For about a week in October, ChatGPT stopped tagging its links, so AI visits quietly landed in your "direct" bucket. The tags are back, and nothing real changed.
A week when ChatGPT visits hid as direct traffic
For roughly a week in mid-October 2026, the links inside ChatGPT's answers largely stopped carrying the `utm_source=chatgpt.com` tag that normally marks them, and they opened without passing a referrer. With neither a tag nor a referrer, analytics tools had nothing to name the source, so they filed those ChatGPT visits as "direct" rather than as AI referrals. On a dashboard the effect looked dramatic: ChatGPT traffic appeared to slump while direct traffic climbed, even where the real number of visits had not moved.
The wobble lined up with the rollout of OpenAI's GPT-6 and its new "Intelligent UI" way of rendering answer links. Search Engine Roundtable flagged the missing parameters on 14 October 2026. One tracker, Profound, put numbers to it: the share of ChatGPT answers whose cited links still carried a tag fell from about 99% to roughly 67% by 14 October, then as low as 5 to 11% across 14 and 15 October. Adobe's Cedric Huesler shared data suggesting that across about 500 pages where ChatGPT drove at least 5% of visits, ChatGPT referrals fell by around 31,000 a day while direct visits rose by around 26,000, almost mirror images.
OpenAI never announced the change, and it is not in ChatGPT's release notes. Huesler later said OpenAI had told him privately that it was a bug and that a fix was rolling out, and by about 16 October the tags were back to roughly 99%. Read all of that as unconfirmed. This is a pattern independent trackers observed and a secondhand explanation, not something OpenAI has stated publicly, a point the trade coverage of the episode (Search Engine Land, which is owned by Semrush) also stressed. The UTM tag was reportedly added by OpenAI around June 2026, per trade press, though that origin is not something the primary record here confirms.
Read the trend lines together before you worry
For any firm that watches where its visitors come from, the lesson is about measurement, not panic. During the window, a drop in "ChatGPT" traffic in your analytics did not mean buyers had stopped arriving from ChatGPT. The visits were simply misfiled as direct. So when AI referral traffic appears to fall, look at referral, direct and landing-page trends together before concluding that anything real happened. A matching rise in direct traffic, concentrated on the same pages, is the tell: on two in three of the pages that lost ChatGPT traffic, direct rose to compensate, where the week before it had been closer to a coin flip.
This is why defensible AI metrics matter more than a single alarming line on a chart. AI referral numbers are young and noisy, and a one-day cliff with an equal-and-opposite rise somewhere else is usually a plumbing problem, not lost business.
One detail outlasts the fix. Even after the tags returned, most ChatGPT clicks still arrived with no referrer, so the UTM tag remains the signal your analytics can actually count. If you strip or rewrite query strings at the edge, keep `utm_source=chatgpt.com` intact so these visits keep their label. The same discipline helps across assistants, whether you are tracking ChatGPT brand visibility or AI-citation reporting elsewhere. None of this needs a project; the work of getting found in AI answers is simply better judged over weeks than over a single surprising day.
Frequently asked questions
What happened to ChatGPT's UTM tracking in October 2026?
For about a week in mid-October 2026, links inside ChatGPT's answers largely stopped carrying the utm_source=chatgpt.com tag that identifies them, and they opened without passing a referrer. Analytics tools then logged those ChatGPT visits as "direct" traffic instead of AI referrals, so dashboards showed ChatGPT traffic falling and direct traffic rising even where real visits were unchanged.
Why did ChatGPT referrals show up as direct traffic in analytics?
When ChatGPT dropped the utm_source=chatgpt.com tag during the October 2026 window, its links carried neither a UTM parameter nor a referrer. Analytics platforms rely on one or the other to name a traffic source, so with both missing they defaulted those ChatGPT visits to the "direct" bucket, the same place a typed-in address or an untagged bookmark lands.
Did OpenAI confirm that ChatGPT dropped its UTM tags?
No. OpenAI did not announce the October 2026 change and it does not appear in ChatGPT's release notes. The explanation that it was a bug now fixed is secondhand: Adobe's Cedric Huesler said OpenAI told him so privately. Treat the episode as unconfirmed, corroborated by several independent trackers rather than by OpenAI itself.
When did ChatGPT's UTM tags come back?
The tagging began slipping around 8 to 9 October 2026, alongside the GPT-6 rollout, and bottomed out on 14 and 15 October when the share of ChatGPT answers with a tagged link fell as low as 5 to 11%. By about 16 October 2026 the tags had returned to roughly 99%, so the measurement gap was short-lived.
What should a firm do when its ChatGPT traffic appears to drop?
When ChatGPT traffic appears to drop in your analytics, check referral, direct and landing-page trends together before assuming real visits fell. A matching rise in direct traffic on the same pages usually means visits were only misfiled, as they were in October 2026. Keep the utm_source=chatgpt.com tag intact in any URL handling so AI visits keep their label, and judge AI visibility over weeks rather than a single day.
