Google now tells sites to manually fact-check all AI-generated content before publishing, warning that generative models "predict a likely sequence of words" rather than retrieve facts. The company revised its guidance page on Oct. 1, adding three sentences to the "Focus on accuracy, quality, and relevance" section that make manual review an explicit, mandatory step.

The new wording sits right after the section's opening line. "Keep in mind that generative models don't retrieve facts, but predict a likely sequence of words based on their training data," the guidance reads. "Because of this, generative AI outputs may contain inaccuracies (also known as hallucinations). It is critical to manually factcheck and review all AI-generated content for accuracy and trustworthiness before publishing."

Google's documentation updates page says the change brings the guide in line with the company's talks at developer events. The Oct. 1 changelog entry states the generative AI content guide was updated "with information from the Search Quality Raters guidelines." It is the only page listed for Oct. 1.

Metadata falls under the same review

The update also tightens the metadata line. Where the previous version said "This includes metadata," the page now reads "This review also applies to metadata." The guidance still lists