Two separate reporting disruptions this month show why cross-referencing data matters more than treating any single dashboard as authoritative. Between mid-September and early October, a significant proportion of automated Google result checks returned incomplete data, and ChatGPT referral attribution dropped sharply after inline links stopped carrying their tracking parameter. Both affect how you read performance reports, and neither reflects an actual change in traffic or rankings.
According to research published by rank-tracking firm Nozzle, 8% of Google results pages checked between 13 September and 9 October were missing most of the sites that would normally appear for the tracked keyword, compared with 0.17% over the preceding three weeks. The share peaked at around a quarter of pages between 27 and 29 September, then settled at 5% to 11% daily from 1 October onwards. Nozzle collected the affected pages from 17.8 million automated searches and attributes the pattern to Google's response to AI-driven scraping, though Google has not issued a statement on the matter.
For keywords where a client site would typically rank, that site failed to appear in the top ten results on 58.6% of the affected pages, against 2.7% under normal conditions. The disruption has eased since early October but has not returned to the August baseline.
What this means for reported ranking drops
If a keyword shows a sudden decline in late September or early October in your rank tracker, compare it against Search Console impressions and clicks for the same query and page. Where Search Console shows a similar drop, the decline is more likely to be genuine. Where it does not, you may be looking at a data collection artefact rather than a true loss of visibility.
The disruption also underlines a structural limitation in how rank-tracking tools operate. They simulate searches at intervals and record the results returned. When Google's anti-scraping measures block or distort those simulated searches, the tool records an incomplete or altered page as though it were the real result a user would see. Search Console, by contrast, logs actual impressions served to real users, which makes it a more reliable cross-check during periods when automated collection is compromised.
ChatGPT referral attribution fell 45% after GPT-6 launched
Separately, ChatGPT-attributed sessions in Google Analytics 4 fell sharply this month. Analysts link the drop to inline links in ChatGPT answers that stopped carrying the utm_source=chatgpt.com tag after GPT-6 rolled out.
Graphite CEO Ethan Smith reported an approximate 45% drop in ChatGPT sessions across sites tracked by Graphite with meaningful AI Assistant Channel traffic in GA4. The untagged visit data is likely to appear under the Direct label instead. Adobe's Cedric Huesler added that OpenAI confirmed a known bug to him on 14 October and began deploying a fix the following day. By around 9am Pacific Time on 15 October, tagged ChatGPT clicks had returned to roughly 95% to 100% of the same hour from the previous week across thousands of websites Huesler tracks. SEO consultant Glenn Gabe noted he had not yet seen the change, and OpenAI's release notes, while updated on 15 October, did not reference the bug or solution.
Huesler places the affected window between 8 and 15 October. He and Smith both recommend noting it in reports so the dip is not mistaken for lost traffic. Since most ChatGPT clicks still arrive without a referrer, Huesler suggests tracking them by their UTM tag rather than relying solely on referrer strings.
Why both disruptions share the same root problem
Both incidents expose the same structural weakness: relying on a single method to measure performance. Rank trackers simulate searches and record what they see. Google Analytics records visits with the attribution tags attached to them. When the simulation is blocked or the tag is missing, the tool reports what it can see, not what is actually happening.
The practical fix is the same in both cases. Cross-reference rank-tracker data with Search Console impressions and clicks for the same query and page. Cross-reference ChatGPT attribution with Direct traffic patterns, particularly if you have seen a recent increase in Direct sessions that coincides with a drop in ChatGPT-attributed ones. Where the two sources agree, the signal is stronger. Where they diverge, investigate before concluding anything has changed.
A wider pattern across AI search and reporting platforms
The same principle applies to other reporting limitations that have become visible recently. Google is moving inline links higher in AI Overviews and AI Mode, but in Search Console each link is given the same position as the overall AI Overview, so moving an inline link higher does not by itself change its reported position. Bing's new Topics Portfolio feature in the AI Performance report groups queries your site is cited for into broader topics and lists opportunity queries where other sites appear and yours does not, but Microsoft describes the metrics as directional rather than absolute, which makes them better for comparing topics than tracking exact counts.
Google job listings published this month describe teams generating and recommending AI content for Discover. A Staff ML Engineer listing states the Discover Feed Recommendation team leads by recommending AI-generated content to fulfil user interests, and describes agentic flows that generate content for Discover and Notifications. A director-level design listing describes turning Discover into a proactive, hyper-personalised concierge experience. Neither listing gives a timeline or indicates how much of the feed AI content would fill, and Search Console's generative AI report for Discover lists impressions as the only metric its help page describes.
The operating principle this leaves you with
No single reporting platform shows the full picture. Rank trackers simulate searches, which makes them vulnerable to anti-scraping measures. Analytics platforms record the attribution tags attached to visits, which makes them vulnerable to tagging errors. Search Console logs actual impressions served to real users, but only for Google properties. Bing Webmaster Tools provides directional AI search citation data, but rounds it into ranges. Each platform has a different method, a different scope and a different set of limitations.
The only reliable way to read performance is to cross-check them against one another. Where they agree, the signal is strong. Where they diverge, the divergence itself is the story, and it usually points to a data collection issue rather than a genuine change in visibility or traffic. Treating any single dashboard as authoritative is a category error, and October has made that clearer than most months do.