OpenAI fixes ChatGPT bug that stripped UTM tags and skewed GA4 attribution for a week

What happened

OpenAI has fixed a ChatGPT bug that stripped tracking parameters from outbound links, leaving roughly a week of referral traffic misclassified in Google Analytics 4. According to a write-up on dev.to, links displayed in ChatGPT answers frequently lacked the utm_source=chatgpt.com parameter from around October 8 through October 15, 2026, and often failed to send an HTTP referrer as well. With both attribution signals missing, GA4 sorted many of those sessions into the Direct channel, producing a misleading pattern in which reported ChatGPT referrals fell while Direct sessions climbed.

The timing matters: the dev.to post ties the behavior to OpenAI's broader GPT-6 rollout, which included an interface update called Intelligent UI. The write-up connects the issue to how the new interface opened links, including a window.open call using noopener and noreferrer, which can suppress both campaign parameters and referrer information. Search Engine Roundtable reported that OpenAI identified the behavior as a bug and began deploying a fix on October 15.

How the attribution broke

The two affected signals work differently. A UTM parameter travels inside the URL itself, so analytics software can identify a visit's source whenever the parameter survives the click. An HTTP referrer is context supplied by the browser, indicating the page or service that sent the visitor along. When neither arrives, GA4 generally falls back to Direct, subject to its channel rules and whatever other data is available.

During the affected window, both failed at once: UTM tags were frequently absent from ChatGPT links, and referrer data was often omitted. That combination meant the problem could not be worked around simply by leaning on referral reporting instead of campaign-based attribution.

What the fix does and does not repair

Tagged clicks began returning toward earlier levels in sizable samples on October 15 and 16, according to the dev.to write-up, restoring a usable attribution signal for future traffic. Two caveats remain. First, HTTP referrer data stays unreliable for many ChatGPT clicks even after the fix, so channel definitions built only around referrers may keep undercounting the assistant. Second, the fix is forward-looking: GA4 cannot retroactively attach source data to sessions that were recorded without it. A chart showing a ChatGPT traffic decline during the affected week should therefore be read as a measurement gap, not as evidence that visibility, user interest, or content performance dropped by the same amount.

What analytics teams should do now

The dev.to post suggests a three-part cleanup:

• Annotate the reporting window. Flag October 8 through 15 in GA4 dashboards, internal reports, and campaign analyses as a period when ChatGPT outbound-link attribution was degraded by missing UTM tags and unreliable referrers.
• Audit the Direct traffic spike. Compare Direct sessions, landing pages, conversions, and dates in the affected period against adjacent periods. This will not reconstruct every lost ChatGPT session, but it can show whether the Direct increase lines up with the attribution failure.
• Separate measurement from performance decisions. Avoid cutting content, shifting channel budgets, or concluding that ChatGPT visibility declined based on the anomaly alone, and resume monitoring tagged traffic now that the fix is live.

The post also recommends reviewing how analytics reports define AI-driven traffic more broadly, since a channel group that depends solely on referrer data may continue to undercount ChatGPT even after the repair.

Why it matters

ChatGPT is becoming a genuine discovery channel, and this incident shows how fragile measurement of that channel is. Site owners do not control how an AI product opens outbound links, whether it passes referrer information, or whether it appends source parameters, and a single platform-side change shipped without notice quietly redirected a week of traffic into the least informative bucket GA4 has. For anyone comparing acquisition week over week, attributing leads to channels, or reading Direct traffic as a proxy for brand demand, the lesson is to build analytics workflows that can absorb these platform-level shifts and to treat third-party referral metrics as one signal among several rather than a source of truth.