ChatGPT sent real people.

Your visit is counted the same way every row here was: a page path, a coarse country and city, a timestamp, and dwell time. Your IP is hashed at ingest to resolve city-level geography, then discarded — never written to the database. No cross-site cookie, no email, no name, no device fingerprint.

Two signals, either sufficient: the browser reports chatgpt.com as the referring document, or the landing URL carries utm_source=chatgpt.com, which ChatGPT appends to links it surfaces. The via tag on each row names which one proved that session — the evidence is on the page, not in our word for it.

Because the table lists pageviews, not visits. One visitor who lands on pricing, opens signup, then returns to pricing produces three rows. The sessions counter counts that person once; the pageviews counter counts all three. Every pageview in the session is attributed to ChatGPT, because the referral is what put them on the site at all.

Known crawlers and automated agents are flagged at ingest and excluded from both the table and the counters. No bot filter is perfect — but a visitor who loads four pages over ten minutes with real dwell on each is behaving like a person, and the dwell column is published so you can judge that yourself.

Yes. In GA4: Reports → Acquisition → Traffic acquisition, add Session source as a dimension, and look for chatgpt.com and perplexity.ai. Most owners have never filtered for them — if yours are bucketed under Referral or Direct, you are probably undercounting AI traffic.

Because “AI search is a real acquisition channel” is usually asserted by people selling something and almost never evidenced. This is the evidence for our own domain. The numbers are modest and we publish them anyway — a small real number argues better than a large unsourced one.