Did ChatGPT stop tagging its links with GPT-6? Our ledger: utm_source=chatgpt.com on 9 of 9 arrivals from October 7 to 11 and none since; the chatgpt.com referrer was already gone on September 27; ChatGPT-User fetches unchanged; the API still tags 249 of 249

What was reported, and by whom

On 2026-10-14 Barry Schwartz at Search Engine Roundtable wrote that OpenAI "seems to have removed or replaced the UTM parameters it placed on links within the ChatGPT results", that it "seems to have started with the release of GPT-6", and that he is "not sure if it was directly intentional". The observation he quotes is Cedric Huesler of Adobe on LinkedIn: "GPT-6 opens links with window.open(url, '_blank', 'noopener,noreferrer') and drops ?utm_source=chatgpt.com. There's no tag and no referrer, so your analytics files the visit under direct." Huesler compared about 500 pages where ChatGPT drove 5% or more of visits, 2026-10-09 to 10-13 against the week before: ChatGPT down 31,000 visits a day, direct up 26,000 a day on the same pages. A second observer quoted there, Sébastien Billard, reports ChatGPT human traffic and ChatGPT-User fetches falling since 2026-10-09 while OAI-SearchBot fetches rose. We have not read either original post; the quotes and numbers are as the article carries them.

The release is dated: OpenAI's "GPT-6 and Intelligent UI for everyone" is 2026-10-07, and the ChatGPT release notes carry entries for 2026-10-07 and 2026-10-09. Neither mentions links, parameters or referrers. With no OpenAI statement, the claim rests on what sites measure. Below is what ours measures.

ChatGPT arrivals on askedabout.com, by week and by what identified them

Our analytics stores, for every session, the URL it entered on and the referring domain the browser sent. A ChatGPT arrival is a session whose entry URL carries `utm_source=chatgpt.com` or whose entry referrer is `chatgpt.com`. The table counts sessions since 2026-07-01 by which of the two was present. Weeks start on Sunday; the last row is 2026-10-11 to 10-14 16:46 UTC. Our own test loads are excluded.

Two readings. The tag was present on every ChatGPT arrival in the GPT-6 week: the nine sessions from 2026-10-07 to 2026-10-11 (two on 10-07, three on 10-08, two on 10-09, one on 10-10, one on 10-11 at 01:40 UTC) all entered on a `utm_source=chatgpt.com` URL. The referrer went missing before GPT-6, not with it: the last session with a `chatgpt.com` referrer entered on 2026-09-26, and from 2026-09-27 on 0 of 16 tagged sessions carried one, against 11 of 21 in the three weeks before. On our ledger the "no referrer" half of the report predates GPT-6 by ten days; the "no tag" half has not shown up.

Day by day since October 1

Sessions per day by how they entered, beside the number of pages the ChatGPT-User fetcher requested from us that day (our crawler ledger, which records every bot-shaped request by path and user agent). "Direct, untagged" is a session with no referrer and no `utm_source`, which is where an untagged ChatGPT click would land. 2026-10-14 is partial, to 16:46 UTC.

The direct column looks like the Adobe pattern, 0 to 3 a day rising to 8 to 12 from 2026-10-09, but on our site it is not people. Our analyst reads every session by hand each morning: the 10-12 to 10-14 direct rows are mostly one automated actor loading `/` from six countries minutes apart (8 of 10-14's 9 sessions held) plus hosting-network addresses; real persons by hand were three on 10-12, three on 10-13 and one on 10-14 to the morning read. A site at our volume cannot carry the comparison Huesler ran on 500 pages; the column is shown so the limit is visible. The tagged column's three zero days, 10-12 to 10-14, have precedent in the same window (10-03 and 10-04, 09-18 to 09-20), so they are not evidence of a change. The fetcher column does not move: ChatGPT-User, the user agent ChatGPT sends when it reads a page for an answer, requested 24 to 49 pages a day from us before and after 2026-10-07.

The API did not change either

Our weekly four-engine run asks 12 buyer questions of ChatGPT, Perplexity, Gemini and Claude through their APIs, three samples each, and stores every cited URL. The 2026-08-07 census found ChatGPT's API citations tagged `utm_source=openai` on 252 of 252 URLs and the other three engines tagged none. The same census on the two runs either side of GPT-6:

Two different tags, two different surfaces. The API appends `utm_source=openai`; the ChatGPT app appended `utm_source=chatgpt.com` to the links people click. The report concerns the second. Tools that measure ChatGPT through the API, including ours, would see no change from this, and that absence is not evidence about the app.

What this means if you watch ChatGPT traffic

• Check the entry URL on your most recent ChatGPT sessions before concluding anything. If sessions with `utm_source=chatgpt.com` stopped on a date and direct rose on the same pages the same day, that is the Adobe pattern. If the tag is still arriving, as it was for us through 2026-10-11, the change has not reached your links, or not yet.
• Referrer-based attribution of ChatGPT was already undercounting. On our ledger no ChatGPT arrival has carried a referrer since 2026-09-27. Any report that counts ChatGPT by referring domain alone has been reading zero for weeks while tagged clicks kept arriving. The tag was the only reliable key, and it is the thing now reported missing.
• Separate what an AI visibility tool measures from what your analytics measures. A tool that queries the API sees `utm_source=openai` on every citation and nothing changed. Your analytics sees the app's links. If the app's tag is gone, the two will disagree and the tool is not wrong, it is measuring a different surface.

How this was verified

The Search Engine Roundtable article, OpenAI's GPT-6 post and the release notes were read directly on 2026-10-14; the Huesler and Billard observations are quoted as that article carries them. Our numbers come from three instruments read the same day: the PostHog session ledger (two queries, saved with the run), the Convex crawler ledger (11,634 rows 2026-09-24 to 10-14) and the API citation census script of 2026-10-11 over the 2026-10-05 and 10-12 runs; the query file and cut outputs are committed beside this post. A session is not a person; the analyst's hand reads are quoted where they differ. We did not observe the ChatGPT app's link handling ourselves and hold no ChatGPT account for it. API samples are samples, not proof of what every chat user sees. Pre-registered read, 2026-10-28: tagged ChatGPT sessions on this site 2026-10-15 to 10-28 beside the ChatGPT-User fetch count. Zero in 14 days with the fetcher at its rate means the tag is gone from our links too; one or two is not read; three or more means it persists on at least some surfaces. This post's own 14-day ChatGPT reader count is secondary: it reads zero if the tag is gone.