ChatGPT traffic disappeared from analytics: what to check
If your ChatGPT referrals vanished in October 2026, they probably became direct traffic. Here is how to confirm it in GA4 with the utm_source tag, and how to count those visits again.
The report arrives on a Monday. ChatGPT sessions are down sharply, direct traffic is up, and nobody changed anything. The first instinct is to blame a ranking shift or an engine update. Before you do, check the boring explanation: the visits may still be arriving, just under the wrong label.
According to the Promptwatch changelog, ChatGPT stopped sending a referrer on October 8, 2026. Visits that used to carry a chatgpt.com referrer began showing up as direct traffic. ChatGPT has acknowledged the bug. The changelog also notes that ChatGPT still adds utm_source=chatgpt.com to the links it generates, and that detail is what makes the visits recoverable.
This post is the check we run, in order, and what each result tells you.
Step 1: find the date of the break
Open your analytics and plot ChatGPT sessions by day for the last two months. A referrer bug looks like a cliff, not a slope. If the drop begins around October 8 and happens within a day or two, that fits. If the decline is gradual over weeks, you are looking at something else, such as fewer citations or a change in the prompts people use.
Now plot direct sessions over the same window. If direct jumps by roughly the amount ChatGPT fell, the traffic moved rather than disappeared. It won't match exactly, since direct traffic is noisy, but the shapes should mirror each other.
Step 2: look for the UTM tag in GA4
Since ChatGPT still tags its links, the tag should be visible in your data even when the referrer is missing. In GA4, build a free-form exploration, or use a custom report, with these pieces:
- Dimension: session source (or session manual source) and landing page with query string.
- Filter: the page location contains
utm_source=chatgpt.com, or the session source equalschatgpt.com. - Compare it against direct sessions for the same dates.
What you want to know is how many sessions carry the tag but sit in the Direct channel group. If you see them, you have confirmed the cause. If GA4 already reads the tag into the source field, those sessions may be classed as ChatGPT already, and your drop has another explanation.
One caution. UTM handling depends on how your analytics is configured, and tag management setups sometimes strip or overwrite parameters. If you see no tagged sessions at all, check that your redirects and consent tooling aren't dropping the query string before blaming ChatGPT.
Step 3: rule out the other causes
A referrer bug is a neat story, and neat stories are dangerous. Run these quick checks so you don't stop at the first plausible answer.
Landing pages. Which pages lost the traffic? If every page lost a similar share, a tagging issue is likely. If one cluster of pages collapsed, you probably lost citations on those pages.
Crawl access. Check that nothing changed in your robots.txt, firewall or CDN rules around the same date. Our guide to why AI referral traffic gets misattributed lists the other common leaks: privacy browsers, apps that open links without referrers, and copy-pasted URLs.
Visibility. Check whether your brand still appears in answers for your tracked prompts. Traffic can fall because mentions fell. That is a real drop, and it needs a different fix.
If the date matches October 8, direct rose by about the same amount, and tagged sessions exist inside direct, you can say with confidence it's the referrer problem. Write the date and the check into your report, so next quarter nobody wonders why the October chart looks odd.
Step 4: fix the reporting, not just the explanation
Explaining the gap in a footnote works once. The better move is to reclassify. In GA4 you can create a channel grouping rule or a custom dimension that treats utm_source=chatgpt.com as an AI referral when the referrer is empty. Keep one rule: a real referrer should always win over the UTM tag. Otherwise you risk counting a visit under two labels.
Also annotate the October 8 date in your dashboards. Historic ChatGPT numbers before and after that date aren't comparable until you've backfilled the tagged sessions.
If you need the longer treatment of AI attribution in general, see how companies attribute traffic and leads to ChatGPT, Perplexity and Gemini, and the roundup of tools for tracking AI referral traffic.
Where Promptwatch fits
The manual route works, but it asks you to rebuild the same custom report in every property and to remember the exclusion rules. The place we'd do it is Promptwatch Visitor Analytics, which collects AI-referred visits and conversions through a lightweight script, the GTM template or the WordPress plugin.
On October 16 the changelog added a setting called Include ChatGPT UTM referrers. Turn it on, and visits that arrive without a referrer are attributed to their UTM source, then added to the ChatGPT card and the trend line. Per the changelog, a real referrer always wins, so nothing is counted twice. That is the same logic as the GA4 rule above, applied for you.
The benefit is less about cleverness and more about consistency. The same toggle applies to every report people look at, so the October gap doesn't reappear in a chart someone built months ago. The changelog doesn't name a plan for this setting, so check your plan on the pricing page if you aren't sure it's available to you.
Visitor Analytics also ties AI visits to conversions. That matters, because the useful question is rarely "how many ChatGPT sessions?" and more often "did those visitors sign up?". If the answer is yes, a missing referrer was hiding your best channel.
What we wouldn't claim
We can't tell you ChatGPT has fixed the bug, because we haven't seen a source that says so. Treat the UTM method as a safeguard that is worth keeping even if the referrer returns. We also wouldn't assume every direct visit since October 8 is ChatGPT. Only the tagged ones are.