ChatGPT Traffic Drop Likely a Tracking Glitch

A significant drop in ChatGPT-attributed traffic within Google Analytics 4 has been identified not as a loss of user interest, but as a likely tracking glitch stemming from changes in how OpenAI formats its links. Starting on October 9, sites tracked by Graphite saw ChatGPT sessions plummet by 22% to 57%, with CEO Ethan Smith estimating the decline at approximately 45% from the previous baseline. This anomaly was corroborated by LLM Pulse, which reported that sessions from ChatGPT dropped by roughly 46% between October 9 and 11 compared to the preceding two weeks, while total site traffic remained stable.

The core issue lies in the modification of link attributes. According to Smith, ChatGPT began opening inline links in answer text without the standard utm_source=chatgpt.com tag or a referrer header. Consequently, these visits are misclassified as Direct traffic in GA4 rather than being attributed to the AI assistant. While this explains the apparent disappearance of referral data, analysts caution that there is insufficient evidence to determine if actual visitation numbers have decreased or if they have merely been mislabeled.

Migration to Direct Traffic Channels

Data from Adobe senior director of product management Cedric Huesler supports the theory that traffic is being misattribed rather than lost. Analyzing approximately 500 pages where ChatGPT previously drove at least 5% of visits, Huesler found that daily ChatGPT referrals fell by about 31,000 between October 9 and 13. In direct contrast, Direct visits surged by approximately 26,000 per day during the same period. Search traffic declined by only 7%, consistent with broader market trends, indicating that the shift was specific to the AI channel.

Huesler noted that on two-thirds of the pages that lost ChatGPT attribution, Direct traffic increased, a ratio significantly higher than the previous week’s average of half. Behavioral metrics further suggest these new Direct visitors are actually coming from ChatGPT. On mobile devices, the percentage of Direct visitors who encountered the consent banner rose from 36% to 44%, moving closer to the 55% rate typically observed among known ChatGPT visitors. These findings, based on four days of sampled data, strongly imply that the traffic source remains unchanged despite the reporting error.

Technical Shifts in Link Formatting

The tracking disruption coincides with the rollout of GPT-6. Huesler explained that after GPT-6 launched, ChatGPT began opening links in new tabs using browser settings that withhold the referrer information. Additionally, the utm_source tag, which had been present since ChatGPT’s search launch in 2024, was removed from inline links. However, links located in the Sources panel and shopping integrations retained their tags. Smith highlighted that 95% of mobile ChatGPT visits to Graphite’s sites relied exclusively on this tag, making mobile app traffic particularly vulnerable to the tracking failure.

Profound linked the change to a new answer format that prioritizes Intelligent UI elements over traditional citations. Data collected by Profound showed that the share of answers containing at least one tagged source link dropped from 98% on October 8 to 66% by October 14. During the same window, Intelligent UI components, such as charts and comparison tools, grew from 2% to 35% of all answers. Links accessed via product or information panels continued to carry tags, suggesting the issue is isolated to specific inline citation formats.

Independent Verification and User Behavior

Tomek Rudzki at Peec AI conducted an independent test using one million prompts repeated between October 2–7 and October 9–13. He observed that ChatGPT inserted direct URLs into 42% of responses, up from a previous rate of 8-10%. Furthermore, 54% of these direct URLs pointed to homepage addresses rather than specific content pages. Rudzki confirmed that these links lacked both tags and referrers, although he clarified that his data reflected displayed links rather than verified user click-through rates.

The authors of LLM Pulse noted that their aggregated data did not show a detectable increase in overall Direct traffic. They argued that because Direct traffic volume is vastly larger than ChatGPT traffic, a relatively small shift in attribution would be statistically difficult to isolate. This limitation made it impossible for them to definitively distinguish between genuinely lost traffic and traffic that was simply no longer labeled as originating from ChatGPT.

Resolution and Industry Response

Smith initially characterized the issue as a bug on OpenAI’s end, predicting a rapid resolution. By October 14, Huesler confirmed that OpenAI acknowledged the problem. A fix began rolling out early on October 15, and by 9 a.m. PT, tagged ChatGPT clicks had returned to 95% to 100% of pre-glitch levels across thousands of domains. Despite the restoration of tags, most Clicks still arrive without a referrer, leading Huesler to recommend relying on UTM parameters for accurate counting.

However, the fix has not been universal. SEO consultant Glenn Gabe reported that he had not yet seen the change reflect in paid or free accounts at the time of his assessment. Profound stated it had received no official confirmation of a fix, and OpenAI’s release notes for October 1 through 15 did not mention changes to link tracking. It is worth noting that several entities providing data, including Graphite, LLM Pulse, Profound, and Peec AI, sell SEO or AI search services, and Huesler is employed by Adobe. Sample sizes for LLM Pulse and Profound were not disclosed.

Strategic Implications for Analysts

The abrupt 50% decline in GA4 referrals serves as a critical reminder that reporting anomalies can mimic genuine shifts in user behavior. Both Smith and Huesler advise annotating the October 8 date in reports to prevent stakeholders from misinterpreting the tracking error as a loss of audience engagement. Huesler extended the relevant window to October 8–15.

Until the tracking infrastructure stabilizes, analysts should review ChatGPT and Direct traffic side-by-side for each page, particularly homepages, which Rudzki indicated receive more than half of the direct links generated by the AI. When Google Analytics introduced the AI Assistant channel in May, it was noted that sessions without a referrer would default to Direct. The ChatGPT tag was essential for capturing this traffic, especially on mobile. Smith clarified that UTM parameters set manually within ChatGPT Ads should remain unaffected by this bug. Looking ahead, the industry awaits confirmation that tags have fully restored across all account types, as OpenAI has not officially documented the fix. Meanwhile, competitors like Gemini have taken different approaches, with Roger Montti covering their recent addition of UTM parameters to links.