Does Google penalise AI content? What its policies say, and how to use AI safely

Google put its position in writing in February 2023, in a Search Central post on AI-generated content. The core sentence is short: appropriate use of AI or automation is not against our guidelines. The post says Google rewards high-quality content however it is produced, that using AI does not give content any special gains, and that using automation, including AI, to generate content with the primary purpose of manipulating ranking in search results is a violation of its spam policies.

The current documentation says the same from both sides. Google's guidance on generative AI content, last updated on 1 October 2026, calls AI particularly useful when researching a topic and adding structure to original content, and warns that generating many pages with it without adding value for users may break the scaled content abuse policy. Its guide to helpful content repeats the warning about manipulating rankings. So the question Google asks of a page is not whether a model wrote it, but why it was made and whether it helps the person who lands on it.

If you use AI to draft anything, you may worry that one update could take your traffic with it. This section shows where Google draws the line, so you can tell which side your pages sit on. In short, the line runs around purpose and value, and it applies to people as much as to models.

Scaled content abuse became a named spam policy in March 2024. Google's announcement says it builds on the earlier policy about automatically generated content and applies no matter whether content is produced through automation, human efforts or a combination of both. The policy describes many pages generated for the primary purpose of manipulating search rankings and not helping users, and gives examples:

• using generative AI or similar tools to generate many pages without adding value for users;
• scraping feeds, search results or other content to generate pages, including through automated transformations such as swapping in synonyms or translating, where little value is provided;
• stitching together content from different pages without adding value;
• creating several sites to hide the scaled nature of the content;
• creating many pages that make little sense to a reader but contain search keywords.

The same update added two policies worth knowing if you publish at volume. Site reputation abuse covers third-party content published on an established site mainly to benefit from that site's ranking signals; Google said enforcement would start on 5 May 2024. Expired domain abuse covers buying an expired domain and repurposing it mainly to manipulate rankings with content of little value. If you build pages from templates and data, the guide to programmatic SEO shows how to tell a useful set from a scaled one.

Google acts on spam in two ways, and they look different from your side. Its spam policies say it detects violations through automated systems and, as needed, human review, and that sites that violate them may rank lower or not appear in results at all.

• A manual action is issued when a human reviewer at Google has decided that pages on a site do not comply with its spam policies. You are told in the Manual actions report in Search Console and by a Search Console message. The types closest to AI-written content are Thin content with little or no added value (thin affiliate pages, scraped content, low-quality guest posts, doorways), Major spam problems (aggressive techniques such as scaled content abuse and cloaking) and Site reputation policy. An action can cover part of a site or all of it.
• Automated systems, including SpamBrain, which Google describes as its AI-based spam-prevention system, act on their own, and what they do never appears in the Manual actions report. Google's help page says that in most cases spam is discovered automatically. Its page on spam updates says a site affected by one should review the spam policies, and that changes may help once its systems learn, over a period of months, that the site complies.

So check the Manual actions report first. If it is clear, there is nothing to appeal: a drop came from automated ranking or spam systems, from competitors or from changes on your own site, and the fix is in your pages.

Google Search does not require a label on AI-assisted articles. Its 2023 post says disclosures are useful where someone might wonder how a piece was created, and its guide to helpful content turns that into three questions about your content: Who, How and Why.

• Who: accurate bylines where readers would expect them, leading to information about the people behind the content. Google's post adds that giving AI an author byline is probably not the best way to make its part in the work clear.
• How: is the use of automation, including AI-generation, self-evident to visitors, and do you give background on how it was used? One sentence often does it, such as a note that a comparison was drafted with an AI model from your own test results and checked by an editor.
• Why: Google calls this perhaps the most important question. Content made to help people who come to your site passes it; content made primarily to attract search engine visits does not.

Shops have firmer rules. Google's generative AI guidance points to Google Merchant Center: AI-generated images need IPTC DigitalSourceType metadata set to TrainedAlgorithmicMedia, and AI-generated product titles and descriptions must be specified separately and labelled as AI-generated.

Google's guidance gives the reason in one line: generative models don't retrieve facts, but predict a likely sequence of words based on their training data. A draft can be fluent and wrong in the same sentence, and nothing in it will warn you. Google calls it critical to manually fact-check and review all AI-generated content for accuracy and trustworthiness before publishing, and extends that review to titles, meta descriptions, structured data and image alt text, since all of them can appear in search results.

• List every checkable claim: numbers, dates, names, prices, quotes, rules, and anything said about what a company, product or law does.
• Check each one against a primary source, such as the official page, the regulation or the product's documentation, and keep the link. If you cannot find a source, cut the claim.
• Open every link the draft cites and confirm the page exists and says what the sentence claims.
• Read the metadata and markup: does the title promise what the page gives, and does the structured data describe what visitors can see? Marking up invisible, irrelevant or misleading content has a manual action type of its own.
• Add what only you can add: your data, your experience, your tested result. Google's helpful-content questions ask whether you are mainly summarising what others say without adding much value, and a well-checked summary can still fail that test.

Say you write the blog for an accountancy practice and ask a model for a post on filing deadlines. The draft gives a late-filing penalty and a deadline with complete confidence, but the penalty comes from an older rule and the deadline applies only to paper returns. Ten minutes on the tax authority's own pages catches both, before a wrong figure reaches the clients you most want to impress. Which steps can run unattended and which need a person is covered in the guide to SEO automation.

Google's written position is clear. The mechanics behind it are not published, so be wary of anyone who claims to know them.

• No threshold. The scaled content abuse policy speaks of many pages and gives no number, no publishing rate and no share of AI text that tips a site over the line.
• No statement on detecting AI text. Google's 2023 post says its systems analyse patterns and signals to identify spam however it is produced. It does not say whether Google identifies AI-written text, or whether it uses that as a ranking signal.
• Studies are estimates. Any study that sorts ranking pages into AI-written and human-written has to guess which is which, usually with a detection tool, so treat its findings as hints rather than rules.
• Raters are not rankings. Google points to its Search Quality Rater Guidelines for how raters judge main content created with little to no effort, and says their ratings don't directly influence ranking.

Google's own words give you the safe uses: researching a topic and adding structure to original content. The rest follows from the policies above.

• Start from something you have that others do not, such as data, tests, cases or a process you run, and let the model draft around it.
• Publish a page because a reader needs it, not because a template can produce it; if two pages would say the same thing, publish one.
• Keep a person between the draft and the live page who checks every claim, link and title.
• Change a page's date only when its content has substantially changed. Google's helpful-content questions list changing dates to make pages seem fresh as a warning sign.
• Publish in batches and watch Search Console: the Page indexing report shows how much of each batch Google indexes, and the Manual actions report should stay clear.

If you want help with the volume, AutoPSEO's content engine does these steps for you. It takes topics from searches your site already appears for in Search Console, sets aside those your pages already cover, writes in your site's own format, and by default waits for you to approve every article before it goes live. The writing documentation lists the checks each draft passes and what you approve.