Sean Walsh
Sean is a Director at Intelligency heading up our digital marketing and client services operations. Sean has 15+ years experiencing working both in-house and agency with brands including Lloyds, Alstom, Hitachi, Lufthansa, Viaplay, DFDS Seaways and Mercedes-Benz.
If your ChatGPT referral traffic seemed to fall sharply in the second week of October, you are not alone, and you probably did not lose those visitors. For about a week, ChatGPT stopped labelling the links it sends people to. Analytics tools could not tell where those visits came from, so they filed them as “direct” traffic instead.
OpenAI has now started fixing the problem. But the episode is a useful warning for anyone who reports on AI search performance: the data is more fragile than it looks, and a dip in a chart is not always a dip in demand.
What happened
In June, OpenAI began adding a small tracking tag (utm_source=chatgpt.com) to links inside ChatGPT answers. That tag is what lets Google Analytics 4, Matomo and similar tools recognise a visit as coming from ChatGPT. It has quickly become the main way marketers measure AI referral traffic.
Shortly after GPT-6 launched on 8 October, that tag disappeared from many links. Cedric Huesler at Adobe found that links were also being opened in a way that strips out the referrer, the other signal analytics tools use to identify where a visitor came from. With no tag and no referrer, the visit lands in the direct bucket.
The scale was significant. Across around 500 pages where ChatGPT normally drove at least 5% of visits, Huesler saw ChatGPT visits fall by about 31,000 a day between 9 and 13 October, while direct visits on the same pages rose by about 26,000 a day. Other trackers told the same story. LLM Pulse recorded around 46% fewer ChatGPT sessions, and Profound found that the share of ChatGPT answers containing a tagged link fell from about 98% to 66% in less than a week.
It was a bug, but the gap in your data is permanent
On 14 October, OpenAI told Huesler privately that this was a known bug, and a fix began rolling out the next day. Tagged ChatGPT clicks returned to near-normal levels within hours. OpenAI has not made a public statement, and some analysts were still seeing untagged links while the fix rolled out.
Two points matter for reporting:
- Visits recorded as direct between roughly 8 and 15 October will not be reclassified. No backfill has been announced, so October reports will understate ChatGPT and overstate direct traffic.
- The referrer is still missing on most ChatGPT clicks. The tracking tag is now effectively the only signal, so if it breaks again, the same thing will happen.
Why this matters beyond one bad week
AI search is becoming a meaningful source of high-intent visitors. Someone who clicks through from a ChatGPT recommendation has often already shortlisted you. That makes it tempting to treat the “AI assistant” line in GA4 as a reliable performance metric.
This episode shows its limits. AI platforms control how their links behave, they change them without notice, and they are under no obligation to help you measure them. A product update on their side can make your AI channel appear to collapse overnight, even when nothing has changed in how many people find you.
The bigger risk is a bad decision. A business that saw ChatGPT traffic halve in October might conclude that its AI visibility work had stopped paying off and cut budget, when the visitors were still arriving all along.
What to do now
- Annotate 8 to 15 October in GA4 and Matomo so nobody misreads the dip in future month-on-month or year-on-year comparisons.
- Report ChatGPT and direct traffic together for that week. If the combined total held steady, you lost measurement, not visitors.
- Check the pages ChatGPT normally sends people to. A rise in direct entrances to deep pages, such as a specific treatment or course page, is a strong sign of misattributed AI traffic, because people rarely type those URLs directly.
- Judge AI visibility on more than one number. Pair referral data with visibility tracking (how often you are mentioned and cited in AI answers), branded search trends and enquiry or conversion data.
- Do not add UTM tags to your own internal links. They overwrite the original source and make attribution worse.
