If your reported ChatGPT traffic dropped on October 9, this is the blog post that will disambiguate what changed, when it happened, and why.
At Profound, we’ve been investigating a change in how ChatGPT links to websites through its new Intelligent UI. Most of the links we checked directly in these answers were missing the tracking tags that tell analytics tools where a visitor came from.
Those tags are now back. By the morning of October 16, approximately 99% of the answers we collected with cited links included at least one tagged source link again.
But their return does not automatically recover the missing history. For marketers looking at traffic and conversion reports across these dates, understanding the sequence still matters.
Here is what we observed as tracking declined, fell much further, and then returned.
What actually changed
With the launch of Intelligent UI, ChatGPT began answering more questions with interactive elements, alongside its familiar text-based responses.
Many of the links in those new answers were missing a small piece of information: utm_source=chatgpt.com.
That might look like an obscure detail at the end of a URL. But it’s what tells an analytics tool, “This person came from ChatGPT.”
Imagine someone asks ChatGPT which running shoes to buy. Your brand gets recommended, they click through to your product page, and they place an order.
The website can still record the visit and purchase. But without the information identifying ChatGPT as the source, the visit may appear as Direct, and ChatGPT’s role in that purchase becomes invisible. A missing UTM does not always mean a visit becomes Direct, because a referring page can sometimes identify the source too.
Our assessment was that missing attribution was driving much of the reported referral decline. The timing and scale were consistent with what customers were reporting.
To understand the sequence, we looked at two things over time: how often ChatGPT used Intelligent UI, and how often its answers included tracking tags on cited links.
All times below are UTC, and the data is grouped into 30-minute intervals. The UTM rate refers to the share of answers with cited links that included at least one tagged source link.
October 8, 18:30 UTC: tracking began to fall
Before the decline, approximately 99% of the answers we collected with cited links included a tracking tag.
In the interval beginning at 18:00 UTC on October 8, that rate was approximately 98.5%. At 18:30 UTC, it fell to approximately 97%, beginning a more sustained decline.
The first change was relatively small. Over the following hours, it became much more visible.
As more answers used Intelligent UI, more answers appeared without attribution on their cited links.
October 9: roughly a quarter of answers had no tagged source link
Across October 9, approximately 75% of the answers we collected with cited links included at least one tagged source link.
Intelligent UI accounted for approximately 25% of those answers.
Nearly every answer had some identifiable attribution before the drop. By October 9, roughly a quarter had none on the cited links we collected.
For a marketer seeing reported ChatGPT referrals fall around this date, that was a meaningful connection. A similar share of answers had moved into a format where many links no longer identified ChatGPT as the source.
It did not mean that a quarter of visitors had stopped arriving. It meant that a quarter of the collected answers with cited links no longer carried the tracking information we were measuring.
October 14, before 19:00 UTC: the initial decline continued
By October 14, before the much sharper decline that evening, approximately 67% of answers with cited links still included a tagged source link.
Intelligent UI accounted for approximately 34% of those answers.
This was the pattern behind our initial assessment. We were hearing from marketing teams seeing roughly 30–40% declines in reported ChatGPT referrals, while a similar share of collected answers was appearing in the new format without attribution.
We also found exceptions.
Some links rendered after opening an additional product or information panel still carried tracking tags. A link displayed directly in an answer could be missing its tag, while a link reached further inside the interface still had one.
Attribution could depend on which link a person clicked, even within the same answer.
That mattered because it meant Intelligent UI did not necessarily remove every tagged referral. Some visits could still be identified, even when many of the links displayed directly in answers were missing attribution.
October 14, 19:00 UTC: tracking fell much further
At 18:30 UTC, approximately 68% of answers with cited links included a tracking tag.
In the next interval, beginning at 19:00 UTC, that fell to approximately 10%.
At 19:30 UTC, it was approximately 5%.
The share of answers using Intelligent UI changed just as sharply. It went from approximately 33% at 18:30 UTC to 92% at 19:00 UTC, then 96% at 19:30 UTC.
The relationship we had observed during the initial rollout became much more pronounced. As the new format appeared in nearly all of the answers we collected, tagged source links became much less common.
For roughly the next 27 hours, the UTM rate was generally around 5–12%, with some fluctuations.
The reporting gap had become substantially larger than the decline marketers first noticed around October 9.
October 15: tags began returning in some checks
On October 15, public reports began appearing that ChatGPT was tagging its links again. Search Engine Roundtable updated its coverage to reflect those observations.
But our collected responses had not yet returned to their earlier level of tracking.
At 22:00 UTC on October 15, approximately 11% of answers with cited links included a tagged source link. At 22:30 UTC, that rose to approximately 16%.
The data was beginning to improve, but the recovery was still partial.
Finding a tagged link showed that attribution was available in that instance. We still needed to look across collected answers to understand how consistently it had returned.
October 16, 02:30 UTC: the recovery became much larger
At 02:00 UTC on October 16, approximately 25% of answers with cited links included a tracking tag.
At 02:30 UTC, that rose to approximately 51%. By 03:00 UTC, it had reached approximately 82%.
The rate continued to fluctuate over the following hours. At 05:00 UTC, it was approximately 58%.
Tracking was returning, but it had not yet settled at the level we had seen before the decline.
October 16, 06:30 UTC: tracking returned to approximately 99%
At 06:00 UTC, the UTM rate reached approximately 93%.
By 06:30 UTC, approximately 99% of answers with cited links included at least one tagged source link again. It remained around that level in subsequent collected intervals through 16:30 UTC, the latest interval in this analysis.
The tags had also returned inside Intelligent UI itself.
At 06:30 UTC, approximately 96% of Intelligent UI answers with cited links included a tagged source link. The format was still appearing in answers, and those answers were now carrying attribution much more consistently.
The recovery was therefore more than a change in how often ChatGPT used the new format. Tracking had returned within the format too.
What this means for your reporting
The return of tracking tags improves reporting for new visits. It does not automatically recover source information that was missing from earlier visits.
That leaves a period between October 8 and October 16 where reported ChatGPT referrals need to be interpreted carefully.
A decline during that period may reflect missing attribution. An increase as the tags returned may reflect restored attribution. Neither change, on its own, tells you how much actual visitor demand changed.
The percentages above describe tracking in collected answers. They do not provide a direct multiplier for correcting your traffic or revenue reports. People click different links, and attribution could vary depending on where those links appeared.
This affected Profound’s UTM-based human referral reporting too. Every analytics provider relying on the missing tag faced the same limitation.
Why visibility and citations still matter
With AI search, more of the discovery process happens inside the answer itself: people compare products, learn about companies, and narrow their choices before visiting a website.
During this period, even some of the visits that did happen became harder to identify.
Visibility tells you whether your brand appears in answers and recommendations, and how often competitors appear alongside or ahead of you. Citations tell you which websites and pages the answer engine chooses to reference.
Together, they help answer questions marketers still need to ask: are we becoming easier to find? Is ChatGPT using our content? Which sources are shaping the answers about our category?
Those measures don’t depend on a tracking tag arriving with a visitor. They don’t replace revenue attribution, but they become incrementally more important when traffic reporting gives you a less complete picture.
What I would do now
Annotate the changes in GA4 and any other reports used to evaluate ChatGPT performance. Use the dates and times above, converted into the timezone of your reporting.
Explain the period to anyone comparing performance before, during, and after the missing tags. Look at overall visits and conversions alongside your visibility and citation trends.
If the ChatGPT referral number was the only thing that suddenly fell, I would not use that alone as a reason to change your strategy. I would apply the same caution to a sudden increase as tracking returns.
Keep looking at whether your brand is being recommended, whether your content is being cited, and where you have opportunities to improve.
