Organic sessions have stopped being a fair scorecard, because Google now answers many queries on the results page. Ahrefs measured a 58% lower clickthrough rate for the top-ranking page when an AI Overview is present, comparing December 2023 with December 2025 (Ahrefs), and Pew found people click a result 8% of the time when a summary appears versus 15% when it does not (Pew Research Center). Measure instead: AI impressions in Search Console, citation share across engines, referral sessions from chat domains, branded demand, and conversions per question group.
How much traffic are AI Overviews actually taking?
Two independent measurements point the same way. Ahrefs compared 300,000 keywords, half with AI Overviews and half informational without, using aggregated Search Console data. For AI Overview keywords the average desktop CTR at position 1 fell from 0.073 to 0.016, against a forecast of 0.037 without the feature; position 10 lost 19.4%. Non-AI Overview keywords also fell, from 0.076 to 0.039, which Ahrefs attributes to the wider shift toward zero-click features (Ahrefs).
Pew tracked 900 US adults through March 2025, covering 68,879 searches. 18% of searches produced an AI summary, and 58% of participants saw at least one that month. With a summary on screen, 8% of visits led to a click on a traditional result, against 15% without, and 26% of visits with a summary ended the browsing session, against 16% without (Pew Research Center).
| Measure | With AI Overview | Without |
|---|---|---|
| Position 1 desktop CTR, Dec 2025 (Ahrefs) | 0.016 | 0.039 |
| Visits with a click on a result (Pew) | 8% | 15% |
| Visits that ended the session (Pew) | 26% | 16% |
| Clicks on a link inside the summary (Pew) | 1% | n/a |
Why organic sessions stopped being a fair scorecard
Your Search Console data blends two different things. Google counts a click on an external link in an AI Overview or AI Mode as a click, applies standard impression rules, and gives an AI Overview "a single position in search results" with "all links in the AI Overview assigned that same position" (Search Console Help). A page can hold position 1, appear in more answers than ever, and lose half its clicks.
So a falling sessions line does not distinguish between losing visibility and gaining it in a format that pays in influence rather than visits. The fix is to report visibility and outcomes separately, and to stop treating sessions as the proxy for both.
What should you measure instead?
Five numbers, each with a home.
1. AI impressions. Search Console's generative AI performance report shows "how many times links to your site were shown to a user in a generative AI feature on Google Search", broken down by page, country, device and date, for AI Overviews and AI Mode. It excludes clicks, queries and Search Labs experiments (Search Console Help). This is your Google citation count.
2. Citation share. Fix a list of 20 to 50 questions your buyers ask and run them monthly in ChatGPT, Perplexity, Claude, Gemini and AI Mode. Record whether you are cited, which URL, and who else is. No vendor publishes this, so you build it.
3. Referral sessions from chat domains. Add a channel in your analytics for referrers such as chatgpt.com, perplexity.ai, claude.ai, gemini.google.com and copilot.microsoft.com. Expect undercounting: Cloudflare notes that traffic from Claude's native app carries no Referer header (Cloudflare).
4. Branded demand. Track branded queries in Search Console and direct traffic month over month. When a model names you in an answer, the next step is often a search for your name, not a click on the citation.
5. Conversions per question group. Group landing pages by the buyer question they answer and report leads or revenue per group, not sessions. A page that loses 40% of visits and keeps its leads is not a problem.
| Metric | Where | Blind spot |
|---|---|---|
| AI impressions | Search Console generative AI report | No clicks, no queries |
| Citation share | Your own monthly test set | Manual, sample based |
| Chat referrals | Analytics referral channel | Native apps send no referrer |
| Branded demand | Search Console queries, direct traffic | Slow to move |
| Conversions per group | CRM or analytics goals | Needs clean page grouping |
How do you set this up in a week?
Day one, open the generative AI report and export the last three months by page; it has been live for all properties since August 31, 2026 (Search Engine Land). Day two, write the question set with sales, not with a keyword tool, and run the first pass. Day three, create the referral channel and a branded query filter. Day four, map landing pages to question groups. Day five, build one dashboard with the five lines and retire the sessions chart from the top of it.
Qubba Marketing runs this scorecard for partners because it changes the conversation: from "traffic is down" to "we are cited on 14 of 30 questions and the gaps are these". The second sentence is something a team can act on.
Should you opt out of AI Overviews?
You now can. The Search generative AI control in Search Console defaults to Include; Exclude means your content "is prevented from being visible to users in Search generative AI features, including being linked to within these features and helping with grounding AI responses", across AI Overviews, AI Mode and Discover, with regular search results and the Gemini app unaffected and the change applied within one to two days (Search Console Help). Google says the setting is not used as a ranking signal (Google).
Weigh it honestly. Cloudflare's 2025 review found crawl-to-refer ratios as high as 500,000:1 for Anthropic, up to 3,700:1 for OpenAI, generally below 400:1 for Perplexity, and as high as 30:1 for Google (Cloudflare). For a publisher paid per pageview, exclusion can be rational. For a company that sells a service, an AI answer that names you at the moment of decision is the visibility you were buying with SEO in the first place, and the numbers above say the click was already going away.
Frequently asked questions
Does Search Console show clicks from AI Overviews?
Yes, but not separately.
Clicks on external links in AI Overviews and AI Mode count as ordinary clicks inside the Web search type of the Performance report.
The dedicated generative AI performance report shows impressions, pages, countries, devices and dates only, with no clicks or queries.
To estimate AI clicks you have to compare the two reports over the same period.
Are AI Overview citations counted as position 1?
Not exactly.
Google assigns the whole AI Overview a single position in the results, and every link inside it inherits that position.
If the overview sits at the top, your citation reports the same position as the top result, which is why average position can look stable or improve while clicks fall.
AI Mode positions follow normal results page methodology.
Which referrers identify AI traffic in analytics?
The common ones are chatgpt.com, perplexity.ai, claude.ai, gemini.google.com and copilot.microsoft.com.
Group them into one channel so the trend is visible.
Treat the figure as a floor: Cloudflare has documented that traffic from Claude's native app arrives without a Referer header, and other apps behave similarly, so some AI driven visits land as direct traffic.
Will blocking Google-Extended remove my site from AI Overviews?
No.
Google-Extended controls whether crawled content is used to train and ground Gemini models, and Google states it "does not impact a site's inclusion in Google Search nor is it used as a ranking signal".
The switch for AI Overviews, AI Mode and Discover's AI features is the Search generative AI control in Search Console, alongside nosnippet and max-snippet directives.