How to Track AI Overviews Traffic in Search Console

AI Overviews traffic has no dedicated report in Google Search Console, so you infer it. Segment likely overview-triggering queries with a regex filter, compare their click-through rate at stable positions with a control group of queries that never trigger overviews, then pair that view with a GA4 expected-clicks formula. The result is a defensible click-loss number instead of a vague worry. This tutorial walks through the full workflow with screenshots, formulas and a monthly reporting template.

Why nobody can find AI Overviews traffic in their reports

Google folds every AI Overview impression and click into the standard web search totals in Search Console. You will not find a filter to flip, a tab to open or a dedicated line item. The data sits inside the blended totals, and Google has given no indication that a native segment is coming soon.

That absence makes anxiety around AI Overviews traffic feel hard to pin down. You cannot point to one report and say, there is the damage. The pattern still has a shape you can measure. Picture a how-to article that has held position 4 for a year. Rankings look healthy. Then an AI Overview starts triggering for its main query, and over three months CTR slides from 8% to 3% while position stays put. Nothing in the default view screams. The clicks simply halve.

That stable-position, falling-CTR signature is the evidence. The rest of this post extracts it cleanly through query segmentation in GSC, landing-page validation in GA4 and two formulas that turn a feeling of decline into a number you can defend in a meeting.

Does Google Search Console show AI Overviews impressions separately?

No. Say that plainly so the rest earns your trust: every method in this post is inference from combined data. Search Console has no separate AI Overview segment today, and a dashboard claiming to read one directly is adding steps to tea leaves.

Google documents how impressions and positions behave when overviews appear. A page cited inside an overview earns an impression whenever that overview is visible. A page ranking below it earns an impression as usual. Both land in the same bucket as classic blue-link impressions. Citations can inflate impressions while clicks barely move, so impression growth beside flat clicks is a soft signal worth noticing.

The featured snippet era used the same measurement-by-inference playbook. If you lived through it, the muscle memory transfers. I unpacked the parallels in AI Overviews vs Featured Snippets: What Actually Changed, and the mechanics of how these answers get assembled live in our AI Overviews hub. Waiting for Google to ship a native report is a losing strategy either way.

How to spot AI Overview impressions hiding in your totals

Look for query clusters where impressions rise or hold steady while clicks fall in the same window. On informational queries, that split is the most common trace an overview leaves. Confirm it by searching the query in an incognito window and checking whether an overview renders.

Why average position lies to you in the AI Overview era

Links cited inside an overview inherit the position of the entire block, which usually sits at or near the top of the page. A page that used to report position 6 can suddenly report position 1 the moment it gets cited, even though the visual link is a small chip inside a paragraph of AI-written text. Average position can improve while traffic falls. If you take the position number at face value, you will diagnose the opposite of what is happening.

How to measure AI Overviews traffic in Google Search Console

Now we work at the screen. Open the Performance report and secure your baseline before anything else.

That sixteen-month window is precious because it likely reaches back before AI Overviews expanded in your market. Export it through the Export button in the top right with full query-level detail. The window rolls forward every day, and your pre-overview baseline erodes with it.

The regex patterns that isolate AI Overviews in Google Search Console data

Click the Query filter, switch it to "Custom (regex)," and paste a pattern that catches question-shaped queries. Those trigger overviews far more often than most other query types:

(?i)^(how|what|why|when|where|which|who|can|does|do|is|are|should|best way)\b

That one line carves your query universe into a flagged segment of likely overview triggers. It will catch false positives. Regex gives you candidates, and human verification makes it honest: take 20 to 30 flagged queries, search each in an incognito window, and record two things in a spreadsheet - whether an overview triggers and whether you are cited in it. Twenty minutes of tedium hardens every conclusion you draw later.

Building your control group of overview-free queries

Then invert the logic. Filter for branded and navigational queries where searchers type your product name and no overview appears. This control group matters more than the flagged segment. If both segments decline together, something else is moving: a title change, a new SERP feature or a seasonal dip. The control group keeps you from blaming AI Overviews for your own meta descriptions.

Worked screenshot walkthrough: from filter to first finding

Take one query end to end. Suppose "how to clean cast iron skillet" is in your flagged segment and your page has sat at position 4 the whole time. Use the Compare tab with the same three months this year versus last year, keeping the regex filter applied. Last year: 9,200 impressions, 736 clicks, 8.0% CTR. This year: 10,100 impressions, 313 clicks, 3.1% CTR. Position 4.2 versus 4.1. Impressions rose, position held, CTR fell 61%. Run the same comparison on the control group and see branded CTR flat at 31%. That contrast on one screen is your first finding.

The GA4 side: sessions, landing pages and the expected-clicks formula

GSC alone is not enough because clicks can redistribute across pages in ways the query view hides. Open GA4 and inspect landing-page sessions filtered to organic search. Confirm that the decline is real at the page level and specific to organic. Add an annotation on the date AI Overviews expanded in your market so future reports carry that context without relying on memory.

Calculating your own position-band CTR curve

Resist the urge to borrow an industry-average CTR curve. Your share of branded queries, snippet ownership and title style makes your curve yours. From the sixteen-month export, use only pre-overview months, group queries into position bands (1-2, 3-4, 5-6, 7-10) and compute total clicks divided by total impressions for each band. That historical CTR curve becomes the honest denominator for everything that follows.

The click-loss formula, worked with real numbers

The formula is almost embarrassingly simple:

1. Expected clicks = current impressions x your historical CTR for that position band
2. Click loss = expected clicks minus actual clicks

Run it on the flagged segment. Suppose the segment logged 12,400 impressions this quarter at an average position in your 3-4 band, where your historical CTR is 6.8%. Expected clicks: 843. Actual clicks: 512. Click loss: 331 clicks this quarter, attributable with reasonable confidence to overviews absorbing demand above you.

How do you know if AI Overviews are stealing your clicks?

You now have four data points per query cluster: position trend, flagged CTR trend, control behavior and citation status. The verdict comes from how they combine.

The four verdicts your data can return

The first row is what this workflow exists to catch. The third row is what it exists to rule out, because blaming AI for a plain ranking decline wastes a quarter on the wrong fix.

Cited in the overview vs left out: two different stories

Track cited and uncited clusters as separate populations because they behave differently. Citation softens click loss and can lift branded search weeks later, when people who read the overview return by name. Uncited pages under an overview pay the full tax without collecting a dividend. Before you conclude anything, respect one threshold:

Below that level, month-to-month noise drowns the signal. Aggregate individual queries into clusters until each segment clears the bar. If you are choosing a downstream metric once citations enter the picture, start with our breakdown of mentions versus citations in the generative engine optimization hub.

Build the monthly AI Overviews click-through rate report

All of this should fit on one page because the audience has about thirty seconds.

The one-page report template

The verdict line is the part people skip. It should not be. "Overview impact contained to two clusters, estimated 331 clicks lost, both clusters queued for optimization" beats any chart you can draw.

Adding assistant citation checks to the same report

Google is only one answer engine. Pair the GSC view with prompt-based visibility sampling across ChatGPT, Perplexity and Gemini so the report covers where answers actually happen. Sample size matters more than people expect. Read the math on how many prompts you need to measure AI visibility before trusting a single check. Third-party trackers like theairankings.com and checkthat.ai provide useful triangulation against your own sampling. Two independent methods agreeing is worth more than either alone. Monthly is the right cadence for CTR trends, while weekly fits citation checks on launch-critical queries.

This is also the loop AstroFabric's AI visibility agent runs on a schedule. Metered tool calls handle sampling, the code sandbox handles exact computation, and the finished report lands in Slack or email. Every write stays approval-gated, so nothing on your site changes without sign-off.

From measurement to AI Overviews optimization: acting on the numbers

Measurement without triage is organized worrying. End each report with a decision.

Which pages to fix first

Defend pages with high click loss and no citation first. They pay the AI tax without collecting the visibility dividend, which makes them both your biggest losses and your fastest potential wins. The fix playbook is generative engine optimization in its full form. It calls for extractable answers near the top, first-hand specifics an overview wants to quote and structure clean enough for a machine to lift.

Proving the fix worked with the same formulas

Give re-measurement at least a full monthly cycle before judging anything. Then run the identical expected-clicks math on the identical segments. The same formulas that quantified the loss become your before-and-after proof. That symmetry makes the whole system credible. Marketers who measure this now will own the citation slots their competitors are still anxious about. The setup takes an afternoon, maintenance runs about an hour a month, and that is a fair price for replacing dread with a dashboard.

Put the workflow on a schedule

If you would rather have the regex segments, the sandbox math and the monthly report run themselves, AstroFabric's AI visibility agent performs this loop with scheduled checks, exact computation, delivery to Slack or email, and approval gates on anything that touches your site. Start free at /signup and have your first click-loss number before the month closes.

Frequently asked questions

Sources

• theairankings.com
• checkthat.ai