Google's Spam Update Is Live: Check Your AI Content

Google released its September 2026 spam update on the 24th. It is still rolling out. One week in, Google also rewrote its guidance on AI-generated content. If your website has AI-written pages, you have a short window to find out where you

Google released its September 2026 spam update on the 24th. It is still rolling out. One week in, Google also rewrote its guidance on AI-generated content.

If your website has AI-written pages, you have a short window to find out where you stand before you start fixing the wrong thing.

Here is what Google has said, what it hasn't, and what we would check this week on a small-business site.

Google has not said what this update hit

The Search Status Dashboard entry is one sentence. Google released the update at 09:15 Pacific on September 24, it applies globally and to all languages, and the rollout may take up to two weeks.

No target is named. Not links, not AI content, not scraped pages. Anyone telling you this update "hits AI sites" is guessing, and some of them are selling you a fix for the guess.

What Google does say, in its explainer on spam updates, is that it occasionally improves how its automated detection works, including SpamBrain, its AI-based spam-prevention system. Sites that see a change should review the spam policies. Sites that violate them may rank lower or not appear at all.

The dashboard gives one more useful fact. The earlier 2026 spam updates ran 19 hours (March), about two days (June) and about two and a half days (August). This one is budgeted for up to fourteen. Your traffic chart will wobble for longer, and a wobble on day six tells you very little.

Your traffic chart is only readable after the rollout ends

The mistake we see most is a business owner opening Search Console on day three, seeing a dip, and rewriting twenty pages the same afternoon. Do not do that.

Start with the dates. Compare September 24 onward against the four weeks before it. Then look at your ten pages that bring in the most visits, one at a time, and ask two questions.

Did impressions fall, or only clicks? Impressions steady with fewer clicks usually means a result page changed around you, not that you were demoted. We covered how to read that split in our Search Console piece on AI Overviews.

Did you fall a few places, or out of the top results? Position 2 to 4 is ordinary movement. Position 4 to 29 across a wide set of queries is the pattern worth a real investigation.

If the whole site slid together, that points at a site-level quality judgment. If three pages dropped and the rest held, it points at those three pages. The fixes are different, so the diagnosis comes first.

Google rewrote its AI-content guidance a week into the rollout

On October 1, Google's documentation changelog logged an update to its guidance on using generative AI content. The stated reason: "To get our documentation in sync with our presentations we use at our developer events." The changelog says the update adds information from the Search Quality Raters guidelines.

Google does not connect that edit to the spam update. We won't either. But the page it edited is the clearest statement Google has made about what an AI-assisted site owes its readers, and it is worth reading in full.

It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing.

— Google Search Central, guidance on using generative AI content

The guidance page says four things that matter to a business owner.

First, using AI tools to generate many pages without adding value for users may violate the spam policy on scaled content abuse. The test is value to the reader. Nothing on the page bans the tool, and a human copy-pasting the same thin page 200 times fails it just as badly.

Second, AI models predict likely wording instead of retrieving facts, so outputs can contain inaccuracies. That is the reason Google gives for the human review.

Third, the review is not only for body copy. Google says it also applies to title elements, meta descriptions, structured data and image alt text.

Fourth, if you generate content automatically, consider telling readers how it was created.

None of that bans AI. It describes a process: draft with a tool, verify with a person.

Small sites forget the fields no one rereads

When an owner thinks "AI content," they picture blog posts. Google's own list is wider, and the wider list is where small businesses are exposed.

Meta titles and descriptions. Plenty of site-builder plugins and AI tools bulk-generate these. Two hundred pages with near-identical descriptions that promise things the page does not deliver is exactly the many-pages-no-value pattern.

Structured data. Review markup, price, availability and business hours generated by a plugin should match the page. A schema block that claims a four-star rating your page never shows is a trust problem, not a technicality.

Image alt text. Auto-captions that describe a "smiling person holding a phone" on a plumbing page add nothing for the reader.

Location and service pages. A page for every suburb, each swapping only the place name, is the oldest scaled-content move there is, and AI makes it cheap to produce 80 of them in an afternoon. We cover how to do local pages that earn their place in our local SEO guide.

Product descriptions. On a store, the guidance adds a stricter rule. It says Google Merchant Center has its own policies for AI-generated content, and that AI-generated images must carry the IPTC DigitalSourceType TrainedAlgorithmicMedia metadata. AI-generated product titles and descriptions must be specified separately and labeled as AI-generated.

That last item is the one most stores will miss, because the metadata is added at the file level, not on the page.

Every weak page ends up in one of three piles

Once you have the list, the decision on each page is simpler than the spreadsheet makes it look.

Keep. The page answers a real question a customer asks, the facts are right, and someone on your team could add one detail only you know: a price range from last month's jobs, a photo from your own van, a name. Fix the metadata and move on.

Rewrite. The topic is right and the page is generic. This is where an AI draft earns its keep: give it your real details, then have a person check every line against what you actually do. A rewrite takes an hour. Doing it to 200 pages at once is how you recreate the original problem.

Cut or merge. Three pages that answer the same question in slightly different words become one page, with redirects from the other two. A suburb page with no local detail becomes a section on a broader service page. Merging keeps the one useful thing the pages had and removes the pile.

Keep a note of what you changed and when. If traffic moves in November, you want to know which of your edits it followed.

A page is safe when a human can defend every claim on it

Here is our operator test, and it is ours, not Google's. For each page on your site, can the person who owns the business point to where each claim came from?

Prices, hours, service areas, certifications, "serving the area since 2009," a stat in a blog post. If the answer is "the AI wrote it and I skimmed it," that page is a liability whether or not this update touches it.

We run our own blog under that rule. Every post here is drafted with AI help, and none ships until each stat is checked against the source it came from and linked in the text. If a number has no primary source, it does not go in. That is slower than pressing publish on twenty drafts. It is also the only version of AI content we would put our name on for a client.

We have no client data yet on how this specific update sorted AI-assisted sites. It is ten days old and unfinished. If a vendor already claims a pattern, ask to see the sample size. The honest read right now is: Google has not told us what it caught, and we are not going to pretend we know.

If you want the broader reasoning on what gets sites demoted, our earlier piece on why AI-generated blog content gets deindexed holds up. The constructive side, building pages that rank and get cited, is in how to build a blog that ranks and gets cited by LLMs.

What to do this week, in order

1. Do nothing to your top pages until the dashboard marks the update complete. Our rule: leave pages that are performing alone. The rollout is not finished and a rewrite now muddies the diagnosis.
2. Export your pages and sort them by traffic. Put your top 20 in one list and your bottom 100 in another.
3. Read the top 20 as a stranger would. Fix wrong facts first. Wrong facts are the one thing no one disputes.
4. Check the bottom 100 for sameness. Near-identical descriptions, templated location pages and thin product text are the candidates to merge, rewrite or remove.
5. Audit the hidden fields. Pull your meta descriptions, schema output and alt text into a spreadsheet and read them in one sitting.
6. Write down how you publish. One line: who drafts, who checks, where sources live. If that line does not exist, you have a content process problem, not a Google problem.
7. Leave the rollout alone, then judge. Re-check Search Console a week after Google marks it finished.

Most owners can do steps 1 to 3 themselves in an afternoon. Steps 4 to 6 are where the hours go: hundreds of pages, a plugin nobody remembers installing, and a decision on each one. That is the part we do for people, and it is the reason our AI content service starts with an audit instead of a content calendar.