Google revised its generative AI content guidance on Oct. 1, adding a manual fact-check requirement that did not exist in the Dec. 10, 2025 version of the page. Three new sentences now sit in the document's "Focus on accuracy, quality, and relevance" section, right after the opening line.

Google now states:

"Keep in mind that generative models don't retrieve facts, but predict a likely sequence of words based on their training data. 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."

What changed in the metadata line?

The previous wording told publishers that "This includes metadata." The Oct. 1 version replaces it with: "This review also applies to metadata."

The list of covered metadata types stayed the same: <title> elements, meta description elements, structured data, and image alternate texts, "which can appear in Search results."

The expansion matters because many content operations run a fact-check process only on article drafts. Under the revised language, the same manual review has to extend to AI-generated title tags, schema, and alt text, or the policy is incomplete.

Why did Google add it now?

The Oct. 1 changelog entry attributes the change to the Search Quality Raters guidelines. The generative AI guide is the only page Google listed for that date.

Google employees have made the case publicly before. In an April 2024 Search Off the Record episode, Gary Illyes said generative AI output always has to be fact-checked. In an August 2025 interview, Illyes said AI content "should be human curated," with someone checking that it is accurate.

The Oct. 1 update is the first time Google has written that instruction into its own content guide.

What the guide still says

The rest of the Oct. 1 page matches the Dec. 10, 2025 version. It opens by saying generative AI can help with research and with adding structure to original content.

The same paragraph warns that using AI tools to mass-produce pages without adding value for users may violate Google's spam policy on scaled content abuse.

The document still points readers to sections 4.6.5 and 4.6.6 of the Search Quality Raters guidelines, which cover scaled content abuse and content made with little effort, originality, or added value. The page notes that raters use those guidelines to evaluate Google's ranking systems and that their ratings "don't directly influence ranking."

Why this matters operationally

The Dec. 10, 2025 version asked for accuracy but stopped short of a manual fact-check step. The Oct. 1 version adds that step and gives Google's reason for it in the same paragraph: generative models predict words, they don't retrieve facts.

That distinction is the basis for the new requirement. It also gives publishers a quotable line they can attach to a review policy, an SLA, or an editorial brief.

Teams that outsource metadata generation to AI tools now need a review path for title tags and alt text on top of the article workflow.

What to watch next

The documentation updates page is the public record Google uses for changes to its Search documentation. Any further edits to the AI content guide will appear there first.

Publishers running AI-heavy production should also track whether Google attaches enforcement language to the fact-check step in a future spam policy update, and whether the Search Quality Raters guidelines add rating examples that target AI-only metadata.