Search Console Counts AI Mode in Web Totals: 7 Workarounds Rated

Google counts a click from AI Mode exactly like a click on a blue link, and files both under the same Web search type. There is no filter in the Performance report that pulls them apart. For Search Console AI Mode measurement, that leaves you 1 of the 4 standard metrics you can isolate, and only since June 2026. The other 3 stay blended.

Short answer

Search Console does not separate AI Mode clicks from regular web clicks. AI Mode and AI Overviews are counted inside the Web search type of the Performance report. The Generative AI report, added in June 2026, shows AI impressions by page, country, device and date. Clicks, queries and position from AI features cannot be isolated anywhere in Search Console.

How Search Console counts AI Mode and AI Overviews in Web totals

Google's page on AI features states the rule in two sentences. Sites appearing in AI Overviews and AI Mode "are included in the overall search traffic in Search Console". They are "reported on in the Performance report, within the 'Web' search type," according to Google Search Central, last updated 10 December 2025.

That means AI traffic is not missing from your reports. It is present and unlabelled. A rise in Web clicks could be a blue link moving up, an AI Overview citation, or an AI Mode card, and the report treats all three as one stream.

The detail sits in a separate help page, "What are impressions, position, and clicks?" It has a dedicated section for each AI feature. The rules differ between the two features, and that difference is the root of most measurement confusion.

AI Overviews: one block, one position for every link

For AI Overviews, Google's help page is direct. "Clicking a link to an external page in the AI Overview counts as a click." Impressions follow the standard rules. On position, it says "An AI Overview occupies a single position in search results, and all links in the AI Overview are assigned that same position".

So your page can be the sixth source cited and still report the same position as the first source. If the Overview sits at the top of the page, every cited link reports a high position. If it sits lower, every link inherits that lower slot.

John Mueller explained the reasoning on Reddit in September 2026, as reported by Search Engine Journal on 13 September. "Position for these is hard to do in a way that makes it useful, so we're currently tracking it like we do for many search features (as a block)," he wrote. He added that it is "not separated out in the Gen-AI performance report."

AI Mode: each element gets its own position, and follow-ups reset the count

AI Mode works differently. Google says "Position in AI Mode follows the same methodology as a Google Search results page." Carousels and image blocks inside AI Mode use the standard position rules for those elements. A click on an external link counts as a click, and standard impression rules apply.

The rule with the largest practical effect concerns follow-ups. Google's help page says "If a user asks a follow-up question within AI Mode, they are essentially performing a new query". Every impression, position and click in the new response is credited to that new query.

That is why fragments such as "yes" or "tell me more" have started appearing in query reports. Nothing in Search Console marks a query as a follow-up. Glenn Gabe noted this when Google first documented the change on 16 June 2025: there is no indication that a query was part of a conversation.

What the Generative AI report adds, and what it leaves out

On 3 June 2026, Google announced Search Generative AI performance reports in Search Console. The post was written by Hillel Maoz and Moshe Samet of the Search and Search Console teams, and published on the Search Central blog. It began with a subset of sites, and trade press reported it on most properties by mid August.

The report shows impressions from generative AI features, covering AI Overviews and AI Mode in Search and AI features in Discover. You can break those impressions down by page, country, device and date. Google defines an impression as how often URLs from your site appeared in these features.

It does not show clicks, click-through rate, queries or position. That is 1 metric of the 4 you normally get. In our view this is the most important fact about the report, because it shapes every question you can answer with it.

The report is a slice of your totals, not an addition to them

The Generative AI data is still included in the overall Performance report. Adding the two together double counts every AI impression. Treat the new report as a lens on data you already had, not as a second source of traffic.

Mueller described the impression trigger on Bluesky in June 2026, reported by Search Engine Journal on 25 June. "The impressions are based on links to your site being shown in AI Overviews / AI Mode," he wrote. "If something needs to be 'activated' to see the link, it would only count when users do that."

That last sentence matters for AI Overviews. Links hidden behind a "Show more" control only register once a user expands them. A page cited deep in a long Overview may therefore record fewer impressions than its citation frequency suggests.

Two logging errors you need to know before trending AI impressions

Any trend line for AI search traffic runs across two disclosed data problems. Both are listed on Google's data anomalies page. Neither is widely mentioned in the dashboards that vendors sell for AI visibility.

The 349-day impressions error that covers AI Mode's first year

In April 2026, Google disclosed that a logging error "prevented Search Console from accurately reporting impressions from May 13, 2025 until April 27, 2026". Google says clicks were not affected and that the issue touched data logging only. Secondary coverage describes the error as inflating impressions, so corrected figures read lower.

We counted the span at 349 days. That window covers almost the entire first year of AI Mode in the US. The error is not attributed to AI features, and we are not claiming a link. The point is narrower: any year-on-year impression comparison that crosses April 2026 compares numbers logged under two different conditions.

The August 2026 gap in the Generative AI report

The new report had its own fault within weeks. Google's anomalies page says a logging error "caused a decrease in impressions on the Generative AI performance report in Search for data from August 13 - August 17, 2026". An update dated 21 August says the missing data was restored.

A separate entry covers Discover on 13 August, including its Generative AI slice. If you exported data during that window, your copy may still hold the lower figures. Re-export the dates rather than trusting a cached sheet.

7 ways to measure AI Mode traffic, rated by what they isolate

We collected every method we found recommended for separating AI search traffic, then checked each against Google's own counting rules. The test was simple. Does the method isolate traffic from AI features, or does it isolate something correlated with them and call it AI?

Ratings are our assessment against Google's documented counting rules, read 29 October 2026. We did not run a controlled test of each method on a large property, and the verdicts should be read as an analytical rating, not a benchmark.

The one method that isolates something real: page-level AI impression share

Because the Generative AI report is a subset of Web totals, you can divide one by the other. Pick a page, a date range and a country. Take the page's impressions in the Generative AI report and divide them by the same page's Web impressions over the same range.

The result is the share of that page's search visibility that came from AI features. It is the only number in this list that comes from Google's own logging and refers specifically to AI surfaces. It is also the one we would put in front of a director, with the caveats attached.

Here is how the arithmetic works on an illustrative page. Say a pricing guide logs 4,000 Web impressions in 28 days and 1,400 Generative AI impressions. Its AI share is 35%. If it earned 60 clicks, you still cannot say how many came from that 35%, and that is where the method stops.

Run the ratio across your top 50 pages by impressions and sort it. Pages with a high AI share and a falling click-through rate are your strongest candidates for the zero-click effect. The reasoning on which metrics replace the click is set out in the analysis of metrics for zero-click search.

Two counting differences limit the precision. The Performance chart aggregates by property, so two results from the same site on one query count once. The Generative AI view may not deduplicate the same way, and Google has not published the matching rule. Treat a ratio above 100% as a counting artefact, not as a finding.

Workarounds that only appear to isolate AI traffic

Three popular methods produce a confident number from a mixed population. Each one has a sound idea behind it. Each one also fails on a rule Google has already documented.

Long-query regex filters catch length, not surface

The common advice is to filter the Performance report with a regular expression that matches queries of 10 or more words. The theory is that AI Mode prompts are longer and more conversational than typed searches. That much is plausible.

The filter fails in three places. Long queries also appear in classic search, so the match includes blue-link traffic. Follow-ups such as "yes, pricing" are short, so the match misses AI Mode traffic Google has told you exists. And long queries are the ones most likely to be hidden.

Google withholds queries that, in the words of its 2022 performance data deep dive, "aren't issued by more than a few dozen users over a two-to-three month period". A long, unique prompt is close to the definition of that. The filter therefore samples the long queries common enough to survive privacy thresholds, which are the least AI-like ones.

Our own property shows the floor. Over 28 days to late October 2026, zandigital.in recorded 125 impressions, 0 clicks and no query rows at all. On a small site, a regex filter has nothing to match against.

GA4 referrer segments merge AI Mode back into organic

In May 2025, AI Mode links carried a noreferrer attribute, so clicks landed in analytics as direct traffic. Mueller called it "a bug on our side" on LinkedIn, and Search Engine Journal reported it fixed within days, on 28 May 2025.

The fix restored attribution and removed the only accidental signal. AI Mode clicks now arrive as Google organic, like every other Google search click. A channel group named "AI search" built on the google referrer will contain your blue-link traffic too, and will look healthy for the wrong reason.

Text fragments are a sample with a bias you cannot measure

Google often appends a highlight fragment, beginning with #:~:text=, to links from AI Overviews and featured snippets. Browsers do not send fragments to the server, so they never reach your logs. A script on the landing page can read them, and some teams tag those sessions as AI.

The trouble is coverage in both directions. Featured snippets add fragments too, so the tag overcounts. Not every AI link carries one, so it also undercounts. We found no Google documentation stating when fragments are added, so the size of either error is unknown. For non-Google agents, reading agent traffic in your server logs is the more reliable route.

Average position belongs in the same category. A blended average mixes AI Overview block positions with element positions from AI Mode and ten blue links. A rising average can mean an Overview moved up the page while your citation order inside it fell.

Where this argument is weakest

The case above rests on Google's documentation, and documentation describes intent, not behaviour. Google's own help page says its heuristics "are subject to change." Any rule quoted here could be revised without notice, and the ratio method depends on the report staying a strict subset of Web totals.

We also did not test each workaround on a large property. Our first-party data is a small site with almost no impressions, which tells you about the floor and nothing about scale. A large publisher with millions of impressions could find that a regex filter tracks AI Mode volume closely enough to be useful. That is an empirical question we cannot settle.

The case that blended data is good enough

Google's position, stated on the AI features page, is that AI traffic is simply part of search. On that view, separating it is a vanity exercise. What matters is whether total search clicks and conversions rise or fall, and those are fully measured.

We think that view is half right. For budget decisions, total search outcome is the number that pays the bills. For content decisions, it fails, because a page that loses clicks to an Overview needs a different fix from a page that lost its ranking. Blended data cannot tell those two failures apart. The case for treating them separately is made in the piece on measuring discovery that attribution misses.

What to report to your team about AI search traffic

This post is written for the practitioner who owns the search report. Your director wants one number for AI search. The honest answer is that one number does not exist, and the best response is a small set of numbers labelled by how much they can bear.

We would report three items. The site-wide AI impression share, computed from the two reports. The 10 pages with the highest AI share, alongside their click-through rate trend. And a written note that AI clicks are not separable, with Google's documentation linked so nobody mistakes the gap for laziness.

Do not build a dashboard that implies more precision than this. A chart labelled "AI Mode clicks" will be quoted in a board deck within a quarter, and it will be wrong. If you are wiring the data into a reporting tool, the guide to building an AI visibility dashboard covers the plumbing. Keep the labels honest when you do.

For context on what the Overviews themselves select, our earlier test of Google's AI Overviews claims against citation data covers ranking and citation. This post deliberately stays on the reporting side of that line.

Frequently asked questions

Where to start this week

Export the Generative AI report and the Web report for the same 28 days, by page. Join them in a spreadsheet and compute the AI impression share for your top 50 pages. Avoid any window that crosses 27 April 2026, and re-export 13 to 17 August if you pulled it before the fix.

Then rename one thing. If any report you own carries a label such as "AI Mode clicks" or "AI search traffic" built on referrers or regex, relabel it as an estimate and state the method underneath. That single edit prevents the most expensive mistake in this whole area.

Related reading