Google's August 2026 Spam Update, and What It Actually Means for Agency SEO

Google rolled out a spam update starting 18 August 2026. Per Google's own Search Status Dashboard it ran for two days and sixteen hours, completing on 21 August.

Within a week the usual thing happened. Half the SEO commentary called it the AI content update. Sites that had lost traffic announced that Google had finally cracked down on AI writing. Sites that had gained announced that their human writers had been vindicated.

Almost none of that is what the update was, and getting it wrong leads people to the wrong remedy. Here is our read, and we will take a position rather than list both sides.

What Google actually shipped

It was a spam update, which is a specific category in Google's own taxonomy and distinct from a core update. It shipped with no new spam policies attached, which means Google was enforcing existing rules more effectively rather than introducing new ones.

The rule it enforced is scaled content abuse. Google's spam policies define it directly:

"Scaled content abuse is when many pages are generated for the primary purpose of manipulating search rankings and not helping users."

The examples Google lists under that heading include:

"Using generative AI tools or other similar tools to generate many pages without adding value for users"

and

"Scraping feeds, search results, or other content to generate many pages (including through automated transformations like synonymizing, translating, or other obfuscation techniques), where little value is provided to users"

Read the first example carefully, because the whole argument sits inside it. The object of the sentence is many pages without adding value. Generative AI is named as one way of producing them. It is not the thing being prohibited.

Three things that follow, and all three matter

1. Using AI to write is not penalised. Using AI to mass-produce interchangeable pages is.

Google has been consistent on this for years. Its position is that it rewards helpful content regardless of how it was produced, and it acts against content produced primarily to manipulate rankings. The August update did not change that position. It made the enforcement of it bite harder.

The practical implication is that "we need to stop using AI" is the wrong response to a traffic drop. So is "we need to have a human edit it." Neither addresses what was actually flagged.

2. The method of production is irrelevant. The outcome is what is judged.

A hand-written thin city page with 400 words of generic copy and the town name swapped in is caught by the same policy, for the same reason, as an AI generated one. Human authorship is not a defence. Nor is a better prompt, a longer output, or a fact-checking pass.

This is the part agencies resist most, because a lot of SEO service delivery over the last decade has been built on producing volume. The volume was the product. If the pages were always thin, the fact that a person typed them does not make them safe now.

3. The unit of judgement is the pattern, not the page.

This is the single most important structural point and it is the one most people miss. Google evaluates groups of pages for repeated structure and functionally identical content. A page that reads perfectly well in isolation can still be caught as one of forty near-identical siblings.

That means a page-by-page audit will not find the problem. You have to look at your location pages, or your service pages, or your programmatic set, side by side, and ask whether they are genuinely different documents or one document wearing forty hats.

The related trap is doorway abuse: sets of pages aimed at different towns or regions that all funnel into the same place, with no real browseable hierarchy behind them. That pattern is what a lot of local SEO packages actually are.

Our position

We think the update was correct and overdue, and we think most of the agency industry has been describing it dishonestly because the honest description is commercially inconvenient.

Selling two hundred location pages is easy. Explaining why the client should buy ten real ones instead is a harder conversation, because ten sounds like less work and the client is used to counting deliverables. But two hundred near-identical pages were never an asset. They were a liability that had not been collected on yet, and in August a lot of it got collected.

We will say the thing agencies do not like saying: if your content programme stops working when Google enforces the scaled content abuse policy properly, your content programme was the thing the policy was written about. That is not bad luck and it is not Google being unfair to small businesses. It belongs on the list in why most lead generation agencies fail, alongside every other deliverable that was easy to count and hard to justify.

If your content programme stops working when Google enforces the scaled content abuse policy properly, your content programme was the thing the policy was written about.

The test we apply before creating any page

We use one question, and it is deliberately hard to pass:

Can you write two genuinely specific paragraphs about this exact topic, for this exact place or audience, that could not be true of any competitor or any neighbouring town?

If yes, the page deserves to exist, and the rest of the writing is execution. If no, the page will be thin no matter how many words you put on it, and adding words makes it worse rather than better.

For a niche like manufactured housing, where we set out the approach in SEO for mobile home investors, passing is easier than it sounds, because the operators genuinely know things the wider internet does not.

For a location page, passing means naming things that are true of that place and nowhere else. The specific corridor, the named communities, the county level process that actually differs, the local economic driver and how it changes this market in particular, the questions customers from that area actually ask on the phone. Not the city name swapped in. Not synonyms. Not a map embed.

For a blog post, passing usually means the source material came from the business rather than from the internet. The operator's own process, the constraints they work under including the inconvenient ones, the objection they have answered a thousand times and never written down, the measurement they took on their own accounts. Use AI to organise and express that knowledge. Never use AI to invent it. If you cannot source a claim, it does not go on the page.

That is also why the most defensible content any business can publish is its own operational data. Nobody else has it, a model cannot generate it, and it clears the test by definition.

What we would actually change

If you are reviewing a content programme after this update, in order:

1. Group your pages and look at them together. Location pages next to location pages, service pages next to service pages. Judge the set, not the page.
2. Measure uniqueness against siblings, not against the web. Shared navigation, footers and standard FAQ blocks do not count against you. The body does. Aim for a substantial majority of the body being genuinely different writing.
3. Fix, do not delete. A thin page that has links and history is worth more improved than removed. If a page truly has to go, redirect it to the closest genuinely equivalent page rather than deleting it or adding a noindex tag.
4. Stop publishing in large batches. A sudden mass of new near-identical URLs is itself a pattern associated with scaled content abuse. Spread substantial content out.
5. Re-source the thin pages from the business. Most thin pages are thin because nobody interviewed the operator. That is a fixable problem and it is the highest return work available.

What this has to do with AI search

One connected point, because the two questions get asked together. The same properties that make content survive a spam update, genuine specificity and information that exists nowhere else, are the properties that get a page retrieved and cited by AI assistants. We cover the technical side of that in getting cited by ChatGPT, Claude and Perplexity.

There is no version of this where the durable answer is more pages. It has been fewer and better for a while, and August made that expensive to ignore.

Sources: Google Search spam policies, scaled content abuse; Google Search Status Dashboard, ranking updates history; Google Search and AI-generated content; Google Search Essentials, creating helpful content

Keep reading

• Getting cited by ChatGPT, Claude and Perplexity
• SEO for mobile home investors
• Why most lead generation agencies fail