SEO · 13 min read

Does Google penalise AI content? What the 2026 evidence shows for businesses scaling content

Google doesn't penalise AI content for being AI, but scaled, low-value pages get hit. What its policies, 2026 spam updates and ranking data show.

A pottery inspector holds a blue glazed jug up to window light beside a long bench of identical jugs
On this page — 14 sections
  1. TL;DR
  2. Table of Contents
  3. What does Google actually say about AI content?
  4. What is scaled content abuse?
  5. How do Google’s quality raters judge AI content?
  6. What actually gets hit in 2025 and 2026?
  7. What do the ranking studies show?
  8. Where is the line for a business scaling content?
  9. How Castle uses AI in its own content
  10. Our perspective
  11. Worried your content is more volume than value?
  12. FAQ
  13. Sources
  14. Recommended

Does Google penalise AI content? Not for being AI. Google’s spam policies go after pages made “for the primary purpose of manipulating search rankings”, however they were produced. The catch for anyone scaling up: in Ahrefs’ 2026 study, heavily AI-written pages earned a half to a third of the impressions that mostly human pages did.

The maths is tempting, we get it. A content tool and a free afternoon can turn out 300 service pages that once would have cost a small fortune in copywriting. Meanwhile Google has shipped four spam updates in 2026, against one in 2025, and started a November core update on the 10th.

Below: what Google’s own documents say, what’s been hit, what the ranking data can and can’t tell you, and how we use AI without betting the domain on it.

If you want more content without handing Google a reason to bury it, talk to us. Our SEO team plans and writes content built to rank and to be cited.

Explore our SEO services

TL;DR

  • Google doesn’t penalise content for being AI-written. Its scaled content abuse policy applies “no matter how it’s created”.
  • What gets hit is volume without value. AI just makes that cheap.
  • Google ran four spam updates between March and October 2026, and consultants reported a rise in manual actions this autumn.
  • In Ahrefs’ 2026 sample, fully AI pages could rank, but mostly human pages took 82.2% of top-three positions.
  • Google now calls it “critical” to fact-check AI content by hand before publishing, metadata included.
  • A hand rewrite won’t rescue a site full of thin AI pages. Google’s John Mueller says to treat it as “starting over”.

Table of Contents

What does Google actually say about AI content?

Google cares whether content helps people, not which tool wrote it. Its February 2023 guidance on AI-generated content said “Appropriate use of AI or automation is not against our guidelines”, while using it mainly to game rankings breaks the spam policies.

That position hasn’t softened, but it’s become more specific. Google’s page on using generative AI content, updated on 1 October 2026, says AI “can be particularly useful when researching a topic, and to add structure to original content”. It also warns that generating “many pages without adding value for users” may breach the scaled content abuse policy.

The October edit matters. Search Engine Journal spotted three new sentences explaining that models “predict a likely sequence of words” rather than look facts up. Google now says it’s “critical to manually factcheck and review all AI-generated content” before publishing, including title tags, meta descriptions, structured data and alt text.

Google’s guide to people-first content adds a “Who, How and Why” test. It “strongly” encourages accurate bylines and warns against deceptive authorship, so an invented author with an AI headshot is a liability. It suggests disclosing automation where readers would “reasonably” expect it. And it repeats that AI used “for the primary purpose of manipulating search rankings” is spam.

What is scaled content abuse?

Scaled content abuse is Google’s term for publishing lots of low-value pages to win search traffic. The spam policies define it as “when many pages are generated for the primary purpose of manipulating search rankings and not helping users”, offering “little to no value to users, no matter how it’s created.”

Google’s examples include:

  • Using generative AI “to generate many pages without adding value for users”
  • Scraping content and disguising it through “synonymizing, translating, or other obfuscation”
  • Stitching together other people’s pages without adding anything
  • Spreading content across several sites to hide how much of it there is

The policy arrived in March 2024, covering content made “whether automation, humans or a combination are involved”. Google later reported 45% less low-quality, unoriginal content in results. So a room of freelancers churning out 400 near-identical pages is as guilty as a script. Google doesn’t much mind whether the forgery was done with a quill or a printer.

Enforcement comes from “automated systems and, as needed, human review”, and offending sites “may rank lower in results or not appear in results at all.” Human review produces a manual action, and in Google’s Manual Actions report documentation, scaled content abuse sits under “Major spam problems”.

How do Google’s quality raters judge AI content?

Raters are told to judge effort and value, and to treat AI as a tool that cuts both ways. The September 2025 rater guidelines call generative AI “a helpful tool for content creation” that “can also be misused.”

If you’re planning to scale, three passages matter:

  • Thousands of pages made by running existing content through translation software, without “oversight, manual curation, etc.”, don’t count as having effort (section 3.2).
  • Scaled content abuse gets the Lowest rating “no matter how they are created” (section 4.6.5).
  • “The use of Generative AI tools alone does not determine the level of effort or Page Quality rating”, but it’s “possible to use Generative AI tools to create Lowest quality content” (section 4.6.6).

Raters don’t rank your pages, mind. Google’s Danny Sullivan has written that their ratings “don’t directly impact how a page or site appears in Search”. They test whether Google’s systems are working. Think of the guidelines as the marking scheme the algorithms are meant to pass.

Quality controller at a Sheffield cutlery works inspecting a finished knife under a bench lamp beside a row of blades
Like a quality controller in a cutlery works, Google judges the finished blade, not which machine ground it. AI-generated image.

What actually gets hit in 2025 and 2026?

Sites publishing at volume with nothing new to say are the ones in trouble, and enforcement has picked up. Here’s the evidence, labelled by how solid it is.

Spam updates (Google’s own record)

Google’s ranking update history lists one spam update in 2025 (from 26 August) and four in 2026: 24 March, 24 June, 18 August and 24 September. The September update “applies globally and to all languages” and, as Search Engine Journal noted, took nearly 14 days. Google hasn’t said which policy any of them targeted, so anyone claiming they were aimed at AI content is guessing.

Manual actions (consultant reports, not Google data)

In October 2026, Search Engine Roundtable reported consultants seeing more manual actions. Marie Haynes said requests for help “went from essentially no requests to several per week”, mostly thin content, pure spam and scaled content, adding that “most of these sites are legitimate long standing businesses.” It’s anecdotal, not UK-specific, and Google gave no figures. That last line should still make an established business pause.

What Google’s people have said

  • Danny Sullivan, April 2025: per Search Engine Journal’s notes, Google doesn’t “really care how you’re doing this scaled content, whether it’s AI, automation, or human beings.”
  • Gary Illyes, August 2025: he told Kenichi Suzuki that “human created” was the wrong phrase: “it should be human curated.”
  • John Mueller, November 2025: “just rewriting AI content by a human won’t change that, it won’t make it authentic,” he said of a site with thin AI pages.
  • John Mueller, October 2026: he likened mass-produced AI content to keyword stuffing and white text on white backgrounds, tricks the systems eventually catch.

What do the ranking studies show?

AI content isn’t penalised as a category, but heavily AI-written pages do worse on average. Both studies below are from Ahrefs, using its own detector, and neither says which country’s results it measured. Treat them as global, not UK, data.

July 2025. Across 600,000 pages ranking for 100,000 random keywords, the correlation between AI content and ranking position was 0.011, “effectively zero”. Only 13.5% of top-20 pages were purely human and 4.6% purely AI.

July 2026. The follow-up study found that 5.3% of top-ranking pages scored 100% AI, yet pages under 50% AI took 82.2% of top-three rankings. Low and moderate AI pages got two to three times the impressions of heavily AI ones, and were more likely to be indexed.

Share of sampled pages indexed by Google, by AI content level

Ahrefs, 2026. Scale: 0 to 100%.

Low AI content (under 20%): 49.28%

Very high AI content (80% or more): 40.35%

Source: Ahrefs, “Google Doesn’t Punish AI Content; It Punishes Bad Content” (July 2026). AI levels measured by Ahrefs’ own detector; market not stated.

Ahrefs is upfront about the limits: detection “is not perfect”, and heavy AI users may simply be newer, weaker sites. Its author reads the data as “increasing use of AI correlates with decreasing content quality”. We’d agree. Google punishes the shortcuts people take with the tool.

When the 2025 study landed, consultant Glenn Gabe posted a warning: “Please do not read this post and run to publish AI-generated content at scale.”

Hedge layer with a billhook in a frosty field, half the hawthorn hedge neatly laid and staked, half still overgrown
Fewer, properly finished pages beat a field of half-done work, whatever tools did the lifting. AI-generated image.

Where is the line for a business scaling content?

The line is whether each page gives a reader something the ten pages already ranking don’t. Volume with nothing behind it is the offence.

Lower risk: research and outlines; first drafts a specialist rewrites with real experience; summarising your own data, such as customer reviews (Sullivan’s own positive example); drafting metadata, then checking every line.

Higher risk: hundreds of “[service] in [town]” pages where only the town changes; machine translation nobody checks; reworded competitor articles; publishing straight from the tool; a made-up author.

Take an illustrative installer covering the North West. It could publish 300 location pages in a week, each with near-identical copy. Or it could publish 12, one per area it genuinely serves, each with real jobs, local planning quirks and the questions customers ask. The first looks busier in a monthly report. The second is what Google’s documents describe as worth ranking.

Sullivan’s motivation test is a decent filter: would you publish the page if search engines didn’t exist? If the honest answer is no, it’s a traffic play.

How Castle uses AI in its own content

We use AI throughout our content work, and the process is built around what Google checks for. Our setup runs on Claude, connected through MCP connectors to our research and SEO tools and our own software. The order goes like this:

  1. A real question first. Each article starts from keyword and AI search data, and a question customers actually ask.
  2. Research before writing. Every source is opened and checked. If a statistic can’t be verified at its original source, it’s cut.
  3. Drafting with rules. AI helps draft, but invented figures, quotes and case studies are banned. Where first-hand experience is needed and we don’t have it, the draft leaves a gap for a person to fill.
  4. An edit for substance. A separate pass swaps generic sentences for specifics, or deletes them.
  5. Checks against a written standard. Every post is checked against a house standard covering structure, sources, links and images, and anything that fails is rebuilt rather than waved through.
  6. Sources on show. Every article ends with its full source list, so you can check our working.

It’s slower than pressing “generate” 300 times. That’s rather the point. For the bigger picture, see our complete guide to AI SEO and GEO and our piece on earning LLM citations.

Our perspective

Asking whether Google penalises AI content is a bit like asking whether HMRC penalises spreadsheets. It doesn’t care what you used. It cares whether the numbers are honest.

The businesses that get burned over the next year won’t be the ones using AI. They’ll be the ones using it for more pages per pound instead of more research per page. Four spam updates in 2026, consultants reporting manual actions on long-standing businesses, Mueller’s “starting over”: it all points one way. The gap between cheap content and its consequences is closing.

There’s a second reason to care. The pages Google trusts are increasingly the ones feeding AI Overviews and getting cited by ChatGPT and Perplexity. Mass-produced pages rarely earn citations because they repeat what everyone else says. If your SEO strategy is already quietly failing, 500 more thin pages won’t fix it.

So use AI. We do, every day. Use it to raise the floor on research and accuracy, put a real name on the work, and publish fewer, better pages. That’s the version of AI content SEO we’d bet a client’s domain on, and it’s how our AI SEO team works. If you’ve already published at scale, an AI visibility audit is a sensible first check.

Worried your content is more volume than value?

If you’ve scaled content with AI and want an honest view of what’s helping and what’s a liability, we’ll take a look.

Require AI SEO, GEO, or digital marketing and AI automation within your business? Book a free appointment with a senior consultant.

Talk to Castle

FAQ

Can Google detect AI-generated content?

Google hasn’t said whether it runs an AI detector, and its policies don’t hinge on one. The spam policy targets low-value pages made to manipulate rankings “no matter how it’s created”. Third-party detectors are imperfect too: Ahrefs says its own detection suits large-scale trends rather than verdicts on individual pages.

Is AI content bad for SEO?

Not automatically. Ahrefs’ 2026 study found fully AI-written pages in Google’s top three. But pages under 50% AI held 82.2% of top-three positions and earned two to three times the impressions of heavily AI pages. AI content struggles when it replaces research and judgement instead of supporting them.

Will rewriting my AI content by hand fix a ranking drop?

Probably not on its own. John Mueller said in November 2025 that rewriting AI content by hand “won’t make it authentic”, and that a site built on thin pages should be approached “as essentially starting over”. Focus on what each page adds, and remove pages that add nothing.

What happens if Google gives me a manual action?

A manual action means a human reviewer at Google found pages breaking its spam policies. Affected pages can rank lower or drop out of results. It shows in Search Console’s Manual Actions report, where scaled content abuse falls under “Major spam problems”. Once everything is fixed, you can request a review.

Sources