Does Google penalize AI content? What Google actually says

Two details in that list deserve attention. The 2024 policy was written to cover people as well as models: Google replaced its old “automatically generated content” policy with one about scale and value. And the rater guidelines, which train the people who evaluate search results for Google, say directly that “the use of Generative AI tools alone does not determine the level of effort or Page Quality rating.” Google also notes that rater scores “don’t directly influence ranking.”

The spam policy gives examples. Paraphrased, they are:

• using generative AI or similar tools to produce many pages without adding value for users,
• scraping feeds or search results, or running them through synonym swaps, to make many pages,
• stitching together content from other pages without adding anything,
• spreading the same thin content across several sites to hide the scale,
• pages stuffed with search keywords that make little sense to a reader.

Every example is about volume plus missing value. None is about who or what typed the words. A team that writes 400 city pages by hand from one template is closer to this policy than a team that drafts one well-researched guide with a model and checks every fact.

The same page now also counts “attempting to manipulate generative AI responses in Google Search” as spam, and it keeps two related policies you should know if you buy or host content: expired domain abuse (buying an old domain “primarily to manipulate search rankings”) and the site reputation policy (content placed on a host site “mainly because of that host’s already-established ranking signals”).

We found no public statement from Google saying it uses AI-text detectors for ranking, and none saying it doesn’t. What Google has said is narrower. Its 2023 FAQ says its spam systems “analyze patterns and signals to help us identify spam content, however it is produced.” The rater guidelines tell raters to give scaled content the Lowest rating “even if you are unsure of the method of creation”, which suggests authorship isn’t the thing being judged.

The best outside evidence points the same way. In July 2025 Ahrefs checked about 600,000 pages ranking for 100,000 random keywords with its own AI detector and found a correlation of 0.011 between a page’s AI share and its position, “effectively zero”. Most top-ranking pages, 86.5%, contained some AI-generated text. Two caveats: it’s a correlation, and it relies on a detector Ahrefs itself calls imperfect. Our guide on how AI detectors work explains why any detector score is a signal and not proof.

Google’s people-first page is a list of questions. The ones below are taken from it and from the generative AI page, rewritten as checks you can run on a page today:

Google’s May 2026 guide for its AI search features says the same thing in other words: don’t recycle what others have said or what “could easily be produced by a generative AI model.”

IntactVoice is a tool for drafts that a person or team will stand behind: it rewrites toward a writer’s own habits and checks the facts against the source. It is not a way to produce pages at volume. Our Acceptable Use Policy prohibits using the service to create spam, and to break the content policies of the platform where the output is published, including its rules on AI-generated content. Google’s spam policies are those rules for search. We also don’t promise any search outcome: no tool can, and our terms say so.

• Google doesn’t publish how its ranking systems weigh any signal, so “what moves rankings” is Google’s guidance, not a measured formula.
• The Ahrefs study is one company’s correlation analysis with its own detector, published in July 2025.
• Google ran spam updates in March, June, August and September 2026, according to its status dashboard. We didn’t find AI-specific wording in them, but we didn’t review every announcement in full.
• These pages change. Check the “last updated” date on each before you rely on a quote.

Open Search Console, go to Security & Manual Actions, then Manual actions, and confirm the report is empty. Then take your ten pages with the most impressions and run the checklist above on each. A page that fails three or more items is a better use of an afternoon than any new page.