To measure AI Overview traffic loss, compare click-through rate (not just clicks) for the same pages and queries before and after AI Overviews started showing for them, while holding impressions and average position steady. As of September 2026, Search Console's new generative AI performance report tells you which pages get shown in AI Overviews and AI Mode, but it reports impressions only, so the loss itself still has to be calculated from the standard Performance report.
Key Takeaways
- Clicks falling is not proof of AI Overview loss. The signature is stable impressions and position with falling CTR.
- The Search Console generative AI performance report (rolled out globally on 31 August 2026, per Search Engine Land) shows AI impressions by page, country, device and date. It has no click data and no query dimension.
- Build a "control vs exposed" split: pages that appear in AI features vs pages that do not, then compare CTR change across the two groups.
- Estimate loss as: (baseline CTR − current CTR) × current impressions, per page cohort.
- Ahrefs' December 2025 data found a 58% lower CTR for position-one pages on keywords with AI Overviews. Use it as a sanity check, not as your own number.
Why "traffic is down" is the wrong starting point
Every client call I've had this year starts the same way: "organic is down 30%, is it AI Overviews?" Sometimes it is. Often it is a core update, a lost ranking on three money pages, seasonality, or a tracking change in GA4. If you blame AI Overviews without separating those causes, you will fix the wrong thing.
AI Overview loss has a specific fingerprint. Google still shows your result, so impressions hold or even rise. You still rank, so average position holds. What drops is the share of people who click, because the answer is sitting above you. Ranking loss looks different: position falls first, then impressions, then clicks. Demand loss looks different again: impressions fall across the board, including queries that never show an AI Overview.
So the whole method below is about isolating that one pattern.
Step 1: Open the generative AI performance report
Google announced the Search generative AI performance reports on the Search Central blog in June 2026, and Search Engine Land reported the global rollout on 31 August 2026. According to Google's help documentation, the report:
- Counts impressions: how many times links to your site were shown in a generative AI feature on Google Search.
- Covers AI Overviews and AI Mode, and excludes Search Labs experiments.
- Lets you group by pages, countries, dates and devices.
- Caps tables at 1,000 rows, and chart totals can differ from table totals.
- Only appears for sites with enough AI-feature impressions; not every property got it on day one.
What it does not give you, as of September 2026: clicks, CTR, position or queries. That is the gap. The report tells you where you are exposed; it does not tell you what the exposure cost.
Export the Pages view for the last 3 months. This is your exposed page list.
Step 2: Set clean comparison windows
Pick two windows of equal length that avoid known noise:
- Baseline: a period before AI Overviews were widespread for your market and topics. For many Indian sites this means going back to mid-2024 or earlier; Search Console keeps 16 months, so if your baseline is older, you need an export you saved earlier (another reason to archive GSC data monthly via the API or BigQuery bulk export).
- Current: the latest full comparable period, same months if you are seasonal.
Avoid windows that straddle a confirmed core update. Check Google's Search Status Dashboard for update dates before you lock the windows.
Step 3: Split pages into exposed and control cohorts
In the standard Performance report (Search type: Web), export page-level clicks, impressions, CTR and position for both windows. Then join it with the exposed list from Step 1 in a spreadsheet.
| Cohort | Definition | What you expect if AIO is the cause |
|---|---|---|
| Exposed | Page appears in the generative AI report | CTR falls, impressions and position roughly flat |
| Control | Page does not appear, similar intent and traffic | CTR roughly flat |
| Ranking-loss | Position worsened by 2+ in either cohort | Exclude from AIO analysis; diagnose separately |
The control cohort is the step most guides skip, and it is the one that makes your number defensible. If control pages lost CTR too, something else is going on: a SERP layout change, more ads, a new video carousel, or a site-wide issue.
Step 4: Hold position constant
CTR is heavily driven by position, so a page that moved from 2.1 to 3.4 will lose CTR regardless of AI Overviews. Filter both cohorts to pages where average position changed by less than about 1 place between windows. On small sites this shrinks the sample a lot; that is fine. A clean 40-page sample beats a messy 400-page one.
Step 5: Go to the query level for your top pages
Because the AI report has no query dimension, pick your 10-20 highest-impression exposed pages and open each one in the Performance report with the Queries tab. Then check which of those queries actually show an AI Overview today. You can do this manually in an incognito window with the right country, or with a rank tracker that flags AI Overview presence (Semrush, Ahrefs, SE Ranking and others do this as of 2026).
Tag each query: AIO present / AIO absent. Now compare CTR change on those two query groups on the same page. This is the tightest comparison you can get, because content, brand and page experience are identical.
Step 6: Calculate the loss
For each cohort or query group:
Estimated lost clicks = (baseline CTR − current CTR) × current impressions
Then subtract the control cohort's CTR change so you are not blaming AI Overviews for a general CTR drift:
AIO-attributable loss = (exposed CTR change − control CTR change) × exposed impressions
A worked example with round numbers: exposed pages had 2.8% CTR in the baseline and 1.6% now on 500,000 impressions. Control pages went from 3.0% to 2.8%. Exposed change is −1.2 points, control change is −0.2 points, so the AIO-attributable change is −1.0 point. That is roughly 5,000 lost clicks for the period. Multiply by your conversion rate and lead value and you have a number a CEO can read.
Step 7: Sanity-check against published research
Your number should be in a plausible range. Ahrefs' original study of 300,000 keywords found a 34.5% lower CTR for position-one pages when an AI Overview was present (March 2025). Its updated study compared December 2023 with December 2025 and found a 58% lower CTR for position one and 50.8% lower for position two, on informational keywords, desktop data.
If your measured drop is far beyond that, look for a second cause. If it is far below, your content may already be getting cited, or your queries may be less informational than their sample. Neither study breaks results down by query type, so treat them as a ceiling for informational queries, not a universal rule.
Step 8: Check whether you are cited, not just shown
Being in the AI report means a link to your site was shown in an AI feature. That can mean you are one of the cited sources, which sometimes earns clicks, or that your regular result was simply on a page where AI Overviews appeared. Check your top exposed queries manually: is your page in the citation cards? Pages that are cited tend to lose less than pages that sit below an AI answer built from competitors. That difference is where your recovery work goes.
Step 9: Build a repeatable monthly view
Once the method works, automate it:
- Pull the Performance report via the Search Console API or BigQuery bulk export each month.
- Export the generative AI report pages list monthly (API support for it was not confirmed at the time of writing, so plan for a manual export).
- Keep a sheet with the exposed/control flag, position-change filter and the loss formula.
- Chart three lines: exposed CTR, control CTR, and the gap.
The gap line is your AI Overview impact over time. When it narrows, your citation work is paying off.
Common mistakes I see
- Using GA4 sessions instead of GSC clicks. Consent banners, bot filtering and channel rules make GA4 noisier for this job.
- Mixing branded and non-branded queries. Branded queries rarely lose much to AI Overviews and will hide the effect. Filter them out with a regex on your brand name.
- Ignoring AI Mode. The generative AI report combines AI Overviews and AI Mode impressions. Don't label everything "AI Overviews" in your reporting.
- Reporting loss with no action. A loss number without a list of pages to rework is just bad news.
What to do once you have the number
Sort exposed pages by attributable lost clicks and value. For the top ones: add a short, direct answer near the top, add original data or examples that an AI summary can't replace, tighten structure so sections can be cited, and point the page at the next step (a tool, template, comparison or demo) so the visitors who do click have a reason to stay. When I've done this for edtech clients, the goal wasn't to win back every click; it was to make the clicks that remain convert better.
FAQ
Does Search Console show clicks from AI Overviews?
Not separately, as of September 2026. The generative AI performance report shows impressions only. Clicks from AI features are included in the regular Performance report under Web search, mixed with other clicks.
Can I filter the Performance report by "AI Overview present"?
No. There is no search appearance filter for AI Overviews at the query level. You have to tag queries yourself using a rank tracker or manual checks.
How far back should my baseline go?
Ideally to a period before AI Overviews appeared on your key queries. Search Console only stores 16 months, so archive exports monthly or enable BigQuery bulk export to keep longer history.
Is a 58% CTR drop normal?
That is Ahrefs' figure for position-one pages on informational keywords, comparing December 2023 with December 2025. Your number will differ by query mix, country and whether you are cited.
What if both exposed and control pages lost CTR?
Then AI Overviews are not the only cause. Check for more ads, new SERP features, a core update, or a sitewide snippet or title problem.
Should I block Google's generative AI features to protect traffic?
Google now offers a control to opt out of generative AI Search features, and says it does not affect core rankings. But opting out also removes you from citations in those features. Measure first; for most brands the visibility is worth more than the risk.
How many pages do I need for a reliable estimate?
There's no fixed number, but aim for enough impressions per cohort that small swings don't dominate. A few dozen pages with thousands of impressions each is usually workable.
Does this work for small sites?
Yes, but you may not get the generative AI report if you don't have enough AI impressions. In that case, use the query-level method in Step 5 with a rank tracker.
Want a second pair of eyes on your numbers?
If your dashboard says "down" and nobody can say why, I can help you separate AI Overview loss from everything else and turn it into a plan. I've spent 4+ years in organic growth, SEO and content for edtech and startup brands. See my work and reach me through the contact form at younusfardeen.in.