Google’s September 2026 spam update is finished.

That changes the job for site owners.

While the rollout was active, daily ranking movement was difficult to separate from the update itself. Google has now closed the incident, giving analysts a fixed rollout period: 24 September through 8 October 2026.

The official Search Status Dashboard lists the update as global, covering all languages, with a duration of about 13 days and 16 hours.

Google did not identify a specific spam policy as the target.

That means the useful question is not:

“Did the spam update hit us?”

It is:

Which URL clusters changed differently from the rest of the site, and what evidence explains that difference?

NEMO’s recommendation is to use the completed rollout as a measurement boundary, not as permission to start deleting pages.

Start with three windows, not one scary screenshot

Search Console makes it easy to compare arbitrary dates and accidentally manufacture a story.

Use three defined periods instead.

Window Dates Purpose
Baseline 10–23 September Establish pre-update behaviour
Rollout 24 September–8 October Measure what changed during Google’s update
Post-rollout 9–22 October Re-test after the system has stopped changing

The rollout period contains partial start and end days, so avoid comparing raw totals as if every window contains identical exposure.

Use daily averages for clicks and impressions.

The post-rollout window does not exist yet in full. That is useful rather than inconvenient: run the first audit now, save the sheet, then repeat it after 22 October.

If the same clusters remain weak in the second capture, the evidence becomes much stronger.

Step 1: establish the sitewide baseline

Export Search Console Performance data for the baseline and rollout windows.

At minimum collect:

  • URL;
  • query;
  • country;
  • device;
  • clicks;
  • impressions;
  • CTR;
  • average position.

Keep country and device available even if you initially analyse them together.

A site serving Germany, France and Poland can easily hide a national change inside an aggregate European number.

Calculate daily-average clicks and impressions for the site as a whole.

Those numbers are your baseline.

Do not diagnose pages yet.

The sitewide result tells you whether a cluster’s movement is unusual or simply part of the broader site pattern.

Step 2: group URLs by publishing and intent cluster

The unit of analysis should not automatically be the individual URL.

Group pages that compete for the same type of demand.

Examples:

  • product-category pages;
  • comparison pages;
  • informational guides;
  • programmatic location pages;
  • editorial news;
  • old archive content;
  • affiliate reviews;
  • forum or UGC pages.

Also create publishing cohorts where useful:

  • published in the last 30 days;
  • 31–90 days old;
  • 3–12 months old;
  • older than one year.

This matters because twenty similar URLs moving together is a different problem from one page falling while its neighbours remain stable.

A spam-update investigation should look for patterns, not anecdotes.

Step 3: calculate NEMO Cluster Divergence

For each cluster, calculate its percentage change in impressions per day between the baseline and rollout periods.

Then calculate the site’s overall percentage change.

Use:

Cluster Divergence = cluster impressions/day % change − sitewide impressions/day % change

Example:

  • sitewide impressions/day: −8%
  • comparison-page cluster: −31%

Cluster Divergence:

−31% − (−8%) = −23 percentage points

The cluster is therefore underperforming the site’s general movement by 23 points.

Now compare that with another cluster:

  • sitewide: −8%
  • product categories: −10%

Divergence:

−2 points

Those two situations should not receive the same remediation.

Cluster Divergence is not proof that Google’s spam systems classified those pages in a particular way.

It is a prioritisation metric.

It tells you where to investigate first.

Step 4: separate visibility, demand and CTR

Clicks alone are a bad diagnostic.

A cluster can lose clicks for at least three very different reasons.

Visibility changed

Impressions fall and average position also weakens across many queries.

That is a ranking-loss candidate.

Demand changed

Impressions fall but positions for the cluster’s established queries remain broadly similar.

Seasonality, news cycles or query demand may be contributing.

Search Console impressions are only a demand proxy because ranking affects whether an impression is recorded. Treat this as a diagnostic clue, not a search-volume measurement.

CTR changed

Impressions and positions remain comparatively stable, but clicks and CTR fall.

That points you toward the search-result environment rather than automatically toward page quality.

Ads, AI features, rich results, changing titles or different user behaviour can all alter clicks without producing the same ranking pattern.

Keep these causes separate.

Step 5: inspect query demand before blaming the page

For every major losing cluster, split queries into:

  • branded;
  • non-branded;
  • navigational;
  • commercial;
  • informational.

Then examine whether losses are concentrated in a query class.

A page may appear to have been “hit” when the real movement comes from a small group of declining queries.

Likewise, a cluster can maintain total impressions while losing the commercial queries that actually produce revenue.

This is why NEMO’s audit keeps query data beside URL data.

The business question is not simply whether traffic changed.

It is which demand changed.

Step 6: sample representative pages instead of auditing everything

Do not open 300 URLs because a graph turned red.

For each affected cluster, select a small representative sample:

  • two clear losers;
  • two relatively stable pages;
  • two winners, where available.

Compare them side by side.

Check:

  • search intent;
  • content overlap;
  • originality;
  • page purpose;
  • author or business attribution where relevant;
  • internal linking;
  • canonical status;
  • indexability;
  • template differences;
  • structured data;
  • publication and update history;
  • external references or links;
  • whether the page still satisfies the query it targets.

The stable and winning pages are controls.

Without them, an audit often becomes a list of generic SEO recommendations that could have been written before the update happened.

Step 7: use the Five-Way Triage

Every sampled URL should end in one of five buckets.

1. Hold

The page still satisfies its intent and the evidence does not show a persistent page-level problem.

Do not change it simply because the rollout happened.

2. Improve

The intent remains valuable, but the page has a clear weakness: thin evidence, outdated information, poor structure, weak differentiation or an obvious mismatch with the query.

Improve the identified weakness.

Do not rewrite the page indiscriminately.

3. Consolidate

Two or more pages substantially compete for the same task and neither has a strong independent reason to exist.

Consolidation may be appropriate.

That decision should come from overlap evidence, not from update timing.

4. Retire

The page is obsolete, unsupported, has no meaningful remaining demand and provides little independent value.

Retirement can mean removal, noindexing or another appropriate status depending on the page.

Do not use noindex as a generic spam-update recovery tactic.

5. Investigate

The pattern does not look primarily editorial.

Check technical indexing, canonicals, accidental noindexing, template changes, security issues and Search Console manual-action reporting before editing content.

Build one audit sheet

A useful spreadsheet needs fewer columns than most SEO audits.

Use:

Field Purpose
URL Page being evaluated
Cluster Intent/template group
Publish cohort Age or publishing batch
Baseline impressions/day Pre-update visibility
Rollout impressions/day Update-period visibility
Impressions change Direction and magnitude
Sitewide change Control
Cluster Divergence Relative movement
Clicks/day change Traffic effect
CTR change Search-result effect
Position change Ranking clue
Main query class Demand type
Representative sample Winner / stable / loser
Evidence found Specific diagnosis
Triage Hold / Improve / Consolidate / Retire / Investigate
Action date When a change was made
Re-test date When to evaluate it again

That is enough to recreate the analysis later.

Reproducibility matters because the first explanation is often wrong.

Do not turn correlation into an update diagnosis

Google says the September spam update applied globally and to all languages.

Google did not publicly say that this rollout targeted one particular content format, industry, AI-content pattern or site type.

So avoid conclusions such as:

  • “AI content was targeted”;
  • “affiliate pages were targeted”;
  • “programmatic SEO was targeted”;
  • “European sites were hit harder”;
  • “a core update is next.”

Those claims need their own evidence.

A site losing traffic during a spam update is correlation.

The URL-cluster audit is designed to determine whether there is a coherent site pattern worth acting on.

The most important audit happens after the rollout

Today’s analysis establishes the first post-update record.

Do it now.

Then leave a copy untouched.

After 22 October, run the same extraction again using 9–22 October as the post-rollout window.

Ask:

  1. Did the losing cluster stay weak?
  2. Did the site’s baseline recover while that cluster did not?
  3. Did query demand recover?
  4. Did CTR change independently of position?
  5. Did pages changed after the first audit improve differently from untouched controls?

That second pass is much harder to fool with rollout volatility.

It also creates a dataset worth referring to during the next Google update.

The update is finished. The diagnosis should slow down.

Google closing the rollout removes one source of uncertainty.

It does not identify the cause of every site’s movement.

The advantage now is that analysts have a fixed period to examine.

Use 24 September–8 October as the rollout window, compare clusters against the site baseline, separate visibility from demand and CTR, sample winners as well as losers, and document every intervention.

Then rerun the analysis after 22 October.

That process will usually tell you more than deleting pages while staring at a sitewide traffic chart.

For related monitoring and measurement work, see NEMO’s Search & AI intelligence.

Sources

  • Google Search Status Dashboard — September 2026 spam update incident.
  • Google Search Central — spam policies.