AIO Delta

Guide

How to measure AI Overview impact in Google Search Console, step by step

2026-10-21

Search Console doesn't have a "clicks lost to AI Overviews" number. Google counts AI Overview clicks and impressions inside the normal Web search type of the Performance report (Google, AI features and your website). Since 31 August 2026 there is also a Generative AI performance report for every site. It shows AI Overview and AI Mode impressions by page, but no clicks, CTR or queries (Google).

So you have to infer the impact. The method below uses only Search Console and a spreadsheet. It finds queries where your ranking and visibility held but clicks fell, then estimates how many clicks went missing. Plan on about 30 minutes the first time.

What you're looking for

The pattern, sometimes called the "great decoupling" (Ahrefs), looks like this for a single query:

  • average position is about the same or better
  • impressions held or rose
  • clicks and CTR fell

If position fell, that's a ranking problem. If impressions fell, demand or visibility fell. Neither is the AI Overview pattern, so keep them out of this analysis.

Step 1: Open the right report

In Search Console, choose your property and open Performance > Search results. Check the search type filter at the top. Google now splits web results into "Web: text-based" and "Web: multimodal" (Google, Performance report). Start with text-based web search, since that's where most AI Overview queries happen.

Use a Domain property if you have one. It covers all subdomains and protocols, which avoids the common mistake of looking at the wrong version of the site (Google, traffic drops).

Step 2: Compare two equal periods

Click the date filter, open the Compare tab, pick two date ranges of the same length, and click Apply (Google, filtering and comparison). Good defaults:

  • Last 28 days vs previous 28 days for a routine check.
  • The weeks after a drop vs the same number of weeks before it, if you know roughly when clicks fell.
  • The same period year over year, to rule out seasonality. Google's debugging guide recommends this too (Google).

Google suggests weekly or monthly granularity when comparing ranges, so you don't compare a Monday with a Saturday.

Step 3: Narrow the query set

Open the Queries tab. With a comparison active, the table shows each metric for both periods and a Difference column you can sort by (Google, common tasks).

Two filters cut down the noise:

  • Non-branded queries. Search Console has a branded / non-branded query filter, with data from 11 March 2025. It isn't available for sub-properties or sites with low impressions (Google, common tasks). Branded searches behave differently and can hide the pattern.
  • Informational queries (optional). Ahrefs reports that 99.2% of keywords triggering AI Overviews in its data were informational (Ahrefs). You can add a "Custom (regex)" query filter that matches question words and guide terms, for example ^(what|how|why|when|which|who) |guide|vs|difference between. Don't rely on it alone, though: AI Overviews also show on some commercial queries.

Step 4: Export the table

Use the report's export button to download the chart and table data (Google, Performance report). The download includes the comparison columns for the active tab. Export the Pages tab as well if you want a page-level view.

Two limits to know about, both from Google's docs:

For larger sites, the Search Analytics API or bulk data export gives you more rows.

Step 5: Apply the CTR filter in a spreadsheet

Open the export in a spreadsheet. For each query you'll have clicks, impressions, CTR and position for both periods. Add four columns:

ColumnFormula idea
Position changecurrent position − previous position (positive = worse)
Impression changecurrent impressions − previous impressions
CTR change (relative)(current CTR − previous CTR) ÷ previous CTR
Estimated missed clickscurrent impressions × previous CTR − current clicks

Now keep only rows that meet all of these:

  1. Position change ≤ 1 (didn't get worse by more than about one place)
  2. Impression change ≥ 0 (impressions held or rose)
  3. Clicks fell, and CTR change ≤ −20%
  4. Current impressions ≥ 100, so small numbers don't create noise

These thresholds are our suggestions, not a standard. Google doesn't define one. Pick your own, write them down, and use the same ones every time so you can compare month to month.

Sort what's left by estimated missed clicks. That's your shortlist.

Step 6: Cross-check with the Generative AI report

Open the Generative AI performance report, use the Pages and Dates dimensions, and look up the pages that rank for your shortlisted queries (Google). If AI-feature impressions for a page rose around the time its clicks fell, that supports the AI Overview explanation.

Keep in mind what the report can't show. It counts only times your links appeared in AI features. It doesn't show Overviews that appeared without linking to you, and it has no query data. A page that lost clicks to an Overview citing someone else won't show up there.

Step 7: Look at the live results page

For your top 10–20 queries, run the search yourself in a private window and note:

  • Is there an AI Overview?
  • Is your page one of its links?
  • What else is above your link (ads, videos, forum results)?

Results vary by location, device and history, and AI Overviews change often (Ahrefs), so treat this as a spot check, not a measurement.

Step 8: Repeat on a schedule

One snapshot can mislead you. AI Overviews roll out to new queries over time, and Google changes how they display. In October 2026, for example, Google announced more prominent inline links in AI Overviews and AI Mode (Google, Oct 2026). Rerun the comparison monthly, or weekly for important sites, and track the same shortlist over time.

Skip the spreadsheet

Steps 4 and 5 are the slow part. The free AIO Delta check runs the same filter on a Search Console Performance export made with Compare dates. Upload the export or paste a table. It keeps the rows where position held or improved and impressions didn't fall but CTR and clicks did, then estimates missed clicks with the same formula as above. No login is needed.

FAQ

Which date ranges should I compare? Two periods of equal length, such as the last 28 days against the previous 28 days. Also run a year-over-year comparison for the same weeks to rule out seasonality. Weekly or monthly granularity smooths out day-of-week effects.

What counts as a stable ranking? There is no official threshold. A practical rule is that average position didn't get worse by more than about one place. Set the threshold yourself and keep it the same every time you rerun the analysis.

How do I estimate the clicks I lost? For each flagged query, multiply current impressions by the previous period's CTR and subtract current clicks. That gives the clicks you would have had at your old CTR. It's an estimate, not a number Google reports.

Why don't my exported rows add up to the chart total? Google's docs say the table shows at most 1,000 rows and leaves out anonymized queries, while the chart includes them. Exports from the interface are a sample of your top rows, not every query.

Can Search Console tell me which queries showed an AI Overview? No. The Generative AI performance report shows AI Overview and AI Mode impressions by page, country, device and date. It has no query dimension and no clicks, so you still need to check queries yourself.

Sources