Performance in Google AI Mode can’t be isolated from the data you can actually get. The Search Console report shows impressions for AI Overviews and AI Mode together, traffic from assistants blends together in analytics, and a manual test shows you a single answer. What you can measure is the total and its direction, not the performance of one surface.
This article isn’t a tool walkthrough — those already exist. It’s about what the numbers you collect actually support, so no claim ends up in your report that won’t hold up.
Three sources and what each one tells you about AI Mode
| Aspekt | What it shows | What it won't tell you about AI Mode |
|---|---|---|
| Search Console | Impressions in Google's generative AI features | Doesn't separate AI Mode from AI Overviews — they sit in one view |
| Analytics | Visits from AI assistants and their quality | Won't tell you which Google surface the visit came from |
| Manual test | Whether an answer to a specific query mentioned you | Won't tell you volume, or how often it holds |
How to work with each one is covered in the GSC AI report, AI traffic and conversions, and the AI visibility test. This piece is about what to conclude from them.
Why the total is a weak proxy
It’s tempting to treat the combined number as “close enough to AI Mode.” But it merges two surfaces we know behave differently.
In an Ahrefs analysis of US data from September 2025 that captures individual generations, the cited URLs of AI Overviews and AI Mode overlapped by just 13.7 percent. AI Mode vs. AI Overviews goes through it in detail.
What you can actually claim
This is the practical part. The difference between a defensible claim and an indefensible one usually comes down to wording — but that wording decides whether the report holds up when someone reads it a year from now.
| Aspekt | Defensible | Indefensible |
|---|---|---|
| Impressions | Impressions in Google's generative AI features grew by X% | We have X impressions in AI Mode |
| Visits | AI assistants sent N visits, mostly from ChatGPT | AI Mode sent N visits |
| Test | In a repeated test, 4 out of 5 runs mentioned us on that query | We're visible in AI Mode for this query |
| Trend | The total is growing; we can't tell the surfaces apart | We're growing in AI Mode |
This isn’t hedging for its own sake. The left column says what was measured; the right one assigns a number to something that wasn’t.
Common mistakes
Passing the total off as AI Mode
The most common one. The report merges AI Overviews and AI Mode into a single view, so you can’t read one surface’s performance out of it.
Reading impressions as traffic
The report gives you impressions, not clicks and not CTR. An impression is a signal that you showed up, not a visit.
Drawing a conclusion from one run
Source selection fluctuates between runs. One answer you tried is an anecdote, not a measurement.
Mixing Google AI with other assistants
The Search Console report covers Google’s generative features only. ChatGPT and Perplexity aren’t in it.
What this looks like on real data
Our own store’s measurement shows the limit clearly. We recorded 200,000 impressions in Google’s generative AI features in under two months — but that number doesn’t separate the surfaces. And 1,867 visits from AI tools over twelve months, 90 percent of them sent by ChatGPT — a tool that isn’t in Search Console at all.
Both are useful numbers. Neither one says anything about how the site is doing specifically in AI Mode. The full data is in the case study, and we ran into the same limit in AI Mode for e-commerce.
Where this article stops
It isn’t a tool walkthrough. Where to find the report and how to segment it is covered in the linked articles.
It’s accurate as of publication. The report is in beta and rolling out gradually; if Google starts separating the surfaces, we’ll update this piece.
It doesn’t say measuring is pointless. It says don’t pretend to a precision the data doesn’t have.
Key takeaways
As of publication, AI Mode can’t be isolated from the data you can actually get. Search Console merges it with AI Overviews, analytics can’t tell the surfaces apart, and a manual test shows a moment rather than volume.
That doesn’t mean measuring blind. The total and its trend are a usable signal — you just have to say that it’s a total. The most valuable skill in reporting AI visibility today is telling what was measured apart from what we filled in around it.
And when someone offers guaranteed, measurable AI Mode performance, it’s worth asking what data they’re reading it from.
Want your AI visibility measurement set up so the numbers hold up? This site is run by Sniper Design — an AI SEO audit goes through your measurement setup and names what the data supports and what it doesn’t. We don’t promise to separate AI Mode; nobody can do that today.
Sources: the Generative AI features report in Google Search Console (Beta), an Ahrefs analysis comparing AI Overviews and AI Mode (US, September 2025), and our own store’s measurement. Accurate as of July 18, 2026.
