Read the anomalies page before the AI report
Crawlmind Engineering··4 min read
Google's Search Console data anomalies page is the official log of reporting faults, and it should be the first thing you check when the generative AI performance report shows a sudden drop.
The latest reason is recent. Google's Data anomalies in Search Console page now carries an entry for November 9 to November 11, 2026, saying a logging error "caused a decrease in impressions on the Generative AI performance report in Search" for those dates, and that the issue "affects data logging only." In other words, the chart dipped and your actual presence in AI Overviews and AI Mode did not.
The timing makes that entry easy to misread. The November 2026 core update began rolling out on November 10, inside the same window. A team that sees AI impressions fall on the day a core update starts will reach for the core update as the explanation. For those three days, the more likely cause is the logger.
#A young report with a busy incident log
The generative AI performance report is new. Google introduced it on June 3, 2026 for a subset of sites and says it reached all websites worldwide by August 31, 2026. It reports one metric, impressions, described as how often URLs from your site appeared in generative AI features in Search and Discover. There are no clicks and no queries in it.
In the months since launch, the anomalies page has recorded four entries that touch generative AI data:
- June 24, 2026: Discover and Generative AI in Discover.
- August 13, 2026: Discover and Generative AI in Discover.
- August 13 to August 17, 2026: Generative AI in Search, an impressions decrease from a logging error.
- November 9 to November 11, 2026: Generative AI in Search, the same wording.
Each one says the issue affects data logging only. None of them describes a change in how Google ranks or selects sources.
That is not a criticism of the report. New pipelines break more often than mature ones, and Google publishes the faults instead of hiding them. The practical point is that a report this young needs a habit of checking its incident log before anyone writes a narrative about a trend.
#Restored data is not guaranteed
The two Search entries ended differently, at least so far. The August 13 to 17 entry carries an update dated August 21: "Missing data for this period has been restored." The November entry had no restoration note when we checked it.
Other entries on the same page show that restoration is not automatic. One older entry, for February 28 and March 1, says plainly: "This data won't be recovered." The anomalies page itself only promises to record known issues "in the last 3-16 months," so the log also ages out. If your reporting looks back further than that, the explanation for an old dip may no longer be on the page at all.
That gives you three possible states for any flagged window:
- Flagged and later restored. The numbers you see now are complete, but any export taken during the incident is not.
- Flagged and not yet restored. The numbers are low for a known reason and may change.
- Flagged and declared unrecoverable. The numbers are permanently low and should be excluded from comparisons.
A dashboard that treats all three the same will produce wrong week-over-week figures in two of them.
#What to change in your reporting
None of this needs new tooling. It needs a few rules applied every time the report is read.
Check the log before the chart. When AI impressions move sharply, open the anomalies page first. If the dates overlap an entry, the entry is the working explanation until you have evidence otherwise.
Annotate the window in your own data. Most teams export Search Console data into a warehouse or spreadsheet. Add a column or annotation that marks flagged dates and their state (restored, pending, unrecoverable). Without it, a later analyst will rediscover the dip and blame something else.
Re-pull after a restoration. If Google restores data, your earlier export is still wrong. Re-fetch the affected range so your stored copy matches what Search Console now shows.
Do not compare across a flagged window. For November, a week-over-week comparison that includes the flagged November 9 to 11 window mixes a logging fault with the first days of a core update. Compare the days before the incident with the days after it, or wait until the core update completes and the entry is resolved.
Keep two sources of truth. The generative AI report counts impressions only, so a logging fault there cannot be cross-checked with clicks inside the same report. Pair it with something independent: your own server logs for AI crawler activity, analytics for AI referrals, or repeated prompt testing against the engines you care about. If the first-party report drops and the independent signals do not, the report is the likely problem.
#Separate the three things that moved this week
Mid-November produced at least three overlapping events for anyone reading AI visibility data from Google: a core update that started on November 10 and is still rolling out, a logged reporting fault covering November 9 to 11, and the normal day-to-day variance of AI answers. Only one of those reflects how Google treated your pages.
Teams that blame every dip on the algorithm end up making content changes in response to a logging bug. Those changes then get credited, or blamed, for whatever happens next. Reading the incident log first is cheap, and it keeps a reporting fault from turning into a content decision.
The broader lesson applies to any first-party AI report, from Google or anyone else. A measurement surface is software, and software has incidents. Before a number goes into a slide, check whether its publisher has already said the number is wrong.
Related field notes
November 21, 2026 · 5 min
AI answers now hand out your coupon codes
ChatGPT and Google now show discount codes inside answers, and most codes chatbots return do not work. Make sure the codes they find are yours and current.
November 21, 2026 · 5 min
Shallow pages lose in AI search first
Google's head of Search says AI answers leave little room for pages that only restate facts. How to find your shallow pages and decide what to do with each.
November 20, 2026 · 5 min
The ten blue links are back as a tab
Google, Bing and now ChatGPT offer or test a view with only web links. That view ranks pages, not passages, and AI visibility reports often skip it.
Share or discuss
New posts, no spam. Roughly monthly. Unsubscribe with one click.