How to check in your own Analytics whether AI is affecting your clients' traffic
You've spent months reading headlines: that AI is eating organic traffic, that SEO is dead (again), that your clients are losing visits without anyone noticing. And your reaction, if you're remotely serious, is the right one: "show me the data from MY clients, not from some American study."
Good. This article is exactly that: the method to check it yourself, with the tools you already have — Search Console and GA4 — in an afternoon. And, because scepticism deserves to be met with honesty, we'll also tell you where the method limps and what you won't be able to see no matter how hard you look.
Signal 1: impressions rising, clicks falling (Search Console)
It's the clearest fingerprint of Google's AI Overviews: your client still appears (impression), but the user no longer clicks because the generated answer was handed to them ready-made above the results.
How to look at it:
- Open the client's Search Console → Performance → Search results.
- Period: last 16 months (the maximum). Switch on all four metrics: clicks, impressions, CTR and position.
- Look for the scissors shape: a flat or rising impressions line while the clicks line falls. If the average position holds, even better for the diagnosis: the client hasn't lost ranking, they've lost the click.
- Drill down to query level. Filter the informational queries ("how", "what is", "which is best") and compare them with transactional or branded ones. AI Overviews hit informational queries first; if the CTR on those has collapsed while branded holds up, you've got the pattern.
To calibrate what you see: in Spain an average drop of 61% in organic CTR has been measured when an AI Overview is present (ismajimenez.com), and searches that end without a single click have gone from 56% in May 2024 to 69% in May 2025 (data cited in stucom.com). If your clients show the scissors shape, they're not an anomaly: they're the average.
The seasonality adjustment, which is where false positives sneak in: never compare March with August. Compare each month with the same month of the previous year. A hotel loses clicks every November; that's not AI, it's November. The correct question is: is May 2026's CTR worse than May 2025's at similar positions? If the answer is yes across several informational queries, there's something there.
Signal 2: referrals from AIs (GA4)
The flip side of the coin: AI doesn't only take clicks away, it also starts bringing them. ChatGPT, Perplexity, Copilot and Gemini send traffic with an identifiable referrer (chatgpt.com, perplexity.ai, copilot.microsoft.com, gemini.google.com), and you can see it in GA4 with a filter on the session source.
The full step-by-step — including the "AI Assistant" channel that Google activated in GA4 in May 2026 — is in the tutorial for seeing ChatGPT and Perplexity traffic in GA4. For today's diagnosis, take away this: if that traffic exists and grows month on month, AI is already in your client's funnel, like it or not. And it's worth looking at its quality, not just its volume: the sector data compiled by roymo.es points to traffic from AI answers converting at 14.2% versus 2.8% for classic organic, because whoever arrives from a ChatGPT recommendation arrives with the decision almost made.
Signal 3: the direct traffic that doesn't add up
Here comes the awkward part: a large share of the traffic coming from AIs arrives with no referrer — from the mobile and desktop apps, above all — and GA4 classifies it as direct traffic. You can't attribute it, but you can suspect it:
- Is direct traffic to deep pages (a specific article, a service page) growing with no campaign to explain it? Nobody types
/blog/mejores-gestorias-granada-2026from memory. If that URL is getting "direct" traffic, someone gave it to them: an AI chat is the usual suspect. - Is direct traffic growing in parallel with the AI referrals from signal 2? That correlation is the typical pattern of hidden AI traffic.
Signal 4: branded vs non-branded
In Search Console, separate the queries containing the client's brand from the rest. If branded searches hold but the generic ones ("emergency plumber Seville") fall, the business isn't losing interest: it's losing discovery. And discovery is precisely what's moving into conversations with AI: frequent use of ChatGPT in Spain has gone from 4% to 28% in two years (Funcas, III Survey on AI, 2026).
The limits of the method (read them before you present anything)
We promised honesty:
- Correlation isn't causation. A falling CTR could be AI, but it could also be a core update, a new competitor who's got their act together, or a snippet that changed format. The method gives you converging signs, not a forensic proof. That's why you look at four signals and not one.
- Search Console doesn't label AI Overviews. There's no "this click was lost to an AI Overview" filter. You infer from the impression-without-click pattern, and the inference has a margin of error.
- The most important thing doesn't show up in any analytics. Your Analytics measures what happens on the client's website. It doesn't measure what ChatGPT answers when someone asks "which company do you recommend for X?" and your client doesn't appear — or appears with false data. That loss is invisible by design: the user who never hears of you through AI never generates a session you can analyse. And it's the part that's growing fastest: Gartner projects 25% less traditional search volume before the end of 2026 (press release, Feb 2024).
That last frontier is the reason we measure both sides separately. Your Analytics tells you how much traffic arrives; what's missing is knowing what the AIs say about your client when nobody clicks. For that half we built Surfeo: it asks the 4 AIs every week what your client's customer would ask, and tells you where they appear, where they don't, and what's said about them.
What to do with the diagnosis
If the signals converge, you've got two conversations ahead. One with the client, to explain the drop without sparking panic — here's how to tell it with the data in hand. And one with yourself: if the "do I show up in ChatGPT?" question hasn't reached you yet, it will soon, and it's worth having the script ready.
Frequently asked questions
How much history do I need for the analysis to be reliable?
At least 13 months, so you can compare each month against the same month of the previous year and neutralise seasonality. Search Console keeps 16 months; if you want more, export the data now and start building your own history.
What if I don't see any of the four signals?
Then you've got an equally valuable finding: for that client, in that sector, AI hasn't yet touched their traffic in a measurable way. Document it with a date and repeat the analysis every quarter. "We're watching it and it's not affecting you for now" is a sentence that strengthens a retainer, not one that weakens it.
Can I automate this check for the whole portfolio?
The Search Console and GA4 part, partially: saved explorations and a dashboard per client. The part about what the AIs answer doesn't come out of your Analytics and has to be measured separately, by hand with screenshots or with a tool.
Does this work to prove to the client that they need to invest in AI visibility?
It works for something better: to decide with data whether they need it. If the signals appear, the proposal sells itself. If they don't, you've gained credibility for the day they do.
An afternoon of analysis gives you half the picture. The other half — what the AIs say about your client when nobody clicks — you get in minutes with the free AI visibility test: you enter their site and see where they appear and where they don't.
