The Great ChatGPT Traffic Mystery: How a Glitch Slashed AI Referrals and Sent Analytics Marketers into a Panic

Digital marketers and SEO professionals woke up in mid-October to a startling metric drop: Google Analytics 4 (GA4) dashboards were showing a precipitous, nearly 50% decline in referral traffic originating from ChatGPT. For websites heavily reliant on the burgeoning AI Assistant search channel, the sudden slump threatened to upend traffic forecasts and misrepresent user acquisition strategies.

However, as industry experts, data platforms, and analytics executives quickly uncovered, the vanishing traffic was a phantom. The users had not stopped clicking; rather, they had simply become invisible. Driven by the rollout of OpenAI’s GPT-6 architecture and a subsequent tracking glitch, links generated within ChatGPT answers were suddenly stripped of their vital referral data and utm_source parameters, causing thousands of legitimate AI-driven visits to be miscategorized as "Direct" traffic in analytics packages.

Main Facts: The Anatomy of the Tracking Disconnect

The disruption began to unfold in early October, casting light on the fragile nature of attribution modeling in the age of generative artificial intelligence.

At the core of the issue was a technical shift in how ChatGPT handled outbound links. Following system updates associated with the GPT-6 rollout, inline links embedded within conversational answer text began opening in new browser tabs utilizing settings that actively withheld referrer data. Furthermore, these links omitted the crucial utm_source=chatgpt.com parameter that had previously allowed analytics platforms like GA4 to accurately bucket visitors into the AI Assistant channel.

Key developments defining the incident include:

• The Scale of the Drop: Platforms tracking AI-driven web traffic—such as Graphite, LLM Pulse, and Adobe—independently confirmed drops ranging between 22% and 46% in ChatGPT-attributed GA4 sessions starting around October 9.
• The "Direct" Traffic Migration: Detailed comparative analyses revealed an inverse relationship: as ChatGPT-attributed sessions plummeted, Direct traffic surged on the exact same landing pages.
• The Culprit: Industry leaders eventually pinned the anomaly on a software bug within OpenAI’s linking mechanism, which selectively dropped tracking tags on inline citations while retaining them in dedicated Sources panels and shopping modules.
• The Resolution: OpenAI quietly began deploying a fix on October 15, restoring tagged clicks for many domains, though debate remains over the completeness and speed of the rollout.

Chronology: A Week of Confusion in the Analytics Community

The unfolding of the ChatGPT attribution crisis provides a fascinating case study in how rapidly modern digital ecosystems react to underlying infrastructure changes.

October 2 – October 7: The Baseline

Prior to the disruption, AI search visibility tools operated with a predictable baseline. Tomek Rudzki of Peec AI, running a standardized set of 1 million prompts, noted that direct URLs appeared in roughly 8% to 10% of ChatGPT responses. Concurrently, when OpenAI introduced its GPT-6 architecture—rolling it out to paid tiers on October 7 and to Free and Go users by October 8—initial traffic metrics remained stable. LLM Pulse noted no immediate drops on October 7.

October 8 – October 9: The Cliff

The inflection point occurred on October 8 and 9. According to data collected by AI visibility platform Profound, the share of ChatGPT answers featuring at least one tagged source link plummeted from roughly 98% on October 8 down to 66% by October 14. Simultaneously, the usage of "Intelligent UI"—ChatGPT’s interactive format for charts, comparisons, and small tools—surged from 2% to 35% of collected answers.

Graphite CEO Ethan Smith was among the first to sound the alarm on LinkedIn, noting a 22% to 57% drop in ChatGPT-attributed sessions in GA4, translating to an estimated 45% fall from baseline traffic.

October 9 – October 13: Correlating the Shift

As analytics professionals scrambled to diagnose the drop, Cedric Huesler, senior director of product management at Adobe, analyzed roughly 500 pages where ChatGPT historically drove at least 5% of visits. Comparing October 9–13 with the preceding week, Huesler discovered that daily ChatGPT visits dropped by approximately 31,000, while Direct visits rose by about 26,000.

During this same window, Tomek Rudzki re-ran his set of 1 million prompts, discovering that ChatGPT was now outputting a direct URL for a webpage in a staggering 42% of responses—with 54% of those URLs pointing directly to homepages. However, these links completely lacked tracking tags or referrers.

October 14 – October 15: Bug Confirmation and Remediation

The narrative shifted dramatically on October 14. After days of speculation, Huesler reported that OpenAI had formally confirmed to him that the behavior was the result of a bug.

By the early morning hours of October 15, a fix began rolling out. Huesler noted in a follow-up post that by 9:00 a.m. PT, tagged ChatGPT clicks across thousands of monitored domains had rebounded to roughly 95% to 100% of the volume seen during the corresponding hour the previous week.

Supporting Data: What the Metrics Reveal

While the qualitative reports of missing tags were clear, quantifying the shift required deep dives into platform-specific telemetry. The data paints a picture of a massive redirection of attribution rather than a true loss of user engagement.

The Graphite Findings

Ethan Smith’s tracking of meaningful traffic in the AI Assistant channel revealed that mobile users bore the brunt of the impact. Graphite’s data indicated that a remarkable 95% of mobile ChatGPT visits to their tracked sites relied strictly on the UTM tag for attribution, making mobile app traffic uniquely vulnerable when the tagging mechanism failed.

Adobe’s Behavioral Analysis

Cedric Huesler’s analysis of high-ChatGPT-referral pages provided behavioral proof that the "new" Direct traffic was actually disguised AI traffic. On mobile devices, the share of Direct visitors who encountered a site consent banner—a reliable proxy for a first-time visitor—climbed from 36% to 44%, edging closer to the 55% baseline typically observed among verified ChatGPT visitors. Furthermore, Direct traffic rose on two out of every three pages that suffered a drop in recorded ChatGPT traffic.

Profound’s UI Shift Insights

Profound linked the tracking failures directly to changes in OpenAI’s rendering formats. As Intelligent UI formats scaled up, standard inline text links lost their parameters. However, links accessed via secondary product or information panels successfully retained their UTM tags, explaining why some referral data trickled through uninterrupted.

Official Responses and Industry Skepticism

Despite widespread consensus among data scientists and marketing platforms regarding the nature of the bug, official communication from OpenAI remained sparse.

• OpenAI’s Silence: Notably, OpenAI’s official release notes for the mid-October period did not mention link tracking adjustments or the deployment of a bug fix. The confirmation of the bug came through informal channels to enterprise partners rather than a public developer bulletin.
• Mixed Industry Observations: While Adobe and Graphite reported rapid rebounds following the October 15 rollout, skepticism remained in the broader SEO community. Veteran search consultant Glenn Gabe noted that he had not yet observed the anticipated recovery across all managed accounts. Furthermore, tracking platforms like Profound reported no official documentation confirming a comprehensive fix.
• The Competing Approach from Google Gemini: The incident highlighted a divergence in how major AI platforms handle referral attribution. Just weeks prior to the ChatGPT glitch, Google Gemini made headlines by proactively adding robust UTM parameters to its outbound citations, as documented by search analyst Roger Montti, ensuring cleaner integration with GA4 and other analytics tools.

Implications for Digital Marketers and SEO Strategy

The October 2025 ChatGPT attribution scare serves as a cautionary tale for digital marketers navigating the shift from traditional search engines to generative AI discovery channels.

1. Guard Against False Panics in GA4

A sudden 50% drop in referral traffic from a major channel in less than a week is instinctively alarming. However, as Smith and Huesler advised, marketers must look beyond surface-level dashboard metrics. Correlating drops in AI traffic with reciprocal spikes in Direct traffic—particularly on pages heavily cited by LLMs—should be the first diagnostic step. Marketers are strongly advised to manually annotate October 8 through October 15 in their analytics historical logs to prevent executive misinterpretation.

2. The Vulnerability of UTM-Dependent Channels

The incident underscored the inherent fragility of relying solely on UTM parameters for channel attribution. Because browsers increasingly strip or obscure HTTP referrers for privacy reasons, platforms that fail to hardcode tracking parameters into outbound links leave marketers blind. Until OpenAI and other LLM developers standardize robust, tamper-proof attribution, holistic measurement will require cross-referencing multiple metrics.

3. Re-evaluating the "Direct" Channel

For years, digital analysts have treated the Direct traffic bucket as a catch-all for bookmark visits, typed URLs, and untrackable app traffic. The ChatGPT bug proves that Direct traffic can be artificially inflated by systemic changes in third-party AI interfaces. Marketers should conduct granular reviews of their Direct traffic spikes, cross-referencing landing page distributions with known LLM citation patterns—paying special attention to homepage surges, as highlighted by Peec AI’s findings.

Looking Ahead

As generative search continues to mature, the relationship between AI platforms and content publishers will inevitably face growing pains. While OpenAI’s rapid deployment of a fix restored much of the lost visibility, the episode highlights the urgent need for transparent, standardized attribution frameworks across the entire AI ecosystem. Until then, vigilant monitoring, cross-channel correlation, and a healthy skepticism of raw GA4 channel groupings will remain essential tools for the modern SEO professional.