Did ChatGPT Traffic Drop? Check Your Direct Traffic

You open Google Analytics 4 (GA4) on a Monday, see ChatGPT sessions down by close to half, and your stomach does the thing. Did AI stop sending people to you? Did a model update bury your pages?

Probably neither. For about a week in October, the label on those visits fell off. The visitors showed up anyway, in a different row.

How a ChatGPT visit gets its label

Analytics tools decide where a visit came from using two signals. One is the referrer, a note the browser passes along about the page you clicked from. The other is a UTM (Urchin Tracking Module) tag, a short piece appended to the link itself. ChatGPT adds utm_source=chatgpt.com to the links it shows, and that tag is how GA4 knew to file the visit under ChatGPT.

The referrer is the weak signal. Huesler wrote that it is still missing on most ChatGPT clicks, which leaves the tag as the main thing holding the label in place (Relevant Audience). Take the tag away and the visit has nothing to say about where it came from. GA4 files it as Direct, the bucket for "no idea."

What happened between October 8 and 16

GPT-6 began reaching users on October 7 for paid plans and October 8 for Free and Go, per Search Engine Journal. Around then, analysts noticed ChatGPT sessions sliding.

The sizes of the drop differ by who measured it, and the measurements come from firms that sell SEO or AI tracking tools:

• Graphite (Ethan Smith) saw GA4 ChatGPT sessions fall between 22% and 57% on tracked sites, with an estimate near 45%.
• LLM Pulse saw about 46% fewer ChatGPT sessions on October 9 to 11 against the prior two weeks, with total traffic roughly flat.
• Adobe's data showed the Direct row rising by most of what ChatGPT lost, which is what relabeling looks like.

Profound measured the tag itself rather than traffic. Its hourly figures show the tagged share of answers dropping from roughly 99% to about 75% on October 9, then to roughly 5% in the evening of October 14, then back to about 99% by 06:30 UTC on October 16. Profound's read is that missing attribution drove much of the reported decline. It also warns that its percentages are not a multiplier for correcting your own reports.

One caveat on the timeline: Adobe's account has tagged clicks back to roughly 95–100% of normal by the morning of October 15 Pacific time, and Profound's hourly data shows the recovery completing a day later. They measured different things in different samples, so treat "fixed by October 15 or 16" as the honest range. Search Engine Journal also notes that one SEO consultant, Glenn Gabe, had not yet seen the change on his own accounts when the fix was first reported.

Check your own numbers

You don't need a tool for this. You need two rows in the same report.

1. In GA4, pull sessions by source for September 28 through October 22. Put ChatGPT and Direct side by side.
2. Look at the landing pages. If ChatGPT dropped on a page and Direct rose on that same page by a similar amount, you're looking at relabeling. Compare your homepage too, since Peec AI's data in the Search Engine Journal piece suggests many displayed links point there.
3. Add an annotation covering October 8 to 15 so nobody, including future you, reads the dip as a ranking problem.
4. Check your time zone. A property set to a time zone far ahead of Pacific will show the break starting a day later.
5. Compare sessions after October 16 with late September. If ChatGPT is back near its old level, the story ends there.

If ChatGPT dropped and Direct did not rise, the explanation is thinner. LLM Pulse's aggregated data showed no clear Direct bump, and Search Engine Journal notes that data can't separate lost visits from relabeled ones. A site with small volume can't tell either. A handful of sessions a week is too few to read a pattern into.

It also bears repeating: the sessions that landed in Direct during the window probably won't be reattributed, and no backfill has been announced. Those visits are filed. They are just filed wrong.

Reactions to skip

• Rebuilding your attribution model. Huesler's advice is not to, since the visits were relabeled rather than lost. One bad week isn't a trend.
• Rewriting pages that "lost" ChatGPT traffic. Nothing about the pages changed.
• Buying a visibility tool to explain it. A tool can tell you whether your business is mentioned in answers. It can't recover a tag that was never sent.
• Treating a recovery as proof your AI work paid off. The count returning means the label returned. It says nothing about whether the visits became customers. For that question, see our breakdown of whether AI search traffic converts better than Google.

What to take from it

The lesson is smaller than it feels. Attribution for AI traffic rests on one tag that someone else controls, and that someone changed an interface and the tag fell off. For a site that gets a modest amount of AI traffic, the right habit is a monthly look at ChatGPT and Direct together, not a daily watch on either. Our guide on how to know if AEO is working covers what a reasonable monthly check looks like, and the Answer Engine Optimization (AEO) overview explains the wider picture.

I ran a grooming salon for years, and the lesson there was the same: the number on the report is a clue, and the work is figuring out what produced it. If you'd like someone to watch the reports and flag a glitch like this before it becomes a panic, that's the job of getting found on Google and AI.

FAQ

Did ChatGPT really stop sending traffic in October 2026?

The evidence points to a labeling change, not a traffic loss. Adobe's data showed Direct traffic rising by roughly as many visits as ChatGPT lost on the same pages. Other datasets are less clear, so check your own Direct row for the same dates.

What is a UTM tag?

It's a short piece added to the end of a link, such as utm_source=chatgpt.com, that tells analytics tools which source sent the click. Without it, and without a referrer, a visit usually lands in Direct.

Was the problem fixed?

Adobe's Huesler reported tagged clicks returning to about 95–100% of normal on October 15, and Profound's measurements show about 99% tagged by the morning of October 16. OpenAI hasn't published a statement, so the confirmation is secondhand.

Can I get the lost ChatGPT visits back in my reports?

Probably not. No backfill has been announced, so the best option is an annotation covering October 8 to 15 and a note that Direct was inflated for that stretch.

Why does the referrer matter if the tag came back?

Huesler says the referrer is still missing on most ChatGPT clicks, so the tag is carrying nearly all of the attribution. If it disappears again, the labeling breaks again.

Should I worry if my ChatGPT numbers were low anyway?

A low count can't tell you much either way. With a few sessions a week, one relabeled day changes the picture a lot. Watch the trend over a month or more.

Does this affect Google or Perplexity traffic?

None of the sources I read report that. The reports concern ChatGPT links specifically.