter pulls it apart.
Where does AI traffic actually show up?
The assistants that live on their own domain send their own referrer, which makes them straightforward to segment. Google’s AI features live inside Google, so their clicks arrive looking like search clicks.
| Source | Segmentable by referrer? | Where to look |
|---|---|---|
| ChatGPT | Yes | Referrals from its own domain |
| Perplexity | Yes | Referrals from its own domain |
| Copilot | Yes | Referrals from the Microsoft domain |
| Google AI Overviews | No | Mixed into organic Google in both Search Console and analytics |
| AI crawlers fetching you | Not traffic at all | Server logs, by user-agent |
The last row is the one most sites never look at, and it is free. Your access log already records which AI crawlers fetched which pages and when. A weekly count by user-agent tells you whether the machines that build answers are actually reading you — a leading indicator that arrives long before any citation does.
The zero-click problem, stated honestly
An answer that satisfies the reader without a click is a real outcome with no clean measurement. Two things follow, and they pull in opposite directions.
First, the pessimism is often overstated by the way it is measured. A query that never converted, answered without a click, costs almost nothing. Aggregate CTR decline says little on its own; the question is which queries lost clicks, and whether those queries ever led anywhere.
Second, the optimism is unpriceable. Being named as a source has brand value. Nobody can currently convert that into a number without inventing an assumption, and a report built on an invented assumption is worse than a report that says “we cannot measure this yet”.
Segment by query intent before concluding anything about zero-click. An aggregate hides the only distinction that matters.
A setup that survives the shift
- Export Search Console weekly, to your own storage. Sixteen months goes fast, and the interesting comparison will be to a period you no longer have.
- Split queries into intent groups — informational, navigational, commercial — and track CTR inside each. This is where an AI Overview effect becomes visible, and it is invisible in the aggregate.
- Parse access logs for AI crawler user-agents and chart them weekly, by section.
- Segment assistant referrals from ChatGPT, Perplexity and Copilot as their own acquisition channel.
- Check citations by hand, on a schedule. Ask the same set of questions in the same engines every week and record who is named. It is manual and it is the only direct evidence available.
Five items, none requiring a purchase. Note that the last one is manual on purpose: the tools claiming to automate it are sampling the same stochastic output you would sample, and charging for the impression of rigour.
Tests worth running
Two designs give usable answers on a normal site. Take a set of comparable pages, change one technical thing on half of them, and compare the change in impressions and clicks against the unchanged half over the same weeks — this controls for seasonality, which single-page before-and-after does not. Or record the state of a metric on the day of a known algorithm update and compare a fortnight either side, which at least dates the effect.
Whatever the design, write down the method next to the number. A number without its method is not evidence, and this site’s rule is that it does not get published.
FAQ
Can I see AI Overviews traffic separately in Search Console?
No. Google includes it in the standard Search performance totals rather than reporting it separately, so no filter isolates it.
Why do Search Console clicks not match my analytics sessions?
They count different things. One counts clicks on a result, the other counts sessions that started and were recorded. Consent tooling, redirects and blocked scripts all widen the gap.
Is a falling average position bad?
Not on its own. Gaining impressions on lower-ranked queries lowers the average while visibility rises. Check clicks and the query mix before drawing a conclusion.
