Why is ChatGPT traffic showing as Direct in GA4?

A sudden fall in ChatGPT visits reported by Google Analytics 4 (GA4) can make a useful marketing channel look as though it’s stopped working. If your enquiries haven’t changed, the first question is whether fewer people arrived or your reports lost track of where they came from.

The October 2026 tracking problem could leave ChatGPT traffic showing as Direct in GA4, when links temporarily lost information that helps identify where a visit came from. Observed checks later showed the tags returning, but that doesn’t fill the gap left in earlier reports.

What happened to ChatGPT’s link tracking?

ChatGPT’s link tracking suffered a reported bug in October, and analysts observed the tags returning on 15 and 16 October. The evidence points to a temporary tracking problem, rather than a permanent removal of tags from every ChatGPT link.

Adobe’s Cedric Huesler says OpenAI confirmed a known bug to him on 14 October, with a fix beginning to roll out the next day. He reported that tagged ChatGPT clicks in Adobe’s data had returned to roughly their previous weekly level by the morning of 15 October, Pacific time.

That account comes from Huesler’s conversation with OpenAI, rather than a public OpenAI announcement. His account explains why the early reports needed updating, although other checks suggest the tags didn’t return everywhere at once.

AI Search platform Profound separately tracked missing and returning tags in cited answers. Its observations cover 8–16 October, and it found at least one tagged link in about 99% of the sampled answers containing citations by the morning of 16 October. Profound reports those times in Coordinated Universal Time (UTC), the shared time reference used across countries.

The two reports measure different things: Adobe’s account concerns tagged clicks, while Profound counted answers containing at least one tagged source link. However, neither figure tells you what share of your own visitors went missing from a report, so neither can be used as a multiplier for correcting your traffic or revenue.

Why can a ChatGPT visit appear as Direct?

A ChatGPT visit can appear as Direct when Analytics has no usable information identifying where the visit came from. A person may still reach your website and make an enquiry even though the report can’t give ChatGPT the credit.

A link can identify its source through a short tag added to the web address, such as utm_source=chatgpt.com. These are commonly called UTM tags, short for Urchin Tracking Module tags, and they help reporting tools recognise the source of a visit.

A browser may also tell your analytics where the visitor came from, using information called the referrer. A missing UTM tag alone doesn’t guarantee a Direct visit because the referrer may still identify ChatGPT; losing both removes those two ways of recognising the source.

Google’s guidance on Direct traffic describes visits without a clear referral source. The label can cover someone typing your address, but it isn’t proof that the person already knew your business or deliberately came straight to you.

Why doesn’t the fix repair the earlier reports?

The fix doesn’t repair the earlier reports because restoring tags on new links doesn’t recreate source information that was missing when an earlier visit was recorded. A later visit may be recognised correctly while the earlier one remains unclear.

That can produce a misleading comparison across the affected dates. ChatGPT visits may appear to fall during the interruption and rise as tagging returns, even when the change in real visits was smaller or followed a different pattern.

The safer approach is to keep a note beside the affected period rather than rewrite the unknown visits as ChatGPT traffic. Huesler recommends marking 8–15 October, while Profound’s observations extend into 16 October UTC; whoever reviews your analytics should match those dates to your reporting time zone and your own evidence. The dates flag a period that needs cautious interpretation, rather than an exact correction window for every account. That note helps future comparisons account for the interruption without adding guessed ChatGPT visits back into the figures.

A rise in Direct still needs an explanation

Direct traffic can also rise for reasons that have nothing to do with ChatGPT. A new email campaign, a shared link or more people returning to your website could change the total during the same week, so the timing of a rise is a clue rather than a diagnosis.

Several possibilities are worth keeping in view:

• More people type your website address or use a saved bookmark.
• Links in emails, messages, documents or QR codes bring people to the site without source tags or usable referral information.
• A redirect, app or browser setting removes information that would otherwise identify the source.
• A change to your tracking setup affects what the reports can recognise.

Those explanations need checking against your website and activity, rather than choosing whichever story best fits the chart. A privacy or consent change can also affect what gets measured, but it doesn’t mean every unmeasured visit is quietly counted as Direct. A specialist should check whether a consent or tracking change happened at the same time and assess whether it explains the reporting pattern.

A rise in Direct on the same pages that previously received ChatGPT visits gives a specialist a useful place to investigate. Comparing those pages across the affected dates can show whether the pattern is consistent with lost source information and whether another change offers a better explanation. The comparison can’t recover each historical visitor’s source, but it can help distinguish a reporting problem from a genuine fall in visits that needs a different response.

What should a reporting review establish?

A good reporting review should establish whether the apparent ChatGPT decline reflects fewer visits or mainly a change in how those visits were labelled. The person looking after your analytics should investigate that change proactively, before recommending a change in budget or activity. They should identify the affected dates and pages, compare the period before and after tags returned and check whether campaigns, website changes or tracking changes offer another explanation.

The specialist should also compare the reporting change with what happened in the business. Enquiries or sales staying broadly consistent while visits labelled as ChatGPT fell gives them a reason to investigate lost source information, without proving that the missing visits became Direct. Your role is to supply business context they can’t see in Analytics, such as campaign dates or records of genuine enquiries, when they need it to interpret the figures. If the numbers are too small to support a firm conclusion, their report should explain that and identify what to monitor next.

Google now includes an AI Assistant channel in its default channel groups, so the review should check both where Analytics says a visit came from and the broader channel it was grouped into. Visitor counts, visits and sales measure different things too, and a useful review keeps those measures separate rather than treating movement in one as proof of movement in another.

Google’s AI Overviews and AI Mode aren’t included in that AI Assistant channel because their visits sit within Organic Search. searchDecoded’s guide to Search Console’s AI report covers the separate question of whether your site appeared in Google’s AI features, which isn’t the same as measuring visits from ChatGPT.

Good reporting needs context when tracking breaks

Analytics reports turn visitor behaviour into categories you can compare over time, but those categories depend on the information the measurement system receives. A bug that interrupts that information can change the report without an equivalent change in the people visiting your website.

The October ChatGPT issue shows why reporting needs ongoing interpretation. The person responsible for your reports should record known interruptions beside the affected dates and account for them when comparing later periods. That keeps the reports useful without pretending that every period is equally complete or that missing history can always be repaired.

A mention in an AI answer, a linked source, a visit and an enquiry still answer different questions, as searchDecoded’s AI citation measurement guide explains. A good report connects those pieces where the evidence allows and makes clear where the connection remains unknown, so you can judge performance without treating a change in one label as the whole story.

searchDecoded’s AI Search services can help you assess how your business appears in AI answers and what the available evidence says about useful visits and business outcomes. The aim is to make decisions with the evidence you have, while accounting for known reporting gaps rather than filling them with assumptions.