Google's core update began on day two of a broken AI report
The November 2026 core update started Nov 10, inside the Gen AI report's Nov 9–11 logging gap. The order of operations that separates the two.
If your generative AI impressions fell between November 9 and 11, 2026, you cannot attribute that fall to the November core update. Google says the Generative AI performance report under-logged impressions for exactly those three days, and the core update started on November 10 — day two of the gap. Read the click axis first, hold the update open until the rollout completes, and only then compare AI impressions across a clean window.
That is the whole argument. Everything below is the working order that gets you there without discarding a real signal or protecting a fake one.
Two dated facts, one overlap nobody planned
On November 10 at 1:07 p.m. ET, Google announced the November 2026 core update on the Search Status Dashboard: "Released the November 2026 core update, which applies globally and to all languages. The rollout may take up to 2 weeks to complete." The same wording went out on Search Central's LinkedIn post, with Google calling it "a regular update designed to better surface relevant, satisfying content for searchers from all types of sites."
Two days later, on November 12, Google posted a data anomaly: "A logging error caused a decrease in impressions on the Generative AI performance report in Search for data from November 9 – November 11, 2026. This issue affects data logging only."
Read those two dates next to each other and the problem is arithmetic, not interpretation. The update started on November 10. The corrupted window opens on November 9. So in the AI-reporting axis, the first two days of the update's rollout are inside the known-bad window, and only the third day (November 9) is pre-update.
If you pull a 28-day chart of generative AI impressions right now, the dip you see starts before the update did. That is not a ranking story. That is the day the log broke.
Worth adding for context: Search Engine Roundtable had already flagged heavy ranking volatility on November 8 and 9, before anything was official. Volatility before an announcement is normal. It also means the week contains three separate things people will later collapse into one: pre-announcement churn, a logging defect, and a two-week rollout.
Why the AI axis is the one that can break quietly
The Generative AI performance report has a single metric. Google's help documentation describes it as showing how your site performs in generative AI features on Google Search, broken out by pages, countries, dates and devices, with a search-type split between text-based Web queries and multimodal ones. What it does not give you is clicks, click-through rate, position, or a query list.
That narrowness is the reason this defect is more dangerous than it looks. In the classic Performance report, a logging problem announces itself: clicks, impressions, CTR and average position all move, and the internal arithmetic stops reconciling. In the AI report there is nothing to reconcile against. Impressions are the whole instrument. When they drop for three days, the report gives you no second number that would let you catch the error yourself.
One detail that trips up reporting decks: those same AI impressions are already inside the classic Performance report's Web totals. The two reports overlap, so they should never be added together. If you have seen a dashboard where AI impressions were stacked on top of Web impressions to produce a bigger "total visibility" figure, the number was inflated before this bug existed and is differently inflated now.
The order of operations
Four steps. Run them in this order and the question resolves itself.
1. Mark the window before you look at a chart. November 9–11 goes into your instrument log as known-bad for generative AI impressions, and November 10 goes in as the update start date. Search Console has an annotations feature built for exactly this: a dated note attached to the property, so the next person to open the chart sees the defect instead of rediscovering it as a mystery. This is five minutes of work and it is the difference between an analysis and a guess.
2. Read the click axis first. Open the classic Performance report, filter to the Web search type, and look at clicks and average position for November 9–11. This axis was not affected by the logging error. If clicks and position are flat through the gap and then move on November 13 or later, the defect is irrelevant to your conclusion. If clicks fell on November 9 or 10 specifically, the update is a plausible cause and you are now looking at the right signal at the right resolution.
3. Compare AI impressions only after November 12. Whatever the pre-update frame is — 28 days, 90 days, whatever your reporting standard says — compare it to a window that starts after the gap closes. Do not compute a trend across November 9–11, and do not compute a trend across the rollout either: Google said the update may take up to two weeks from November 10, which puts completion around November 24. Judging an update mid-rollout is how teams end up publishing a recovery plan for a dip that was still in motion.
4. Reconcile the two axes in direction only. The AI report says how often your links were shown in AI features. The classic report says what happened after someone clicked. They agree or disagree on direction; they never agree on magnitude, and adding them together is meaningless. A week where AI impressions rise and Web clicks fall is a real and common pattern: more surface exposure, less traffic per exposure.
| What you see | Report that answers it | Window you can trust |
|---|---|---|
| AI impressions dip Nov 9–11, then recover to a normal band | Generative AI performance report | Nothing before Nov 12 for AI impressions; the dip is the logging error |
| Clicks or average position fall starting Nov 13–14 and stay down | Classic Performance report, Web search type | Nov 1 onward; the click axis was never affected |
| Both axes fall, clicks first | Classic report, then query-level rows | Post-Nov 12 only, and treat the update as open until the rollout completes |
| AI impressions flat, clicks falling | Classic report, then page-type segmentation | Your whole window; this is not an AI story at all |
| Everything flat but the client "feels" a drop | Neither | Re-check position at the query level before spending a meeting on it |
The three reads this week actually produces
In client work, the same three shapes keep coming back, and they mean different things.
The first is the false alarm: AI impressions fall for three days and come back. Three days of under-logged impressions in a two-week rollout is exactly the shape you would expect from the documented defect. Nothing to fix, but it must be annotated or it will resurface in the quarterly deck as "the week we lost AI visibility."
The second is a genuine ranking move: clicks and average position fall from November 13 onward, and the fall holds for more than a week. That is what a core update looks like from the inside. It is not something you undo in a week, and Google's own guidance is that there is nothing specific to fix — the honest answer is that the page is being out-ranked by pages that satisfied the query better, and you find out where by reading query-level rows rather than site-level averages.
The third is the one nobody wants to report: the numbers did not move, and the expectation did. Somebody read a news post about volatility, opened a chart, and found a story in the noise. That is a communication problem, not an SEO problem, and it is best solved with a one-page method statement rather than another dashboard.
The pattern under all three: an attribution argument is only as good as the instrument it rests on. When two dated events land in the same week and one of them corrupts the instrument, the order you check things in is the analysis. Same discipline applies to the counting quirks inside Search Console's AI position data, which we wrote about separately because they distort averages in a similar direction.
What this does not cover
Three honest limits.
It does not cover the Discover variant of the generative AI report, which has its own anomaly history and its own dimensions. If your traffic mix leans on Discover, check that report's notes separately before assuming the same window applies.
It does not tell you why you lost rankings, and no dashboard will. Core updates redistribute visibility according to signals Google does not itemize. The useful work is comparative: which query clusters moved, which page types moved, and whether the pages that moved down share a structural weakness.
It also does not fix the deeper weakness this week exposed: an impression-only visibility metric is fragile by design. There is no click data in the AI report to cross-check it against, so a three-day logging error is undetectable from inside the report. Teams that treat generative AI impressions as the primary KPI are running a single-sensor instrument and calling it a trend.
The habit worth keeping
Keep an instrument log. Dated entries, one line each: when a defect was published, when it was fixed, when an update started, when it completed, when you changed your own tracking. It takes a minute per entry and it is the single cheapest defence against the most expensive failure mode in this work, which is confidently explaining a number that was never real.
Two useful precedents from this year, if you want the shape of the problem. When Google moved the AI Overview source row, the lesson was that the interface can change underneath a stable baseline — that one is worth reading before you re-cut your series. And when the September spam update landed, the useful work was deciding what a dated update piece should actually answer; the September analysis is the worked example of "what changed" versus "what do I do."
Meanwhile, that dip in your AI impressions for November 9–11 is a logging error, not a verdict on your content. If you want a second pair of eyes on the read — which axis to trust, where to cut the series, what the query-level rows are saying — our SEO and GEO team runs this as a scoped attribution review, and you can start here with the window you already export. If you are still deciding whether the drop is even real, start with why a page-2 ranking drop can be a measurement artifact before you commission anything.



