Did ChatGPT traffic fall, or did tracking break?

ChatGPT-attributed sessions in GA4 fell noticeably on many sites from 9 October. According to the data Search Engine Journal compiled much of it looks like a measurement problem rather than lost visitors. The analysis shows how sites that get traffic from AI assistants should read their reports.

What happened?

Graphite's CEO Ethan Smith wrote that ChatGPT sessions in GA4 fell by between 22% and 57% on the sites he tracks, roughly a 45% drop from the previous baseline. LLM Pulse reported that on 9-11 October ChatGPT sessions on the sites it monitors were about 46% lower than on the same days of the previous two weeks. Total traffic on those sites stayed roughly flat.

According to Smith, ChatGPT began opening links in answer text without the utm_source=chatgpt.com tag and without a referrer. Those visits can show up as Direct in GA4.

Where did the missing visitors go?

Cedric Huesler of Adobe compared 9-13 October with the week before on about 500 pages where ChatGPT drove at least 5% of visits. ChatGPT visits fell by about 31,000 a day while Direct visits rose by about 26,000 a day; search traffic fell 7%. Huesler notes his findings are early and based on four days of sampled data.

Profound's data ties the change to a new answer format: the share of ChatGPT answers with at least one tagged source link fell from about 98% on 8 October to 66% by 14 October. Tomek Rudzki of Peec AI measured ChatGPT putting a direct URL in 42% of responses, compared with 8-10% before, and 54% of those URLs point to home pages.

Is it fixed?

Not entirely. Smith called it a bug on OpenAI's side; according to Huesler, OpenAI confirmed it on 14 October and a fix began rolling out early on 15 October. Huesler writes that tagged ChatGPT clicks were back to 95-100% of the same hour a week earlier across thousands of domains. Glenn Gabe said he was not yet seeing the change, and Profound said it had seen no official word on a fix. LLM Pulse did not observe a rise in Direct and said aggregated data cannot separate lost traffic from unlabelled traffic.

What should businesses do?

The episode showed that visits from AI assistants were being measured through a single tag. In CRO and analytics work, these steps help:

• Annotate 8-15 October in your reports; do not read the ChatGPT drop as lost traffic without checking Direct.
• Watch the pages ChatGPT refers to, and your home page, side by side with Direct; a jump in Direct can signal lost tags.
• Mobile app traffic is the most exposed: according to Graphite, 95% of mobile ChatGPT visits relied on the tag alone.
• Set UTM parameters yourself on your own campaign links; according to Smith, your own UTMs on ChatGPT Ads clicks are unaffected.
• Even if the fix lands, attribution already lost will not return; mark this week's data separately in year-on-year comparisons.

For teams measuring AI visibility, the main lesson is not to rely on a single signal: in AI visibility work, track how often your brand is mentioned in answers as well as clicks.