How should you disclose AI assistance on a page? What Google asks and readers expect

To disclose AI assistance on a page, add a short note saying what role AI played, what a person checked and when, and who answers for the page. That wording is our own: Google asks only that you consider a note when readers would reasonably expect one, and we found no statement that Search requires it.

• Google's people-first content guidance suggests considering AI or automation disclosures when it would be reasonably expected (Google, Search Central, read 30 October 2026).
• Merchant Center says AI-generated images must carry the IPTC DigitalSourceType value TrainedAlgorithmicMedia, with the metadata left in place (Google, Merchant Center Help, read 30 October 2026).
• Google's Search Central Blog says giving AI an author byline is probably not the best way to tell readers about AI's role (Google, Search Central Blog, read 30 October 2026).
• In a preregistered survey experiment with 599 participants in German-speaking Switzerland, human-written, AI-assisted and AI-generated news excerpts were rated similarly on credibility, readability and expertise before disclosure (arXiv, read 30 October 2026).

What does Google say about disclosing AI use?

Google suggests considering a disclosure and does not frame it as a requirement for Search. Its people-first content guidance says AI or automation disclosures are useful where someone might wonder how the content was created, and adds Consider adding these when it would be reasonably expected. Its February 2023 blog post on AI-generated content says Appropriate use of AI or automation is not against our guidelines.

What counts is purpose and value, not whether AI was involved: using automation, including AI, primarily to manipulate rankings breaks Google's spam policies (see what Google's policies say about AI-generated content). We found no Google statement that a note changes how a page is judged or ranked, so treat a disclosure as information for readers, not protection.

Google's guidance offers Who, How and Why as a way to stay on course with what its systems seek to reward, and a disclosure answers How. For Who, it says to give accurate bylines where readers might expect them and calls fabricating creator profiles, such as with AI-generated headshots, a form of deception. The blog says Giving AI an author byline is probably not the best way to tell readers about AI's role, so, in our view, keep a real byline and put the AI note beside it.

Where does Google set a format for AI labels?

Google names formats for AI labels on images, product data and ads; for the text of an ordinary web page it only suggests considering a note.

Only the Merchant Center row is worded as a rule, and it covers product data, not ordinary page text. For other images, our reasoning is to keep the IPTC value in the file, since Google lists the values it supports. Google's AI label settings page adds that laws in some regions may require labeling ads that use certain AI-created or AI-edited assets, and advises seeking legal guidance.

What do readers expect from an AI disclosure?

No source we read measures what visitors to ordinary web pages expect; the nearest evidence is three arXiv papers about news readers, two of which draw on the same small experiment, and their findings point different ways. A survey experiment with 599 participants in German-speaking Switzerland found that, after disclosure, participants in the AI-assisted and AI-generated conditions were more willing than the control group to keep reading their article, while willingness to read AI-generated news in future did not differ across conditions. A study with 40 participants found trust fell only with detailed disclosures, though most preferred the detail. A workshop paper cites that study as an experiment with 34 participants and says detailed disclosures risk being scrolled past and one-line labels can leave readers searching for signs of AI involvement; the readers proposed designs such as detail on demand.

Our reading, on weak evidence: a one-line note that links to a fuller one avoids the detail some readers skip and still serves those who want it.

What should an AI disclosure say?

In our view, a disclosure should say three things: what role AI played, what was checked and when, and who answers for the page. No source defines a standard wording, so this is our reasoning.

1. The role AI played. Name the task and, if you can, the tool: spelling suggestions, an outline, a first draft or generated images. The more of the wording the tool wrote, the likelier a reader is to wonder how the page was made (our reading).
2. What was checked, and when. Google calls it critical to manually fact-check and review all AI-generated content before publishing, and does not say who does the checking. Say what a person checked and the date; if a tool did the checking, say so rather than imply a person did (see how to fact-check AI-written content against primary sources).
3. Who answers for the page. Name the author or the organization that publishes it, give a way to report an error, and say who approves it (see what a human should approve before AI-drafted content is published).

Keep the note in proportion. NIST's generative AI profile, a companion to a framework NIST describes as intended for voluntary use, includes a suggested action (MP-5.1-003) to consider disclosing use of generative AI to end users in relevant contexts, weighing the objective of disclosure, the context of use, the likelihood and magnitude of the risk, the audience and the frequency of disclosure.

Two samples in our own wording, with placeholders in brackets:

• One line: Drafted with an AI writing tool. [Name] checked every figure and quote against its source on [date] and answers for this page.
• Fuller note: An AI tool wrote the first draft from our outline. [Name] rewrote it and checked each figure, quote and link against the cited page on [date]. Report an error at [address].

Put the one-line version near the byline or at the foot and link it to a fuller note or a sitewide page on how you use AI; that layering is our reasoning, not a Google instruction.

How do you keep a disclosure true after the page changes?

Our reasoning: a disclosure stays true when it is written from a record of what was checked and carries a date that moves only when the check is repeated. Our methodology page describes the record Penlid keeps for each draft it checks: each checkable claim is verified individually against live sources and gets a stored record: verdict, confidence, source URL and the exact supporting quote. That is Penlid's own method, not an industry standard.

Date the note and keep a dated change line beside it, where you already record changes (see what a content operations ledger should record). The same methodology page says A methodology that quietly moves is worse than no methodology. Our reasoning, applied to a disclosure: a note that quietly goes stale is worse than none, because its date makes a claim about the text beside it, so check a claim again whenever an edit adds or changes it, and change the date only then.

Our own note is a worked example, and this guide carries it too. At the foot of every Penlid guide it says the guide was drafted with Claude, Anthropic's AI model, and that a second pass re-read every source and checked each number, quote and policy statement against it on a stated date. It does not say who ran the second pass, and its Who line names Penlid, not a person.