In short
Disclosure is not required by Google and, for most ordinary business content, not yet required by law, though the EU AI Act and several sector regulators require labelling of synthetic media in specific contexts from 2026. Disclose when a reasonable reader would feel misled to learn how the content was made: fully generated articles, synthetic images of people or places, AI-written reviews or testimonials. Do not clutter pages with notices for editing help. The durable answer is a published policy naming an accountable human author on every piece.
The three parties with an opinion
Google does not require disclosure. Its guidance says content is judged on quality regardless of how it was produced, and that it may be appropriate to add a note where readers would reasonably wonder, such as when AI was a meaningful part of the process. It rewards a clear author and evidence of expertise; it does not scan for AI notices.
Regulators are moving. The EU AI Act requires that AI-generated text published to inform the public on matters of public interest be labelled unless it has undergone human review with editorial responsibility, and that synthetic images, audio and video be marked as such; the obligations phase in from 2026. Advertising standards bodies in several countries treat undisclosed AI-generated reviews and endorsements as misleading. Sector rules in finance, health and legal services increasingly require that advice be attributable to a qualified person. None of this yet covers the ordinary business blog post drafted with help, but the direction is clear.
Readers care about being misled, not about tools. Research on disclosure finds that labelling content as AI-generated reduces trust somewhat, that discovering undisclosed AI use reduces it far more, and that a named human author who takes responsibility restores most of it. The asymmetry is the point: the cost of disclosing is small; the cost of being caught not disclosing is large.
Assisted and generated are different things
Most of the confusion comes from treating one question as two. A piece where a person decided what to say, drafted or edited with AI help, checked every claim and put their name to it is assisted. A piece where a prompt produced the text and it was published with a glance is generated. The first needs no more disclosure than the use of a word processor. The second is the case regulators, platforms and readers are concerned with, and the honest answer is usually not to publish it at all rather than to label it.
We write about how AI-written content ranks elsewhere; the short version is that the generated kind rarely does, which makes its disclosure a somewhat theoretical problem.
When to disclose
Always: - Synthetic images, video or audio depicting people, places or events that look real, including AI-generated imagery presented as photography. - AI-generated reviews, testimonials or quotes attributed to people. - Content in regulated domains where a reader relies on it as professional advice. - Anywhere a law or platform policy requires it.
When a reasonable reader would want to know: - Articles that are substantially generated with light human review. - Chatbots and support agents, so the person knows they are not talking to a human. - Translations produced by machine without a human check, in contexts where accuracy matters.
Not needed: - Editing, restructuring, summarising a source the author read, headline variants, grammar. - Research assistance where every claim was verified against the source. - Code, unless the reader is relying on it being reviewed.
What a good disclosure looks like
Specific, brief, and attached to accountability rather than apology. "Drafted with AI assistance and edited, fact-checked and approved by [name]" says what happened and who stands behind it. "This article was generated by AI" without a name says nobody does, and that is the reading that costs trust. For images: "Illustration generated with AI" in the caption. For a chatbot: a line at the start of the conversation.
A site-wide editorial policy page, linked from the footer and from author bylines, is the cleanest way to handle the general case: it explains how content is produced, what AI is used for, what it is not used for, and who is responsible. Then per-piece notices are only needed for the exceptions.
A three-tier policy you can publish
- Human-led, AI-assisted (the default): a named author decides the content, AI may help draft or edit, every fact is verified, no notice on the page; covered by the policy.
- AI-generated, human-reviewed: text substantially produced by AI and reviewed by a named editor; a short notice on the page naming the editor.
- Synthetic media: any generated image, audio or video that could be mistaken for a record of reality; labelled in the caption or on the item.
This maps onto what regulators are asking for, satisfies the reasonable-reader test, and keeps notices off the pages where they would be noise.
Why this matters for search and AI visibility
Search engines and AI assistants both weigh author authority and accountability. A page with a real author, a real bio and a clear editorial policy is a page that systems designed to detect trustworthiness can trust. A page with no author and a vague AI notice is a page they have every reason to discount. Disclosure done as accountability improves a site's standing; disclosure done as a disclaimer does the opposite.
Our own practice is the first tier: named authors, AI as a drafting and editing tool, every claim checked, and a policy that says so. If you would like help writing an editorial policy that is honest about your process and defensible under the rules arriving now, book a call.
Common questions
Is it illegal to not disclose AI-generated content?
For most ordinary business content, not yet. The EU AI Act requires labelling of AI-generated text on matters of public interest unless it has human editorial review, and of synthetic images, audio and video, with obligations phasing in from 2026; advertising rules in many countries treat undisclosed AI reviews and endorsements as misleading. Check the rules for your sector and market.
Does Google require AI content disclosure?
No. Google judges content on quality and helpfulness regardless of how it was produced, and does not scan for disclosure notices. It suggests a note where readers would reasonably wonder, and it strongly rewards a named author with visible expertise, which is a more effective form of accountability than a disclaimer.
Does disclosing AI use hurt trust?
Slightly, according to research on labelled content, but far less than readers discovering undisclosed AI use afterwards, and a named human author who takes responsibility restores most of the difference. Disclosure phrased as accountability, naming who reviewed and approved the piece, performs much better than a bare label.
Do I need to disclose using AI to edit or proofread?
No. Editing, restructuring, grammar, summarising sources the author has read, and generating headline options are assistance, comparable to a spellchecker or a thesaurus, and no reasonable reader expects a notice. Disclosure becomes relevant when AI substantially produced the content or when synthetic media could be mistaken for real.
How should I disclose AI-generated images?
With a short label in the caption or alongside the image, such as 'Illustration generated with AI', and never by presenting a generated image as a photograph of real people, places or products. The EU AI Act requires marking of synthetic images, and platforms increasingly detect and label them; disclosing yourself is both compliant and more credible.
