The data used daily by SEO professionals requires more verification than it seems. Three topics from this week illustrate this: skewed position tracking results, PageSpeed Insights scores changing without page modifications, and shared control of the robots.txt file for AI training.
- Les outils de suivi de positions peuvent fausser les rapports, car des résultats hors sujet et des liens goto non résolus masquent la page réellement classée.
- PageSpeed Insights utilise désormais Lighthouse 13.6.0, qui peut modifier les scores de laboratoire sans changement sur la page.
- Google, Microsoft et OpenAI prévoient de respecter un contrôle partagé du robots.txt pour l’entraînement de l’IA, mais aucun ne le prend encore en charge.
- Microsoft cessera, après le 1er mars 2027, de considérer NoArchive et NoCache comme des préférences d’entraînement ou de grounding.
Disrupted Tracking Results
Two phenomena affecting Google results collected by position tracking tools can skew reports. On one hand, results unrelated to the query can push a tracked page outside the collected list. On the other hand, unresolved ‘goto’ redirect links can hide the page actually ranked.
A SE Ranking study covering 10,000 US desktop Google keywords, published on November 3, indicates that 84% of reported cases involved results unrelated to a single word in the query. Among these blocks, 96.5% started at position 9 or beyond. The tool Nozzle, which uses a different methodology, reports a weekly share of affected SERPs that dropped to 1.5% in mid-October, then rose to 7.6% during the week of October 26.
According to DataForSEO, Google has implemented a mechanism limiting the number of goto links from the same results page that can be resolved within a given period. The company’s High Priority and Live modes would thus only resolve about half of the goto URLs. Serper’s status page mentions this limitation as of October 30, while SerpApi’s second incident, opened on November 3, still showed the status ‘Identified’ at its last check on November 4.
PageSpeed Scores Changing Without Apparent Reason
PageSpeed Insights now uses Lighthouse 13.6.0, a version that modifies the loading simulation used to calculate lab scores. A score can therefore change even though the page has not been touched. According to a post by Google’s Barry Pollard on Bluesky, the updated version of the simulated FCP (First Contentful Paint) and LCP (Largest Contentful Paint) metrics should be more accurate. JavaScript-heavy sites would be more affected by this effect.
The simulation engine, called Lantern, now relies on Chrome’s render-blocking label for each request, with the fetch priority used as a fallback. The field data in the same report, from real users, is not affected by this change. Since scores before and after the update are not based on the same simulation, part of the increase observed on optimized pages may therefore come from the new version. It is advisable to check that reports indicate the same Lighthouse version and compare field data before concluding an improvement.
Shared Control for AI Training
Google, Microsoft, and OpenAI announced on October 30 their intention to respect a shared control of robots.txt for AI training preferences, inspired by an Internet-Draft by Anthropic’s Kevin Kelley. The mechanism takes the form of a classic robots.txt group, associated with the __usage_ai-training policy token, with Allow and Disallow rules. According to the public timelines of the three organizations, none yet supports this control.
Google states it would treat a shared opt-out as a refusal of Google-Extended, when the site has not defined that setting. Its crawler documentation specifies that Google-Extended also covers grounding in Gemini Apps. Microsoft will use the shared control for training, will introduce a separate page-level preference, AI Grounding, and will cease, after March 1, 2027, to consider NoArchive and NoCache as training or grounding preferences.
In the section dedicated to technical SEO, Google’s John Mueller also stated that AI systems ‘build on SEO’, according to his reported remarks, and that he expects technical SEO to remain relevant. He discussed this during Search Central Deep Dive Europe 2026, in an interview published on October 29 with David Iwanow.
