I spent part of the weekend building a system for my site, tracking which AI platforms are sending traffic, and when. Most of it worked. But one platform stopped me completely: Google AI Overviews. Every detection approach I tried failed, and when I dug into why, it turned out the answer is written into the browser specification itself.
Google AI Overviews is undetectable, and it's written into the browser spec
When Google AI Overviews cites a webpage, Chrome appends a text fragment to the URL — a string beginning with #:~:text= — that scrolls the page to the exact passage cited. I assumed this was readable by tracking code. It isn't. The Text Fragments specification requires browsers to remove it before JavaScript executes. This is an intentional privacy measure. The result: every click from a Google AI Overview registers in your analytics as either direct traffic or organic Google, permanently indistinguishable from someone who typed your URL or clicked a standard search result.
What to do this week: Open GA4 and look at your "Direct" channel volume for any month you published content targeting an AI search query. Compare it to the same month a year prior. If it grew, a portion of that growth is likely Google AIO citation traffic. You won't be able to separate it out. But you can stop treating "no AI traffic in analytics" as evidence that AEO isn't working.
Five AI platforms you can measure today if you know where to look
Not every platform is invisible. Perplexity passes a clean perplexity.ai referrer consistently. I confirmed this in testing. Claude does the same via claude.ai. ChatGPT began appending utm_source=chatgpt.com to citation links in June 2025. Gemini and Copilot pass referrers too. All five are trackable from their web versions but mobile apps strip referrers entirely, so any citation clicked on a phone won't show up in your analytics.
What to do this week: In GA4, go to Reports > Acquisition > Traffic acquisition. Filter by "Session source" contains "perplexity.ai" or "claude.ai." If your firm appears in Perplexity or Claude answers, you'll see it here. This is the most reliable direct signal that AEO work is generating real visits, and it costs nothing to check.
The best proxy for AIO traffic you already have access to
Since Google AIO clicks are untrackable, the next-best signal is the one that predicts whether you'll get them. An AirOps study of 16,851 queries and 353,799 pages found that pages ranking first in Google search were cited by ChatGPT 58.4% of the time, versus 14.2% for pages ranking tenth. A 4× gap, driven by search rank before anything else. The mechanism: AI engines retrieve content through search before generating answers. If your pages aren't on page one for a target query, they're largely invisible to AI citation regardless of how well they're written.
What to do this week: Pick three queries your ideal client would type into Google to find a practice like yours. Check your current ranking for each. If you're not on page one, that's the single highest-leverage variable for AI citation visibility ahead of schema markup, content length, or any technical optimisation.
The Google AIO measurement gap is permanent. It's built into how Chrome handles the URL fragment. The right response is to build a measurement approach that doesn't depend on referrer data: run your target queries weekly, record whether your firm gets named, track that number over time. That signal is available regardless of what any browser specification says.