Why ChatGPT referral traffic dropped in October 2026
ChatGPT stopped sending a referrer on October 8, so its visits look like direct traffic. Here is how to confirm it in your analytics and count those visits again.
If your analytics showed ChatGPT traffic falling off a cliff in the second week of October while your direct traffic crept up, your audience probably did not vanish. The labels on the visits changed.
According to Promptwatch's changelog, ChatGPT stopped sending a referrer on October 8, 2026. A referrer is the piece of information a browser passes along to say which site the visitor came from. Without it, most analytics tools have nothing to attribute the session to, and they file it as direct. ChatGPT has acknowledged the problem as a bug. It still adds utm_source=chatgpt.com to the links it generates, which is the thread we can pull on.
What the symptom looks like
A few patterns point to this cause rather than a real demand drop:
- The ChatGPT line in your referral report falls sharply starting around October 8, with no change in your content or rankings.
- Direct sessions rise on the same dates, often to landing pages that nobody types by hand, such as a deep blog post or a comparison page.
- Branded search and overall conversions hold steady. If real interest had dropped, those would soften too.
None of these alone is proof. Together they are a strong hint. A real traffic loss rarely shows up as a matching bump in direct sessions to deep URLs.
How to check without any new tool
You can verify this in your own analytics in about twenty minutes.
- Pick a date window. Compare October 1 to 7 against October 8 onward.
- Look at direct traffic by landing page. If deep content pages gained direct sessions after October 8, note them.
- Check your URL parameters. Search your landing page reports or raw logs for
utm_source=chatgpt.com. If those visits exist, the tag is arriving even though the referrer is not. - Check how your tool attributes sessions. Some setups credit the UTM source and show ChatGPT correctly. Others rely mostly on the referrer. Consent banners, redirects, and link-cleaning scripts can also strip query strings, which hides the tag. If you see no tagged visits at all, one of those may be removing it.
- Make a custom segment. Build a view of sessions where the source or the landing URL contains
chatgpt.com, and save it so you can watch it weekly.
Be careful about what you conclude. This check shows that tagged visits exist. It does not tell you what share of ChatGPT visits carry the tag, so treat the resulting number as a floor.
Why this matters beyond one chart
ChatGPT traffic is small for many sites but unusually valuable. The visitor has already read a synthesized answer and clicked through with intent. If you cannot see it, you cannot justify the work that earns it. Budget conversations get worse: "AI search brings us nothing" is an easy conclusion when the evidence is sitting in the direct bucket.
It also muddies your attribution for other channels. Inflated direct traffic makes everything else look like it is performing worse relative to it, and it can mislead anyone who reads direct as brand strength. Our piece on dark AI traffic and attribution explains why some of this has always hidden in direct, and the October bug made the hidden share bigger.
Annotate and keep going
Whatever you do next, add an annotation in your analytics on October 8. When someone looks at the year-end chart in January, they should not have to guess why ChatGPT dipped. Also note October 16, the date a fix on the tracking side became available, so the timeline of your data stays honest.
Do not rewrite history in reports. Present the raw numbers, mark the break, and show an adjusted line separately if you have one.
Counting the visits again with Promptwatch
The manual method works, but it leaves you stitching reports by hand every week. This is where a measurement tool earns its place. Promptwatch's Visitor Analytics, which tracks AI-referred traffic and conversions through a lightweight script, the GTM template, or the WordPress plugin, added a setting on October 16 called Include ChatGPT UTM referrers.
Here is how it behaves, per the changelog. Visits that arrive without a referrer are attributed to their UTM source. A real referrer always wins, so nothing is counted twice. With the toggle on, UTM-tagged visits are added to the ChatGPT card and trend line.
That means a ChatGPT line that dips on October 8 can be restored in the same dashboard where you already read Perplexity, Gemini, Claude, Copilot, and other AI referrers. The tool also recognizes more AI referrers than before, including DeepSeek, Kimi, Mistral, and claude.com, with Microsoft 365 Copilot visits counted as Copilot.
Keep the limits in mind. The toggle only helps for visits that carry the UTM tag. Untagged visits with no referrer will still look direct, and Promptwatch cannot invent attribution that is not in the request. It is a more accurate floor, not a perfect count.
If you want conversions next to the traffic, the same Visitor Analytics tracks them, so you can see whether the recovered ChatGPT visits actually convert. That is the number that makes the case internally.
What to do this week
- Run the five-step check above and write down what you find.
- Add the October 8 annotation.
- If you use a tool that counts AI referrals, check whether it has a setting for UTM-based attribution and switch it on.
- Compare ChatGPT, direct, and branded search over the same window before telling anyone traffic fell.
For tools that do this kind of measurement, see our comparison of AI referral traffic analytics tools and our notes on ChatGPT Search referral analytics for publishers.