Google has updated its direction for sites that use generative AI, adding a alert that AI output may merge errors and direction to fact-check it by hand before publishing.
Google revised the direction page on Oct. 1. Its documentation updates page says the alter brings the guide in row alongside the company’s talks at developer events.
The leaf now extends the manual assessment to metadata, so teams using AI to compose titles or depiction alt content are covered too.
What Google Added
Google’s new wording goes in the direction page’s “Focus on accuracy, quality, and relevance” section, correct following the section’s beginning line. Compared alongside the Dec. 10, 2025, version, three sentences were added:
“Keep in intellect that generative models don’t recover facts, but foretell a apt sequence of words according to their training data. Because of this, generative AI outputs may merge inaccuracies (also known as hallucinations). It is crucial to manually factcheck and assessment all AI-generated satisfied for accuracy and trustworthiness before publishing.”
In the metadata row that followed, “This includes metadata” became “This assessment additionally applies to metadata.” It motionless lists <title> elements, meta clarification elements, organized data, and substitute texts for images, “which can appear in Search results.”
How We Got Here
Google’s Oct. 1 changelog entry says the generative AI satisfied guide was updated “with data from the Search Quality Raters guidelines.” It’s the lone leaf listed for Oct. 1.
Googlers have recommended fact-checking AI output in community before. I covered an April 2024 Search Off the Record episode in which Gary Illyes stated generative AI output continually have to be fact-checked. In an August 2025 interview, Illyes stated AI satisfied “should be individual curated,” alongside person checking that it’s accurate, as Roger Montti reported.
What Stays The Same
The remainder of Google’s AI satisfied direction leaf matches the Dec. 10, 2025, version. It motionless opens by saying generative AI can assistance alongside investigation and alongside adding construction to first content. The identical paragraph warns that using AI tools to create many pages without adding value for users may breach the company’s spam guideline on scaled satisfied abuse.
The leaf additionally motionless points to sections 4.6.5 and 4.6.6 of the Search Quality Raters guidelines, which shield scaled satisfied abuse and satisfied made alongside small effort, originality, or added value. It says raters use those guidelines to measure Google’s ranking systems and that their ratings “don’t immediately power ranking.”
Why This Matters
The Dec. 10, 2025, type of Google’s direction asked for accuracy but had no manual fact-check education for AI output. The Oct. 1 update adds that stage and gives Google’s logic for it.
The fact-check stage reaches metadata as fine as part text. AI-generated metadata, specified as heading tags, falls under the identical manual assessment as drafts, so a procedure built lone for part copy misses part of what the direction now covers.
Looking Ahead
Teams that use AI in satisfied manufacturing can now citation Google’s own wording whenever they set a assessment guideline covering drafts and metadata. The business logs important changes to its Search docs on the records updates page, which is the location to observe for additional edits.
Featured Image: Golden Dayz/Shutterstock. Google wordmark: Source: Google.
