Google's September spam update ended. Your first three days of data are misleading.

Google's September spam update finished on October 8, 2026, after about 14 days of rollout. If your traffic chart has moved since, the useful question is not whether the update hit you. It is whether you are reading the chart in a way that can answer that at all.

We pulled our own Search Console data for the window on the day the rollout ended. The raw read of the first three days showed a 29% drop in clicks. The weekday-adjusted read showed the same three days landing within half a click of expectation. Nothing had changed except the calendar.

That gap is the whole article. A three-day total after an update mostly tells you which days of the week fell inside your window.

What is established about this update

Google's Search Status Dashboard records a spam update released on September 24, 2026 and completed on October 8, 2026. Google described it as applying globally and to all languages, with a rollout of up to two weeks.

A few things are known, and worth stating plainly:

• It ran roughly 14 days. The other spam updates of the year finished in about two days; the August one ran from August 18 to 21. This one took four times longer.
• It did not target link spam, and Google announced no new policy with it. It improves the automated systems that detect pages violating the existing spam policies.
• Google would not say what share of queries or searches were affected.
• Google runs periodic refreshes of spam updates afterwards, and has said recovery can take many months.

That last point matters more than the dates. Whatever your chart does this week is not the final verdict, and whatever it does next month is not necessarily the update either.

What we measured on mintec.co

We track mintec.co as a Search Console domain property and export clicks and impressions daily. Three windows, all from the same property:

Read those three rows the way most people read them and you get two stories that contradict each other. During the rollout we were up 7.8% on clicks and 14.7% on impressions against the prior two weeks. After the rollout we were down 29.4% on clicks per day.

Both numbers are correct. Neither describes Google.

The raw read looks like a crash, and it is a calendar effect

October 9, 10 and 11 were a Friday, a Saturday and a Sunday. Our Saturday average during the rollout was 26 clicks. Our Tuesday average was 79.5. Any window that ends on a weekend will read as a collapse next to a window measured across full weeks.

So we built the expectation the other way round: average each weekday across the rollout window, then add up what Friday, Saturday and Sunday should have delivered.

The model expected 114.5 clicks for those three days. We got 114. Against the pre-rollout weekdays it expected 108, so the same three days came in 5.6% above the earlier baseline. Impressions tell the same story from the other side: 21,181 actual against 22,582 expected, a 6.2% shortfall, or 8.3% above the pre-rollout expectation.

One honest caveat inside our own data: Thursdays averaged 52 clicks during the rollout against 63 before. That is four or five days per window. It is not a finding, and we are not going to build a diagnosis on it.

Three more ways to misread the window

The calendar is only the first trap. Three more are live right now.

Search Console itself is late

Search Engine Roundtable reported the performance report running more than 64 hours behind on October 11, the worst delay since December 2025. Our export this morning stops at October 11, two days short of today. Any comparison that treats missing days as low traffic will invent a decline.

Rank trackers are returning phantom results

From September 26, Google began serving incomplete and incorrect results to automated searches. SEOmonitor documented it alongside SISTRIX, SE Ranking and AWR, which describe a dual reality where the page a tool fetches differs from the page a person sees. Nozzle reported losing roughly 80% of the data it could collect. The rule that survives this: a tracker movement is a hypothesis, and it becomes evidence only when it holds across several days and shows matching movement in Search Console clicks and impressions.

The volatility everyone dated wrong

Ranking turbulence on September 23 and 24 preceded the announcement. Plenty of post-mortems have pinned that noise on the update, which puts their start date in the wrong place and their recovery clock in the wrong place too.

The three reads we run after a rollout ends

This is the sequence we use for our own site and for client properties. It is deliberately boring.

First read, day zero

Log the dates in the change freeze note and pause broad editorial work. Fix broken things: a 404, a blocked page, a wrong price. Do not start a 200-page cleanup in the same week you need to read the data. The ten-minute triage we run before touching content covers what to record and how to split branded from non-branded queries.

Second read, day three to seven

Compare like weekdays with like weekdays, and report absolute numbers next to percentages. On a site our size, one good Tuesday is 80 clicks and one bad Saturday is 20; the percentage swings while the site does not. Split branded from non-branded before you decide anything, because a brand problem and a discovery problem look identical in the total line.

Third read, day fourteen

Open the Pages tab and work only from URLs with material movement. We use an operating threshold of at least 20 clicks and 20% against the comparison period, plus any non-branded query tied to a commercial decision. Then classify what you find instead of acting on the shape of the graph. The five-step diagnosis for drops in the AI era separates an update problem from a zero-click problem, and they need opposite responses.

If you also watch your AI visibility panel, same caution

The generative AI report in Search Console shows impressions inside AI Overviews and AI Mode, and nothing else: no clicks, no queries, no position. It is UI-only, and we confirmed that today by calling the Search Analytics API with searchType=generative_ai, ai_overviews and aiMode on our own property. All three returned zero rows with no error, while searchType=web returned rows normally. You cannot script it, you cannot archive it, and a logging error already cut its reported impressions starting August 13.

So a panel that shows only impressions, delayed by days, with a known error in its history, is exactly where a dramatic narrative gets built. Watch it, chart it weekly, and do not let it carry a conclusion on its own. Our guide to running that report as an audit explains what it can and cannot support.

What not to do this week

Do not rewrite pages because a three-day window looked bad. Do not file a reconsideration request; algorithmic spam demotions have no such process. Do not answer an update with a burst of publishing, since scaled content abuse is judged by purpose and value rather than by who typed the lines, which is what the August update actually penalized. And do not repeat a recovery or collapse story from a tracker until Search Console agrees with it.

The update is over. What happens next runs on Google's refresh schedule, not on your publishing schedule.

FAQ

When did Google's September 2026 spam update finish? September 24 to October 8, 2026, completing at 4:37 a.m. ET, global and in all languages. It was the longest of the four spam updates Google ran that year.

How long does recovery take? Google says many months, because its systems need to learn that the site complies with the spam policies, and it runs periodic refreshes afterwards. There is no reconsideration request for algorithmic demotions.

Why did my tracker show a crash on September 26? From that date Google started returning incomplete and incorrect results to automated searches, so tools recorded movement real users never saw. Cross-check against Search Console clicks and impressions, and against several consecutive days, before treating any tracker movement as real.