TLDR;
Industry observers say many links opened from GPT-6 answers have stopped passing ChatGPT's tracking tag or a referrer, so analytics records the visits as Direct. OpenAI has not confirmed the change, and its current publisher guidance still says the tag is added automatically. Enterprise brands should annotate the October reporting break and check Direct traffic before concluding that ChatGPT visibility or visits have fallen.
What happened
On 14 October, Search Engine Roundtable reported that links in ChatGPT answers appeared to be losing the utm_source=chatgpt.com parameter and referrer information. Without either signal, Google Analytics and similar tools cannot attribute the visit to ChatGPT and usually place it in Direct traffic. The report connected the timing with the broad GPT-6 rollout, but OpenAI has not confirmed that the model release caused an intentional tracking change.
Cedric Huesler at Adobe published the clearest early evidence. He reviewed about 500 pages where ChatGPT had previously supplied more than five per cent of visits, comparing 9 to 13 October with the prior week. Tracked ChatGPT traffic fell by roughly 31,000 visits a day while Direct rose by about 26,000 on the same pages. He also found similar behaviour among the new Direct visitors.
The evidence is useful but limited. Huesler described four days of sampled data and deliberately selected pages with a large ChatGPT share, so the figures are not a platform-wide traffic estimate. Sébastien Billard separately reported falling human ChatGPT referrals from 9 October, but that is another observation rather than confirmation from OpenAI.
OpenAI's own publisher FAQ still states that ChatGPT automatically includes utm_source=chatgpt.com in referral URLs. That conflicts with the reported behaviour. OpenAI's GPT-6 launch confirms the rollout date and scale, but it does not mention referral handling. The safest description is an apparent measurement change whose coverage, intent and permanence are unknown.
Why it matters
This can make a healthy discovery channel look as if it collapsed overnight. A dashboard that counts only sessions tagged chatgpt.com will show a sharp decline even when many visitors have merely moved into Direct. That distorts channel reporting, content investment decisions and any attempt to compare AI visibility with visits or revenue.
The effect is larger for enterprise sites with deep informational pages. Those pages often receive little genuine typed or bookmarked traffic, so a sudden Direct increase may be the only visible clue. However, Direct is a catch-all bucket. It also includes copied links, privacy restrictions, apps, email and other untagged sources, which means the missing ChatGPT visits cannot be reconstructed accurately at user level.
This also exposes a weakness in GEO measurement. Referral attribution depends on the platform preserving a signal. Brands do not control ChatGPT's outbound-link behaviour, and a platform update can break a trend line without changing citations, recommendations or customer interest. Traffic data is still useful, but it needs to sit beside first-party citation evidence, landing-page patterns and commercial outcomes.
The documentation conflict matters operationally. If OpenAI restores the tag, teams may see an apparent recovery that is another tracking change rather than real growth. Any report covering October should separate observed traffic from attribution quality and make that caveat visible to senior stakeholders.
How your brand can benefit / be affected
Annotate analytics and reporting from 8 to 10 October as a possible ChatGPT attribution break. Preserve the date by device, browser and landing page so the baseline can be revisited if OpenAI explains or reverses the behaviour. Do not rewrite historical data or allocate all new Direct traffic to ChatGPT.
Build a blended diagnostic view for pages that previously received meaningful ChatGPT referrals. Show tracked ChatGPT sessions, Direct sessions, new-user rates and downstream conversions together. A matching fall and rise can support an attribution hypothesis, but label it as estimated hidden traffic rather than measured ChatGPT traffic.
Run controlled checks across logged-in and logged-out sessions, desktop and mobile, and different answer types. Record whether the destination contains the ChatGPT UTM parameter and whether the browser sends a referrer. Repeat the check because a staged rollout or interface experiment can produce inconsistent results. Keep screenshots and timestamps so later comparisons are auditable.
Finally, avoid using referral sessions as the sole GEO success measure. Combine Bing or other first-party citation reporting, server-side landing-page activity, assisted conversions and customer research. Watch OpenAI's publisher documentation for a correction or clarification. Until the platform confirms the behaviour, the practical response is to protect trend reporting and communicate uncertainty, not to claim that ChatGPT demand disappeared.
News date: 14 October 2026. Editorial review: 15 October 2026.