Google Search Console AI Overviews Traffic Measurement and Clicks: SEO Official

Google Search Console AI Overviews traffic measurement, in the official June 2026 announcement, is a dedicated view of impressions in generative AI features on Search, AI Overviews and AI Mode, and Discover. Google lists impressions, pages, countries, devices for Search, and dates from hourly through monthly. Clicks, CTR, and queries are not in that list. If you searched "clicks" because that is how SEO has always worked, the official report does not give you that column yet, and the absence of the column is the thing to plan around, not the thing to work around.

The distinction between plan around and work around is the distinction that keeps the report honest. Planning around the missing column means reporting what Google does give and labeling what it does not. Working around the missing column means inventing a number to fill the gap. The first produces a report that survives a question. The second produces a report that produces a question, and the question is the one that ends the credibility of the report. The absence is real, and the honest response to a real absence is to say it is absent, not to fill it with a guess.

This post stays with that announcement and the AI features page. It is not a vendor workaround for a column Google did not ship, and a vendor workaround for a column Google did not ship is the kind of post that ages badly the moment Google ships the column.

The aging-badly point is the point that keeps this post useful over time. A workaround post tells you how to estimate the missing column. When Google ships the column, the workaround becomes wrong, because the estimate disagrees with the real number, and a reader who followed the workaround now has a number that does not match the source. A post that stays with the announcement tells you what the announcement says. When Google ships the column, the post updates to include it, and the post stays correct because it never claimed more than the announcement claimed.

The AI features page still says Overview traffic is also inside the overall Performance report, Web search type, and that clicks from pages with Overviews have been higher quality, more time on site. That quality claim is not a GSC click table, and a quality claim is not a click count.

The quality-claim-versus-click-count distinction is the distinction that prevents overreading. A quality claim says the clicks that do come from Overviews tend to be better. It does not say how many clicks came. The first is a statement about the nature of the clicks. The second is a count of the clicks. A team that reads the quality claim as a click count reports a nature statement as a volume number, and a nature statement is not a volume number.

You can believe Google that those clicks have been higher quality. You still cannot open the generative report and export a click column Google did not list at launch, and "cannot export" is the operative phrase.

How to read the official report

Use it as "how often did our URLs appear in Google's generative features." Slice by page and country. Do not treat an impression as a session. Do not treat it as ChatGPT, because ChatGPT is a different product with a different logging model, and treating a Google impression as a ChatGPT mention is how a report gets two systems confused.

The two-systems-confused error is the error that produces the most wrong reports. A Google Overview impression is a Google event. A ChatGPT mention is a ChatGPT event. The two are logged by two different systems with two different definitions of what counts. A report that puts a Google impression in a column labeled "AI mentions" is putting a Google event in a column that suggests it is a ChatGPT event, and the suggestion is wrong. The fix is the label. Label the Google impression as a Google impression, and the column stops lying.

Impressions, pages, countries, devices for Search, and dates from hourly through monthly. That is the official list. Devices apply to Search. Discover is in the generative view. Queries are not. CTR is not, and the absence of CTR is the absence of the click column divided by the impression column.

The CTR absence is the absence that follows from the click absence. CTR is clicks divided by impressions. Without clicks, there is no CTR, because there is no numerator. A team that reports a CTR for Overviews is reporting a ratio with a numerator the source does not provide, and a ratio without a numerator is not a ratio. The absence of CTR is not a separate gap. It is the same gap as the click gap, expressed as a second missing column.

Preview and crawl controls remain as documented: Googlebot via robots.txt, snippet controls, Google-Extended for other Google systems. Those controls are still the crawl story. They do not add clicks to the generative report, and a crawl control is not a measurement column.

Export top pages by impression. That is a useful homework list for "which URLs Google already shows in generative features." It is not a mention log on a prompt you typed, and a mention log on a prompt you typed is a different instrument.

Clicks you can actually measure

GA4 sessions with a usable referrer. Many Overview clicks look like Google organic, and "look like Google organic" is the attribution problem at the center of this measurement. Promptwatch visitor analytics, with a script or GTM, tag AI-referred hits when they arrive. Prompt-level Overview checks on Promptwatch paid plans store whether a prompt cited you. Essential $95 per month. Explore is ChatGPT-only. Site: promptwatch.com. Method: how we rank.

The attribution problem is the problem that makes Overview clicks hard to count. A click from an Overview often arrives in GA4 with the same referrer as a click from a regular blue link, because both come from a Google page. GA4 cannot tell the two apart without help, and the help is a script or a GTM template that tags the AI-referred hits. Without the help, the Overview clicks are hidden inside the Google organic number. With the help, they are a separate line, and a separate line is what lets you see them.

Keep GA4 for sessions. Label Overview organic honestly. If the referrer looks like Google and the user came from a page that also sits in Overviews, you still may not be able to prove the Overview box, and "may not be able to prove" is the honest label. Do not invent a click number from a GSC impression, because an impression divided by a guess is not a click.

Promptwatch paid plans add the prompt and competing URLs, plus non-Google engines. You can skip Promptwatch if you do not care about the prompt, competing URLs, or non-Google engines. That is a small program, and a small program is a valid program, just not the one this directory ranks first.

FAQ

Does GSC show AI Overview clicks in 2026?

Not in the metrics Google listed at launch. Impressions, yes. Clicks, CTR, and queries are not in that list, and the list is the thing to quote when someone asks for the click column. Quoting the list is the honest answer to the click question, because the list is the source, and the source is what a QBR can defend.

Can I skip Promptwatch if I have GSC?

Only if you do not care about the prompt, competing URLs, or non-Google engines. GSC generative reports are Google impressions. They are not ChatGPT, and "not ChatGPT" is the line that decides whether you need a second instrument. A team that only cares about Google can skip the second instrument. A team that cares about ChatGPT cannot, because GSC does not watch ChatGPT, and a Google instrument cannot answer a ChatGPT question.

What to do this week

Open the generative AI performance report. Read the June 2026 announcement if someone on the team still expects clicks.
Export top pages by impression. Slice by country if the business is multi-market.
Load those URLs' buyer prompts into Promptwatch.
Keep GA4 for sessions. Label Overview organic honestly.
Do not report GSC impressions as clicks.

Official GSC documentation is the source for this measurement. Impressions are real. Clicks are not in the launch list. A prompt-level Overview check is a different instrument. Keep the instruments labeled, because a labeled instrument is what survives a QBR and an unlabeled one is what does not.