Your site is already being read by AI.
When ChatGPT or Perplexity assembles an answer, something is fetching the page behind the scenes — a training crawl, a search-and-RAG lookup, a fetch triggered by someone’s question. Different purposes, same fact: a request lands on your server.
None of it shows up in GA4. Bots don’t run JavaScript, so the tracking tag never fires. It doesn’t show up in Search Console either — that tool records Google Search impressions and clicks, not AI crawler behavior.
The dashboards most site owners check every day were never built to see this traffic. It isn’t missing. It’s just invisible from where you’re looking.
The AI-era playbook is already crowded with advice: structure your data, write in a way AI likes to cite.
That’s backwards. Acting on AI traffic before you’ve measured it is SEO without analytics — optimizing for a channel you can’t see.
Plenty of tools now track how AI cites you in its answers — the output side. Almost none track what AI actually read to get there — the input side. That data exists only in your own server logs. It was never missing; there was just no tool to read it.
Here’s one real number.
On one site, during a seasonal event that drew a spike in related search interest, visits from search-and-RAG bots — the type that fetch pages to ground an AI’s answer — rose to roughly 3x their baseline.
The takeaway is simple: when human interest moves, AI-driven information demand moves with it. Someone asks an AI a question; the AI goes and fetches the source. For the length of that event, the site was functioning as a reference source through an entirely separate channel — one that never appeared on a single GA4 screen.
Without measurement, that fact simply wouldn’t exist.
This is where it becomes a method, not just an anecdote.
Once you have a bot-visit timeline, you can plot it against the same window as your Search Console clicks or pageviews. Did AI-driven access move with search traffic, or independently of it? Is the ratio stable, or widening?
That’s what tracking an “AI-to-total traffic ratio” means in practice.
One caveat: this comparison lines up two independent counts over the same period — it doesn’t establish causation or attribution. Bot requests carry no cookies, no sessions; there’s no principled way to tie them to a specific human action. That’s precisely why counting them side by side is still useful. A measurement that refuses to overclaim holds up longer than one that doesn’t.
Getting started doesn’t take much.
If you have raw server logs, filtering known AI bot user agents gets you a first count today. On WordPress, a plugin gets you there even faster — no log access required.
That’s what EdgeShaping is for.
The tool matters less than when you start. Raw logs can’t be recovered retroactively. Only the people measuring today will be able to talk about what changed six months from now.
How AI treats your site isn’t something to guess at. It’s something to measure. Start by counting.