We respect your privacy.

We use strictly necessary cookies to keep you signed in and to protect against CSRF. With your permission we also use a small amount of first-party analytics to improve the product. We do not sell your data and we do not use third-party advertising trackers. See our cookie policy and privacy policy .

← All posts

One week, four changes, one traffic chart

Crawlmind Engineering··5 min read

A confounded week is a period when several independent changes hit the same metric at once, so no single chart can tell you which one moved it, and November 9 to 12, 2026 was one for Google Search.

If your organic or AI visibility numbers shifted in the middle of November, you will be tempted to name a cause from the shape of the line. That is the wrong order of operations. Four separate things happened inside four days, and each one touches a different part of what you measure. The useful work right now is to separate them, not to explain them.

#What changed, and when

Here is the timeline, with the source for each item.

November 9 to 11: a Search Console logging error. Google's data anomalies page says a logging error caused a decrease in impressions on the Generative AI performance report in Search for data from November 9 to November 11, 2026. Google adds that the issue affects data logging only. Search Engine Roundtable reported it on November 12 and noted that it overlaps the start of the core update.

November 10: the November 2026 core update. Google started the rollout on November 10, and its announcement, quoted by Search Engine Roundtable, says it may take up to 2 weeks to complete. Google described it as a regular update.

November 10: Gemini 4 Carbon in AI Mode. According to Search Engine Journal, Carbon began rolling out to Ultra subscribers in AI Mode the same day. That is a paid audience, so its direct effect on most sites is small, but it is still a model change on a surface you may be tracking.

November 12: Gemini 4 Hydrogen becomes the AI Mode default. The same Search Engine Journal report says Hydrogen became the default model in AI Mode for all users worldwide on November 12. Google Search VP of Product Robby Stein said it is also being used for the hardest questions in AI Overviews. Neither Google's post nor Stein's said which model Hydrogen replaced.

So one window contains a measurement defect, a ranking change, and two changes to the model that writes AI answers. Each of those can move a number in the same direction.

#Why the overlap breaks the usual diagnosis

Each change acts on a different layer, and the layers feed each other.

The logging error changes what Search Console reports without changing what users saw. A drop in AI impressions from November 9 to 11 is, by Google's own account, at least partly a counting problem. Google has fixed a similar error before: an earlier logging error affected the same report for August 13 to 17, 2026, and Google later posted that the missing data had been restored. Until something similar is posted for November, treat those three days as incomplete rather than low.

The core update changes which pages rank in classic results. AI Overviews and AI Mode retrieve from Google's index, so a ranking shift can change which pages are available to be cited, even if the model did not change at all.

The model changes alter how retrieved pages get selected and summarized. We wrote about this in model swaps are the new core updates: a new default model can reshuffle citations with no change to your content. This time the model swap and the core update arrived two days apart, which removes the clean before and after you would normally use to tell them apart.

The result is that a citation you lost on November 13 could come from a page that lost rank, from a model that weighs sources differently, or from both. A drop in reported AI impressions could be real, or it could be a logging artifact. None of these explanations is ruled out by the data you have today.

#What to do this week

Annotate before you analyze. Add all four dates to whatever dashboard your team reads. A chart without annotations invites people to attach the dip to whichever cause they already believed.

Exclude November 9 to 11 from AI impression comparisons. Google lists those dates as affected by the logging error, so do not average them in, and do not use them as a baseline. If Google later restores the data, as it did for the August window, rerun the comparison.

Wait for the core update to finish. Google's core updates guidance says to confirm the rollout has finished on the Search Status Dashboard, then wait at least a full week after it completes before analyzing your site in Search Console. It recommends comparing that week with a week before the update started. With a rollout that can run up to two weeks, an early verdict is a guess.

Split classic and AI metrics. Look at classic web clicks and positions separately from AI Overview and AI Mode citations. If classic rankings held steady for a set of pages but their AI citations changed, the model change is the more likely driver for those pages. If both moved together, the core update is a candidate, and the model change may be layered on top.

Compare at the page level, not the site level. A sitewide number blends pages that gained from one change with pages that lost from another. Pick the pages that matter most to revenue and check each one: rank before and after, cited before and after, and which competing URLs took the slot.

Hold content changes on affected pages. If you rewrite a page during an open core update and a fresh model default, you add a fifth variable. Keep a list of what you would change, and act on it once the update has completed and a clean week of data exists.

#What this means for AI visibility reporting

Weeks like this one will repeat. Google ships core updates on its own schedule, model defaults in AI Mode change on a different one, and Search Console reports have had at least two logging errors on the Generative AI report this year, both listed on the data anomalies page. Any reporting process that reads one line and names one cause will be wrong often.

The fix is procedural. Keep a change log of every external event with its date and source. Store page-level citation results with the date and the engine surface they came from. When numbers move, check the log first and ask which changes overlap the window. If more than one does, say so in the report and hold the conclusion until the overlap clears.

For November 2026, that means the honest answer to "why did our AI visibility change?" is: several things changed at once, one of them was a counting error, and the comparison Google recommends is not possible until the core update has finished and a week has passed.

Related field notes

Share or discuss

Field notes in your inbox

New posts, no spam. Roughly monthly. Unsubscribe with one click.