https://baryreview.com/blog/should-you-disclose-ai-assisted-content
baryreview

Blog · 2026-10-15 · baryreview editors

Should you label AI-assisted blog posts? A practical disclosure policy

The question nobody wants to answer in writing

If your blog posts start as AI drafts, you have probably had this conversation: should we say so? One person thinks labeling is honest and readers will respect it. Another thinks it’s an invitation to be dismissed, and besides, nobody labels their spellchecker. Both have a point, and the conversation usually ends without an AI content disclosure policy, so the decision gets made by default by whoever hits publish.

This article tries to settle it practically. Not “is AI writing ethical” but: what disclosure means, where it is actually required, what readers seem to think, and what a small business should write down and follow. Wording is at the end.

First, be clear what you’d be disclosing

“AI-assisted” and “AI-generated” are different things, and most of the confusion in this debate comes from using them interchangeably.

AI-generated means the tool produced the text and a person did little more than approve it. The ideas, structure, claims, and sentences are the model’s. Light edits for typos don’t change the category.

AI-assisted means a person is responsible for the substance and used the tool for part of the work: a first draft that was then substantially revised, an outline, a rewrite of a paragraph they’d already written, research summaries they then verified, headline options. The person could defend every claim in the finished piece because they checked it.

Then there’s the long tail nobody expects you to disclose: grammar checking, spellcheck, a thesaurus, autocomplete. Medium’s content policy, for instance, explicitly exempts AI-assisted grammar and spell checking from its labeling rule.

The useful line is not “did an AI touch this” but “who is accountable for what it says.” If the answer is “a person, who checked it,” you are in assisted territory. If it’s “the tool, we just posted it,” you are in generated territory, which is where most platform rules and most reader distrust are aimed.

Where disclosure is actually required or expected

For a typical small-business blog on your own domain, in most cases no law or platform requires you to label AI-assisted writing. But the picture is not blank, and it has moved in the last two years. Here is what we could verify as of this writing. Policies change; check the linked pages before relying on them.

Google Search. Google does not require disclosure, and its guidance since 2023 has been that it rewards “high-quality content, however it is produced” (Google Search Central, Feb 2023). But its guidance on creating helpful content lists questions it suggests you ask yourself, including: “Is the use of automation, including AI-generation, self-evident to visitors through disclosures or in other ways?” and “Is it self-evident to your visitors who authored your content?” It adds that disclosures “are useful for content where someone might think ‘How was this created?’” and that of the E-E-A-T qualities, “trust is most important” (Google, creating helpful content). Separately, Google’s spam policies name “using generative AI tools or other similar tools to generate many pages without adding value for users” as an example of scaled content abuse (Google spam policies). So: not required, but expected where a reader would wonder, and volume without value is a policy violation regardless of labels.

Medium. If you republish on Medium, its AI content policy requires that stories incorporating AI assistance be clearly labeled, and undisclosed AI-assisted text gets its distribution limited to the author’s own network. Fully AI-generated writing is not eligible for the paywalled Partner Program whether disclosed or not. Grammar and spell checking are exempt (Medium Help Center, AI content policy).

YouTube. Relevant if you also publish video. YouTube requires creators to disclose realistic altered or synthetic content: a real person appearing to say something they didn’t, altered footage of real events, realistic scenes that never happened. AI help with scripts, outlines, titles, or captions is explicitly exempt. Repeated failure to disclose can lead to removal or suspension from the Partner Program (YouTube Help, disclosing altered or synthetic content). The rule is about deception, not about whether a tool helped you write.

The EU AI Act. If you publish for an EU audience, Article 50 applies from 2 August 2026. It requires deployers who publish AI-generated or manipulated text “for the purpose of informing the public on matters of public interest” to disclose that it was artificially generated. There is an important exception: disclosure is not required where the text has undergone human review or editorial control and a person or company holds editorial responsibility for it. Commentary on the rule stresses that the review has to be substantive, not a cursory sign-off (artificialintelligenceact.eu, practical guide to Article 50). Note what that exception rewards: a real human review with a named accountable party is the thing that takes you out of the labeling requirement. A lawn care blog is unlikely to be “matters of public interest”; a health, finance, or legal blog might be. Ask a lawyer if you’re in that territory.

Everywhere else. Industry rules (advertising standards, regulators for lawyers, doctors, and financial advisers) care about accuracy and substantiation regardless of how the text was produced. The Air Canada chatbot case is the reference point: a tribunal held the company liable for a misrepresentation its chatbot made and rejected the argument that the bot was a separate entity (American Bar Association summary). A label doesn’t transfer responsibility. You own what you publish either way.

What readers say, and how much to trust it

The survey numbers get quoted a lot, so be precise about what they measure. These are surveys of stated attitudes, several commissioned by vendors with something to sell, and what people say about AI in a questionnaire is not what they do when reading a blog post. With that caveat:

Read together, these say three things. Readers are suspicious of AI content in the abstract. They are not good at spotting it in practice, and sometimes prefer it when they don’t know. And what moves comfort most is not the label but an accountable human: 12% versus 43% in the Reuters data is the gap between “AI made this” and “a person made this with AI’s help.”

That reframes the question. The decision is not “label or don’t.” It is “can we honestly say a person is responsible for this, and can we show it.”

A policy a small business can actually follow

Here is what we’d write down. It is short on purpose: a policy nobody can remember is not a policy.

1. Nothing goes out without a human read. Every post is read, in full, by a named person who checks the claims, fixes what’s wrong, and cuts what’s generic. This is the step that moves you from generated to assisted, and the step most of the rules above require or reward. If you can’t do the read yourself, get it done; a professional review exists for exactly this.

2. Every post has a real byline. A person’s name, not “Admin” or “Team.” Google’s guidance explicitly encourages accurate authorship information where readers would expect it. The byline is the accountability. That person should be able to answer a reader’s question about the post without asking the AI.

3. Disclose the method once, where readers can find it. A short paragraph on your About page or an editorial policy page saying how you use AI tools and what a human does before publishing. Link to it from the post footer if you like. This satisfies “self-evident to visitors” without a warning label on every paragraph.

4. Label individual posts when a reasonable reader would want to know. Use a per-post note when the tool did most of the work and the human contribution was review rather than authorship; when the reader is relying on your expertise (health, money, legal, safety); or when a platform requires it. For a post where you wrote the substance and used the tool to tighten paragraphs, the policy page is enough.

5. Correct in public. When a reader catches an error, fix it and add a dated correction line at the bottom. This does more for trust than any disclosure, because it shows someone is listening. The news organizations embarrassed by AI errors were usually embarrassed by the lack of review, not by the tool (Press Gazette’s running list).

6. Never claim what isn’t true. Don’t say “written by our team” if it wasn’t. Don’t say “100% human” if a tool drafted it. Don’t label a post “AI-assisted” and then skip the review. An inaccurate label is worse than none.

What this policy does not do: require you to confess to using a word processor, put a disclaimer on every post, or promise anything you can’t keep. It also happens to describe what the EU rule’s editorial-control exception, Medium’s assisted-content tier, and Google’s helpful-content questions are all pointing at: a human in the loop with their name on it.

Sample disclosure wording

Use these as drafts. Adjust the verbs to what you actually do; don’t include a step you don’t perform.

Variant A, for the editorial policy page (general):

How we write: We use AI writing tools to draft, outline, and edit some of the articles on this site. Every article is reviewed in full by the person named in the byline before it is published. That person checks facts, removes anything we can’t support, and is responsible for what the article says. If you find an error, tell us at [contact] and we’ll correct it and note the correction.

Variant B, for a per-post footer (assisted):

This article was drafted with the help of an AI writing tool and reviewed, fact-checked, and edited by [Name] before publication. [Name] is responsible for its content. Last reviewed [date].

Variant C, for a per-post footer (mostly generated, human-checked):

The first draft of this article was generated by an AI tool. [Name] reviewed it, verified the claims against the sources linked, and made corrections. Questions about anything here should go to [contact].

Correction line, to append when needed:

Correction, [date]: an earlier version of this article said [X]. The correct figure is [Y]. Thanks to [reader, with permission] for pointing it out.

Keep them short and true. The one that says “reviewed by [Name]” when nobody reviewed it will eventually be noticed, because the unreviewed post is the one with the invented statistic in it.

Where this leaves you

The label is the least important part. The surveys, the platform rules, and the regulators are converging on the same thing: whether a person checked the work and will answer for it. Do that first, write it down second, and the disclosure question mostly answers itself. Pick a wording variant above, put it on your About page this week, and make “a named person has read the whole thing” the rule you don’t break.

Have an AI-written article you're not sure about?

A professional editor reads it and emails you the corrections as a prompt for your AI. $99.00, usually within one business day.

Start a review

More from the blog