How to Measure ChatGPT Traffic in GA4, and What the October Tag Break Showed

Most ChatGPT visits reach analytics through one signal: the utm_source=chatgpt.com tag on the links in ChatGPT's answers. When the tag disappears, the visits keep coming and analytics files them as direct. That happened for about a week in October 2026, and it is the clearest lesson yet in how fragile AI referral data is.

What happened in October

After the GPT-6 rollout on 7 and 8 October 2026, links in ChatGPT's answer text began opening in a new tab with the referrer withheld and without the tag, Search Engine Journal reported on 15 October. Links in the Sources panel and in shopping results kept their tag. Profound measured the share of answers with at least one tagged source link falling from about 98% on 8 October to 66% by 14 October. Adobe's Cedric Huesler compared about 500 pages and found ChatGPT visits down by about 31,000 a day while direct visits on the same pages rose by about 26,000.

Huesler said OpenAI confirmed to him on 14 October that it was a bug, and that a fix began rolling out early on 15 October, with tagged clicks back to roughly 95% to 100% of the prior week across the domains he tracks. OpenAI has made no public statement. Our news post from 15 October covers the first reports.

How each engine's visits reach your analytics

Two consequences follow. ChatGPT traffic depends on the tag, and on mobile almost entirely: Graphite found 95% of its mobile ChatGPT visits identified by the tag alone. And Google's AI surfaces cannot be told apart from ordinary search clicks in GA4 at all. Gemini's tagging arrived in the same month.

Five ways to measure it that survive the next break

1. Give AI assistants their own channel

If your GA4 property does not already show an AI assistant channel, add a custom channel group with one rule: source matches chatgpt.com, perplexity.ai, gemini.google.com, claude.ai or copilot.microsoft.com. Every tagged or referred visit lands in one line you can chart.

2. Watch direct traffic on deep pages

Direct visits to your homepage are mostly people typing your address. Direct visits to a deep article nobody types are clicks that lost their source. A rise there in the same week a referral source falls is a labelling change, and the visitors are usually still arriving.

3. Read the ChatGPT-User lines in your server logs

OpenAI documents ChatGPT-User as the agent ChatGPT uses when it visits a page to answer a user's question. Its log lines show which of your pages ChatGPT reads while answering. They count pages in play, not visitors, and they depend on no tag.

4. Count impressions where clicks are not reported

For Google, the Generative AI features report in Search Console shows the impressions your pages earn inside AI Overviews and AI Mode. Impressions rising there while organic clicks fall is the pattern of a page being read inside the answer.

5. Track the answer itself

Run a fixed set of the questions your buyers ask, on a schedule, and record whether your brand is named and which pages are cited. Visits follow mentions, and the mention is measurable whatever the link carries.

Mark the week on every report

Annotate 8 to 15 October 2026 in GA4 and in every client report. A month-on-month comparison that includes that week reads as a drop in AI traffic that mostly did not happen. Report the AI channel, direct visits to deep pages and share of answers side by side, so a break in one signal shows up as a gap and is not mistaken for a trend.

Run a free AI visibility audit to see how often AI names your brand today.