Rank Data Problems, PageSpeed Scores & AI Training Opt Out, SEO Pulse

In SEO, we often treat our tools as the absolute source of truth. We see a dip in a ranking report or a jump in a speed score and immediately start looking for what changed on the site. But the reality is that the tools themselves change, and the way they collect data from search engines can shift without warning. The same pattern also shows up in UK Requires Opt Out, where the practical question is how the signal becomes visible.

This week, several updates highlight why we need to be cautious about the numbers we report. From how rank trackers handle redirects to how AI crawlers are viewed by Google, the focus is shifting toward the methodology behind the data rather than the data itself.

The Noise in Your Rank Tracking Data

If you have noticed strange fluctuations in your ranking reports recently, it might not be a change in your actual visibility. There are two specific issues currently affecting how rank tracking tools collect data from Google.

First, there is an issue with unrelated listings. Some tools are picking up results that are based on a single word from a query rather than the full intent. According to a review of 10,000 US desktop keywords by SE Ranking, about 84 percent of these flagged cases were unrelated listings. Interestingly, the vast majority of these, roughly 96.5 percent, appeared at position 9 or lower.

Second, there is the problem of unresolved "goto" links. These are Google redirect URLs that can mask which actual page is ranking. DataForSEO has noted that Google seems to be limiting the number of these links that can be resolved within a certain timeframe. This means some trackers may only be resolving about half of these URLs in their high priority or live modes. This connects with ChatGPT Tags & AI Search Links when the same signal needs a clearer operating decision.

Why this matters for your reporting

The danger here is misdiagnosing a "drop" in rankings. If you see a sudden decline at position 9 or below starting in late September, check if the URLs replacing yours are irrelevant one word matches. For redirects occurring around late October, compare the URL stored by your tracker with the final destination.

The tradeoff is between the convenience of automated trackers and the accuracy of first party data. To get the truth, you should cross reference these drops with Google Search Console's impression and click reports. If GSC shows steady traffic but your tracker shows a drop, the problem is the tool, not your SEO.

PageSpeed Insights and the Lighthouse 13.6.0 Shift

It is a common frustration: you haven't touched a single line of code, yet your PageSpeed Insights (PSI) score has changed. This is currently happening because PSI has updated to Lighthouse 13.6.0.

This update changes the loading simulation used for lab scores. Specifically, the "Lantern" simulation now uses Chrome's render blocking label for every individual request. If that label is missing, it defaults to the fetch priority value. Google's Barry Pollard has indicated that this should lead to more accurate results for First Contentful Paint (FCP) and Largest Contentful Paint (LCP), generally trending toward better scores.

Websites that rely heavily on JavaScript are likely to see the most significant shifts in their scores due to this change in how render blocking is handled.

Why this matters for your performance audits

When you see a score increase, it is tempting to claim a victory for your recent optimizations. However, because the simulation itself has changed, you cannot compare a pre update score with a post update score directly. They are essentially two different tests.

The decision here is to prioritize field data over lab data. Field data is based on real user experiences and is not affected by Lighthouse version updates. Before reporting a performance win to a client or stakeholder, ensure you are comparing reports from the same Lighthouse version and verify the trend against the field data.

A Unified Front for AI Training Opt Outs

Managing how AI models use your content has been a fragmented process, with different tags for different companies. Google, Microsoft, and OpenAI have announced plans to honor a shared robots.txt control for AI training preferences.

This proposed control uses a policy token called __usage_ai-training within the robots.txt file, allowing for both Allow and Disallow rules. While the three companies have announced their intent, they are not yet supporting it in their current live environments.

Google intends to treat this shared opt out as a substitute for Google Extended if that specific tag isn't already defined. It is also worth noting that Google Extended covers grounding in Gemini Apps. Microsoft will use the shared control for training but plans to introduce a separate preference for AI Grounding at the page level. Microsoft will stop recognizing NoArchive and NoCache as training preferences after March 1, 2027.

Why this matters for content ownership

The goal is simplicity, one rule to rule them all. However, the practical reality is that "opting out" doesn't mean the same thing across different platforms. Google's approach to Gemini training versus grounding is bundled, whereas Microsoft is splitting training and grounding into different controls.

If you currently rely on NoArchive or NoCache to keep your content out of Microsoft's AI models, you have a hard deadline of March 2027 to migrate to the new system. You must decide whether you want a blanket opt out or a more granular approach that allows some AI usage while blocking others.

Technical SEO in the Age of AI Search

There is a prevailing fear that AI search will make traditional SEO obsolete. However, Google's John Mueller suggests the opposite. He argues that AI systems actually build upon the foundations of SEO, and technical SEO will remain highly relevant.

Mueller points out that AI systems crawl sites differently depending on their intent, some crawl for general training, while others crawl to find the most current information. He views this as simply "another flavor of technical SEO."

Regarding content formats like listicles, Mueller notes that when a specific tactic works, it is often copied poorly by others until the algorithm adapts. His advice is to focus on a solid technical foundation rather than chasing temporary format trends.

Why this matters for your technical roadmap

This reframes AI crawling as a technical audit item rather than a separate "AI SEO" strategy. You don't need a new playbook; you just need to expand your existing technical audits to include how AI crawlers are interacting with your pages.

The tradeoff is between spending time on "AI specific" content hacks and investing in infrastructure. The decision should be to double down on crawlability, site architecture, and robots.txt management, as these are the levers that control how both traditional and AI search engines perceive your site.

The Gap Between ChatGPT Queries and Google Search

There is a lot of noise regarding the "threat" of ChatGPT to traditional search. Recent comments from OpenAI's Adam Fry suggest that ChatGPT's volume of search like queries is roughly one fifth that of Google's.

It is important to define what "search" means in this context. OpenAI defines a search as any question that triggers ChatGPT to go to the web. However, this is a measure of queries, not traffic. Ahrefs' referral data provides a different perspective: in October, Google accounted for 24.74% of tracked traffic across over 118,000 sites, while ChatGPT accounted for only 0.36%.

Why this matters for your channel strategy

There is a massive difference between a user asking a question in a chat interface and a user clicking a link to visit a website. A "search like query" in ChatGPT does not automatically translate to a referral visit for the publisher.

When analyzing the impact of AI on your traffic, you must distinguish between search volume (the number of queries) and referral share (the actual visits). If you are reporting on the growth of AI search, be explicit about which metric you are using. Relying on query volume alone can lead to an overestimation of the immediate threat to your traffic.

Prioritizing Method Over Metrics

The common thread across all these updates is the necessity of questioning the method before trusting the number. Whether it is a ranking report, a speed score, or a market share percentage, the number is only as good as the process used to generate it.

Rank data can be skewed by one word results and redirect throttling. PageSpeed scores can shift because of a simulation update in Lighthouse. AI market share varies wildly depending on whether you measure queries or referrals. Even the shared AI opt out is interpreted differently by the companies involved. A useful companion note is AI Detected Pages Rank Lower, because it looks at a nearby part of the same system.

When reporting changes in your metrics, the most professional approach is to document the tool version and the specific criteria used to produce the report. This prevents the "phantom" wins or losses that occur when the underlying measurement system shifts beneath you.