Written by Jeremy Souffir Founder, JTS Tech Services
The short version, and the direct answer if you read nothing else. Since OpenAI rolled out GPT-6 (7 October for paid ChatGPT plans, the next day for free users), links clicked inside ChatGPT appear to reach websites without a referrer and without the utm_source tag ChatGPT had been adding. With neither, Google Analytics and most other tools have nothing to go on, so the visit is filed as direct, the same bucket as someone typing your address from memory. The early evidence says the visitors did not disappear: on a sample of about 500 pages, roughly 31,000 daily ChatGPT visits vanished from the reports while about 26,000 daily direct visits appeared on the same pages. If your ChatGPT line dropped this week, read it as a labelling change first and a demand change second. OpenAI has not commented, and nobody yet knows whether this is permanent.
What actually changed?
There are two ways an analytics tool learns that a visitor came from ChatGPT. The first is the referrer, the address of the previous page that a browser normally passes along when you click a link. The second is a tag in the link itself. In June 2025 ChatGPT started adding utm_source tags to more of its outbound links, which is why "chatgpt" became a clean, filterable source in so many reports over the past year.
According to Cedric Huesler of Adobe, whose analysis Search Engine Roundtable reported on 14 October, GPT-6 now opens links with a browser instruction that tells the browser not to send a referrer, and the tag is no longer added. That instruction (noreferrer) is a standard, documented browser feature, often used for privacy reasons. Whatever the motive, the effect on your reports is the same: a click from ChatGPT now looks exactly like a bookmark.
What is confirmed and what isn't
Confirmed: GPT-6's rollout dates, from OpenAI's own announcement, and the fact that ChatGPT added tracking tags in 2025. Reported but not confirmed by OpenAI: the missing referrer and tag, from independent analysts looking at real traffic, and timing that lines up with the GPT-6 release. Unknown: whether it was deliberate, whether it applies to every link type and every app, and whether it will be reversed. We will update this post if OpenAI says anything.
How do we know the visitors are still coming?
Huesler compared 9 to 13 October with the week before, across roughly 500 pages where ChatGPT supplied at least 5% of visits. Three findings make the "they are still coming, just unlabelled" reading the more likely one:
- The volumes roughly offset. ChatGPT visits fell by about 31,000 a day while direct visits rose by about 26,000 a day on the same pages. Search traffic moved in line with normal patterns, so this was not a general traffic slump.
- The pattern is page-specific. On two out of three pages that lost ChatGPT traffic, direct traffic went up. The week before, that was close to a coin flip.
- The new direct visitors behave like ChatGPT visitors. On mobile, the share of direct visitors seeing a cookie consent banner (a rough sign of a first-time visitor) rose from 36% to 44%. For known ChatGPT visitors it sits at 55%. Genuine direct visitors are usually returning people; these look new.
That is one analyst's sample, not a census, and a separate SEO consultant reported a sharp drop in ChatGPT human traffic starting around 9 October without yet confirming the cause. Treat the numbers as a strong signal rather than a settled fact. The useful part is that you can test the same thing on your own store in about ten minutes, and the steps are below.
Why does a reporting quirk matter to a store owner?
Because decisions get made from that line. Over the past year a lot of merchants used "ChatGPT sessions" as the scoreboard for their AI visibility work: did the product data clean-up, the feed fixes and the new collection pages bring in more AI-referred shoppers? That scoreboard just broke, mid-season, in a way that looks exactly like failure. At the same time, direct traffic picked up thousands of first-time visitors that it will happily credit to brand awareness, email or nothing at all. Neither number is now telling you the truth on its own.
The timing makes it worse. Many teams are setting holiday budgets this month. A chart showing AI traffic falling by a third in a week is exactly the kind of evidence that gets a project cut, and here it would be cut for the wrong reason.
The two conclusions that both get this wrong
ChatGPT stopped sending us shoppers, so pause the AI visibility work
The drop lines up with a software release, not with anything you or your customers did. If ChatGPT had actually stopped recommending you, you would expect total visits to those pages to fall. In the sample above, most of the loss reappeared as direct visits on the very same pages. Check your own direct traffic to the pages ChatGPT used to send people to before drawing any conclusion about demand.
Direct is up, so our brand is growing
This is the quieter mistake and it lasts longer. If direct traffic jumped in the second week of October and your landing pages for direct now include a lot of deep product and blog URLs, those are probably not people typing your address. Crediting them to brand, email or offline campaigns will flatter those channels for as long as the change lasts, and it will make AI look less valuable in every budget conversation that follows.
What should you check this week?
- Annotate your analytics. Add a note on 7 to 9 October saying ChatGPT referral tagging appears to have stopped with GPT-6. It takes one minute and saves everyone who opens the report later from drawing the wrong conclusion.
- Look at direct traffic by landing page. People who type your address land on the homepage. A rise in direct sessions that start on specific product, collection or article pages, especially ones ChatGPT used to send traffic to, is the clearest sign of relabelled AI visits. We wrote about why AI shoppers land deep in August.
- Compare the pages that lost ChatGPT traffic with the pages that gained direct traffic. If they are largely the same pages, you have your answer for your own store, which is better evidence than anyone's sample.
- Add a second source of truth that you control. A one-question "where did you first hear about us?" step after checkout costs nothing and is not affected by how any AI app opens links. Your server or CDN logs also show when OpenAI's crawlers fetch your pages, which tells you whether you are still being read even when the clicks are unlabelled.
- Stop reporting AI traffic as a single source line. Report it as a range built from several signals: tagged referrals where they still exist, the direct-to-deep-page estimate, survey answers and crawler activity. A range is less satisfying than a number, and it is far harder for one vendor's update to break.
The genuinely encouraging part
Nothing about the shoppers changed. If the early analysis holds, ChatGPT is still sending people to the pages it sent them to last month, and the work that earned those clicks is still earning them. What broke is the label, and a label can be replaced with better measurement that you own. Stores that build that measurement now will also be ready for the next assistant that sends traffic with no name attached, which on current trends is a matter of when, not if.
Where we fit
This is a leadership problem disguised as a tracking problem. Somebody has to own the question "is AI actually bringing us customers?", decide what evidence counts, and stop a misleading chart from cancelling a project that is working. That is the role our Fractional Head of AI & Digital engagement fills: a few focused days a month of senior AI and digital leadership that sets up AI-channel reporting your team can trust, built from signals you control rather than a tag a vendor can drop overnight, and then uses it to make the budget and build-versus-buy calls. Retaining us means the next time an AI platform changes how it sends traffic, you hear about it from your own dashboard and a person who already knows what to do, not from a quarterly review that has already moved the money. If you want a second opinion on what your own numbers are saying this month, that is a good place to start.
Sources
- Search Engine Roundtable: OpenAI Drops UTM Parameters From ChatGPT Results (GPT-6) (14 October 2026; reports Cedric Huesler's analysis of about 500 pages and Sébastien Billard's observations)
- OpenAI: GPT-6 for everyone (7 October 2026; rollout dates by plan)
- Search Engine Roundtable: OpenAI's ChatGPT Updates To Improve Tracking In Analytics (June 2025; when ChatGPT began adding utm_source tags to more links)
- MDN Web Docs: Window: open() method (how the noopener and noreferrer options stop the browser passing a referrer)
- Google Analytics Help: Traffic-source dimensions (how sessions with no referrer or campaign data are attributed)
- JTS Tech: Roughly half of AI-referred sessions land straight on a product page and what ChatGPT actually reads from your site (earlier context)
