ChatGPT Stopped Telling Your Analytics Where Visitors Come From
On October 9, 2026, ChatGPT links lost their utm_source=chatgpt.com tag. In our own data the tag went from 99.9% of cited links to under 1% within three days. Update, October 17: the tag is coming back, on about half of cited links on October 16 and rising.
Update, October 17, 2026: the tag is coming back. On October 16, 51.9% of the cited links in the ChatGPT answers we collected carried utm_source=chatgpt.com again, rising through the day from 5.8% in our first run at 05:00 UTC to 61.8% by 09:00. A spot check of the same prompts through two different collectors that morning showed the same thing. It is not fully back: before October 9 the tag was on 99.9% of cited links, and product and image cards are still untagged. We have not seen any statement from OpenAI on the cause, and our data cannot show whether links still open without a referrer. Both charts below now include October 16.
If your GA4 shows ChatGPT traffic falling off a cliff on October 9, 2026, you are not alone, and the cause is probably not your visibility in ChatGPT.
Starting that day, most links in ChatGPT answers stopped carrying the utm_source=chatgpt.com tag that OpenAI had been adding to cited links. ChatGPT was also reported to open links with noreferrer, which strips the other signal analytics tools rely on. With neither the tag nor the referrer, GA4 and other analytics tools record the visit as Direct.
We see it in our clients' analytics, and we see it in the ChatGPT answers we collect every day.
- Update, October 17: the tag is returning. It was on 51.9% of cited links on October 16, up from under 1%, but not yet back to the 99.9% it was on before October 9.
- What changed: Since October 9, 2026, almost no links in ChatGPT answers carry the
utm_source=chatgpt.comtag, and ChatGPT opens them withnoreferrer, so the destination site cannot tell the visit came from ChatGPT. - Our data: In a sample check of the ChatGPT answers we track, the tag was on 99.9% of cited links from October 2 to 8 (30,919 of 30,953). From October 11 to 15 it was on 0.75% (317 of 42,325).
- In GA4: ChatGPT-attributed users drop sharply from October 9 on the client properties we looked at. The visits have not necessarily disappeared. Without a tag or a referrer, they are recorded as Direct.
- Intentional or not: OpenAI has not said. It may be a side effect of a new interface, or one more step that makes AI traffic harder to attribute. The effect on your reports is the same either way.
- What to do: Annotate October 9 to 16 in GA4, stop comparing ChatGPT referral numbers across those dates, and measure ChatGPT through visibility and citations rather than referral sessions alone.
What Changed on October 9
Two separate things changed in how ChatGPT sends people to websites.
The tag disappeared: ChatGPT used to append ?utm_source=chatgpt.com to the links it cited. It was the most reliable way to identify ChatGPT visits, because the tag travels with the URL whatever the browser does with the referrer. Most links shown directly in answers no longer carry it.
The referrer disappeared too: as Search Engine Roundtable reported, ChatGPT now opens links with window.open(url, "_blank", "noopener,noreferrer"). The noreferrer instruction tells the browser not to send the page the user came from. So even a site that ignores UTMs and reads the referrer sees nothing.
It happened almost overnight. On October 8, every ChatGPT answer we collected with citations had tagged links: 1,954 of 1,954. On October 9 it was 28 of 138. Since October 11 it has been between 0% and 3% of answers a day, and on October 15 it was none of 1,120.
What Our Data Shows
We collect ChatGPT answers every day for the brands we track, and we keep every link exactly as ChatGPT cited it. That makes the change easy to measure. For this post we ran a sample check over the answers collected from September 25 to October 15, counting the cited links that still carry utm_source=chatgpt.com.
The drop is not gradual in our panel. On October 8 every one of the 5,958 links we collected carried the tag. On October 9 it was 66 of 965. From October 11 to 15 it stayed below 1.5% every day. On October 16 it started coming back.
Some ChatGPT visits still arrived attributed in GA4 during the gap, so not every visitor or every link changed. But in the answers we collect, the tag was effectively gone for a week.
It was not limited to one kind of link, either. We split every link by where it appears in the answer: each type lost the tag, and only some of them have it back:
During the gap, the few tagged links that remained were spread across many unrelated sites, and no link type kept the tag. The return so far covers the source list and the links in the answer text. Product and image cards are still untagged.
What It Looks Like in GA4
On the client properties we looked at, new users with a first-user source of chatgpt.com fall on October 9 and keep falling over the following days. On the larger one, ChatGPT-attributed new users were 3,260 on October 9 and kept sliding, after three months of a steady weekly pattern.
The visits themselves show up somewhere else. Search Engine Roundtable cites one analysis that measured 31,000 fewer ChatGPT visits a day and 26,000 more Direct visits a day after the change, with Direct rising on two out of three pages that lost ChatGPT traffic. In that analysis, most of the "lost" ChatGPT traffic reappeared as Direct.
Pro tipCheck your own site in two minutes. In GA4, compare October 1–8 with October 9–15 for two rows: chatgpt.com as source, and Direct. If ChatGPT fell and Direct rose by a similar amount, mostly on deep pages rather than your homepage, you are looking at the attribution change, not a loss of visitors.
Intentional or Not, the Effect Is the Same
We don't know whether OpenAI meant to do this. We have not seen any announcement from them. It could be a side effect of the new interface: links rendered by different code that never got the tagging logic. It could be a privacy decision. Or it could be one more step in a pattern we keep seeing across AI platforms: show publishers that their content is being used, and make it harder to measure what they get back.
We wrote about the same pattern when Google added AI reports to Search Console with impressions but no clicks. Different company, same outcome for the site owner: less visibility into the one number that ties AI answers to business results.
The tag is now coming back, as the update at the top of this post shows. Until it is fully back and stays back, plan as if it could go again.
How to Report on ChatGPT Traffic Now
Annotate October 9 to 16: add an annotation in GA4 and in any dashboard that shows ChatGPT or AI-assistant traffic. Anyone reading a month-over-month or year-over-year comparison that includes that week needs to know it is not like-for-like.
Stop treating ChatGPT referrals as the full number: referral traffic from chatgpt.com is now a floor, not a measure. Some visits still arrive attributed, but you cannot tell what share of the real total they are.
Watch Direct on deep pages: people rarely type a long article URL from memory. A rise in Direct sessions landing on deep content pages, starting October 9, is the best available signal of where ChatGPT traffic went.
Measure ChatGPT upstream: referral data was always an incomplete picture of AI visibility. It only counted people who clicked. How often ChatGPT names your brand, which of your pages it cites, and for which questions, does not depend on a tag or a referrer, and you can still measure all of it.
Why This Matters for Our Clients
For our clients this changes the reporting, not the strategy. We track ChatGPT visibility and citations by collecting the answers themselves, not by counting referrals, so that measure does not depend on a tag or a referrer.
What does change is how to read the traffic line. Any ChatGPT traffic chart that covers October 9 to 16 needs that week marked on it. While some links still go untagged, ChatGPT referrals are the minimum ChatGPT is sending you, not the total, and Direct traffic to deep pages belongs next to them so the shift is visible rather than mistaken for a loss.
Google's AI reports in Search Console show impressions but no clicks. For a week, ChatGPT dropped the tag from its links, and it is only partly back. Each change makes it harder to see what AI answers actually send you. The brands that keep a clear view will be the ones that measure their presence in the answer itself.
Want to see this running on your brand?
Book a demo and see how our systems turn into compounding organic growth.
