AI disclosure on websites and in outreach: when you have to say it

By Ranjit Bhinge, founder of Blur Studio · Updated · 11 min read

Short answer: AI disclosure means telling people when AI wrote, made or is talking to them. The strongest rule applies to chatbots and other AI that talks to people: under Article 50 of the EU AI Act (the European Union's AI law), people must be told they are dealing with an AI system unless that is obvious. Ordinary AI-assisted website copy and email usually has no general labelling rule, but fake-looking images and video, some platform ad rules and some US state laws do. Check local law and each platform's current rules.

Ranjit Bhinge

You added a chat window, let AI draft your blog posts and send your follow-up emails. Which of these do you have to label, and which should you label anyway?

I'm Ranjit, and I run Blur Studio: 15+ years building software, 120 apps taken live, and clients who have gone on to raise $10M+. I build AI systems and processes that help businesses do a lot more with a lot less time. I also work differently from most of what else is out there, more on that further down this page. Say hello.

Short version: you most clearly need to disclose when a person is talking to an AI (a chatbot or AI voice), when you publish realistic fake images, audio or video, and when a platform's ad or upload rules say so. You usually do not need to label ordinary marketing text that a person reviewed. Everything else is a judgment about trust. The detail depends on where your customers live, so this page is information, not legal advice. Check local law and the current rules of each platform you use.

This page is about websites, chat and outreach. If you want the agency-and-client side (should you tell a client you used AI), read should agencies tell clients they used AI. For the contract wording, see what to put in your contract when you use AI tools.

Three different things people call disclosure

  1. Legal duties. Rules set by a law, which differ by country and state.
  2. Platform duties. Rules set by a site where you publish or advertise, such as YouTube or an ad network. Break them and you risk labels, removal or account trouble, even if no law applies.
  3. Trust practice. No rule forces you, but a short plain note helps a customer feel respected and protects you if they feel misled later.

Most articles blur these. Keep them apart in your head, and in your policy.

The one rule with real teeth: telling people they are talking to AI

The EU AI Act, Article 50, is the clearest written rule. The text on this Article 50 page (an unofficial mirror of the law, not the Official Journal) sets out several duties. These are the ones that matter to a small business:

  • AI that talks to people. Systems meant to interact with people must be designed so people are told they are interacting with an AI system, unless it is obvious to a reasonably well-informed, observant and circumspect person given the context. The information must be clear and given at the latest at the first interaction.
  • Synthetic content from AI providers. Providers of AI systems that generate synthetic audio, image, video or text must mark the output in a machine-readable format. This is a duty for the companies that build the tools, not for you as a user.
  • Deepfakes. Anyone using an AI system to produce a deepfake (realistic fake image, audio or video of real or realistic people, places or events) must disclose that it is artificially generated or manipulated.
  • AI text on public-interest matters. Text generated by AI and published to inform the public about matters of public interest must be disclosed, unless a person reviewed it or has editorial control and a person or company holds editorial responsibility for publishing it.

The article page says Article 50 applies from 2 August 2026, so the chatbot disclosure duty is already in force. The implementation timeline adds that providers of generative systems already on the market before 2 August 2026 have until 2 December 2026 to comply with the machine-readable marking duty. Sources disagree on the exact end of that grace period: one law-firm summary gives 2 February 2027. Ask your tool vendor what it has done instead of relying on a date.

What this means for a small business, in plain terms:

  • A website chatbot that answers visitors is the main case. A short line such as "You are chatting with an automated assistant" at the start covers the idea. Whether your particular chatbot falls under the rule, or is "obvious" from the context, is a legal call. If it might, add the line. It costs nothing.
  • Marketing blog posts and emails drafted by AI and reviewed by a person are not in the public-interest-text duty, which is narrow. That is also why human review matters beyond quality.
  • If you serve EU visitors, do not assume the rule is only for EU businesses. Check where the rule applies to you.

US rules in brief

There is no single US federal law that says "label all AI." Law-firm summaries report two state laws that touch chatbots. California's bot law (Business and Professions Code section 17941, the state's rules for businesses) makes it unlawful to use a bot to communicate with Californians online while hiding that it is a bot, in order to push a sale or influence a vote, and a clear and conspicuous disclosure is the safe route. Utah's AI law, as amended in 2025, is reported to require that a business disclose AI when a consumer clearly and unambiguously asks whether they are talking to AI, and to require more in certain regulated or high-risk uses. The statutes are not reproduced here, so read the current text or ask a lawyer before you rely on either. Check the statute for your state. Laws in this area are changing, and some states, Colorado for example, have passed, delayed or replaced AI laws that are about automated decisions, not website chatbot labels, so do not treat them as disclosure rules.

The safest plain approach also happens to be the most customer-friendly: never claim your chatbot is a human, never invent a human name and photo for it, and answer "are you a bot?" truthfully every time.

Platform rules

These change often. Check the page before you publish.

  • YouTube. Google's video site. Its help page on disclosing altered or synthetic content requires creators to disclose realistic content made or altered with AI, such as a real person appearing to say or do something they did not, altered footage of real events, or a realistic scene that did not happen. It says disclosure is not needed for clearly unrealistic content or for minor edits, and it lists using AI to help with scripts, titles or thumbnails among things that do not need disclosure. Repeated failure to disclose can lead to labels, removal or loss of monetisation.
  • Ad platforms. Google, Meta (the owner of Facebook and Instagram) and others have rules about AI-made or AI-altered ads, especially for political, election and social-issue ads, and they update them often. See AI-generated ads and platform policy traps, and read the current page of each ad platform before you launch.
  • Google Search. Google's guidance on generative AI content warns that AI output may contain inaccuracies and that mass-producing pages can break its spam policy on scaled content. It is about quality, not a label. See does Google penalize AI-written pages.

Required, platform-driven or recommended: a one-page table

As of November 2026. This is a guide for deciding what to do, not a legal opinion.

What you use AI forRule-driven?Recommended practice
Website chatbot or AI assistantYes in the EU (tell people it is AI, unless obvious); reported in some US states, especially when asked or when used to misleadSay it is automated at the start; never pose as a person; offer a human option
AI voice answering callsSame idea as the chatbot; calls and recording have their own rules, so check local lawOpen with a clear line that it is an automated assistant. See AI receptionist
AI-drafted blog posts and web copy, reviewed by a personNo general labelling rule; narrow EU duty for unreviewed public-interest textHuman review before publishing; an author or editor name; note AI help where readers would expect to know
AI-drafted cold or follow-up email, sent by a personEmail laws apply as normal: in the US, the US Federal Trade Commission's (FTC, the national consumer protection regulator) CAN-SPAM guide (CAN-SPAM is the US commercial email law) requires true sender details, an honest subject line, a working opt-out honoured within 10 business days and a postal address; the Article 50 text has no general AI label for ordinary marketing emailDo not pretend a person wrote a note by hand; do not fake personal details; keep the opt-out working
Fully automated outreach (AI chats back to replies)Treat as AI that talks to people: the chatbot rule applies in spiritTell people they are talking to an assistant; hand off to a person on request
Realistic AI images or video of real people, places or eventsYes: EU deepfake duty; YouTube rule; platform ad rulesLabel it; avoid putting real people's faces and voices in without consent
Stylised or clearly illustrative AI art, stock-style imagesUsually noOptional note; keep a record of the tool used
AI-written reviews, testimonials or "customer" quotesDo not do this: the FTC's final rule covers AI-generated fake reviews and testimonialsUse only real customer words with permission

Wording you can copy

Short and plain beats legal. Put it where the customer sees it first.

PlaceSample wording
Chat window first message"Hi, I am an automated assistant for [business]. I can answer common questions and pass you to a person any time."
Phone greeting"You have reached the automated assistant for [business]. I can take a message or book a time. Say 'person' at any time to reach someone."
Blog or guide footer"Drafted with AI help and reviewed and edited by [name], [role]. Last checked [date]."
Image caption"Illustration created with AI." Or for a realistic image: "AI-generated image. Not a photograph."
Email footer (when a tool sends on your behalf)"Sent by [business] using automated tools. Reply to reach a person."

A one-page AI disclosure policy template

Copy this into a document, fill the brackets, and share it with staff and contractors.

AI DISCLOSURE POLICY: [Business name], version [date]

1. Tools. We use these AI tools: [list]. Staff may not add others for client or
   customer work without approval from [name].

2. Talking to customers. Any chatbot, voice assistant or automated reply that
   speaks to a customer says it is automated in its first message. It never uses
   a human name or photo to suggest a person. If asked "are you a person?", it
   answers truthfully and offers a human.

3. Written content. AI may draft text. A named person reviews every page and
   email template before it goes live, checks facts, prices and legal claims,
   and is responsible for what is published.

4. Images, audio and video. Realistic AI images, voices and video of people or
   real events are labelled as AI-generated. We do not use a real person's face
   or voice without their written permission.

5. Testimonials and reviews. We never publish AI-written reviews or quotes as if
   they came from customers.

6. Platforms. Before publishing or advertising, staff read the current AI rules of
   the platform (for example YouTube, and each ad platform) and follow them.

7. Customer data. Customer data is not pasted into AI tools unless the tool's
   terms say it is not used for training, and unless [name] approved it.

8. Records. We keep a log: date, tool, what it was used for, who reviewed it.

9. Review. We review this policy every [6] months or when a law or platform
   rule changes.

Worked example with made-up details

A small plumbing company in Ohio adds a chat window and uses AI to draft a monthly newsletter. The details here are invented to show the method.

  • Chat window: first message says it is an automated assistant, offers "talk to a person" and shows the office phone number. Any question that mentions an emergency, a gas smell or a flood is routed to a person right away. Pass: a test visitor can tell it is a bot, reach a human in two clicks, and the bot never says it is a technician.
  • Newsletter: AI drafts a seasonal tips email. The owner reads it, removes a claim about "guaranteed savings", checks the phone number and sends it under her own name. The footer says "Drafted with AI help, reviewed by [owner name]". Pass: nothing in the email states a result that cannot be backed up.
  • Images: a stock-style illustration of a water heater is created with AI and captioned "Illustration." No AI image shows a real technician, a real customer or a before-and-after job.
  • Review test: the company asks, "Would a customer feel tricked if they learned how this was made?" Three checks fail the test and get rewritten.

Who can keep this light

If you only use AI for private brainstorming, spell-checking or editing text that a person fully rewrites, you do not need labels or a policy. A one-line rule in your staff handbook is enough. Skip the fancy badge. A "powered by AI" banner on pages that a person wrote adds noise and suggests lower quality.

If you run regulated services (health, legal, finance, housing, children's products) or sell into the EU, get a lawyer to look at the policy once. Fair housing, health and financial advertising have their own rules that go far beyond AI.

Common mistakes

  • Naming a chatbot "Sarah" with a stock photo and no mention that it is automated.
  • Labelling everything "AI" as a precaution, which dilutes the label for the cases where it matters.
  • Assuming one country's rule covers all your visitors.
  • Publishing AI text without human review and calling it "edited" because a spell-checker ran.
  • Relying on a tool's own "AI detection" or watermark to be your disclosure. A watermark that a person cannot see is not a notice to a customer.

How I'd approach this

I would not wait for perfect legal clarity, because the cheap, customer-friendly choices are also the safe ones. Say plainly when a customer is talking to a machine, label realistic synthetic media, and put a real person's name on anything published. Keep the log, because that is what you will want if a customer, platform or regulator ever asks.

First-week plan:

  1. Day 1: list every place AI touches the public: chat, calls, email, blog, social, ads, images, video.
  2. Day 2: for each, tick which of the three kinds applies (legal, platform, trust). Where the answer depends on a platform, open its current rules page.
  3. Day 3: add the first-message line to any chatbot or phone assistant, and a human hand-off.
  4. Day 4: add reviewer names and the footer line to AI-drafted pages and templates.
  5. Day 5: fill in the policy template and send it to staff and contractors.
  6. Day 7: test as a customer. Chat with your bot and ask "are you a person?" Pass if it answers truthfully and offers a person. Fail if it dodges.

Decision rule: if a reasonable customer could think they were dealing with a person, or that a fake image was real, disclose. If not, a footer note is optional. Review the rules every six months. If you want AI agents that use your business knowledge, connect to your CRM and calendar, and clearly hand off to your team, Blur Studio builds them, including finishing a half-built setup. See the business page. For how chat tools fit into a business, read AI chatbot for business. Visitor tracking has separate privacy rules: see whether website visitor tracking is allowed under privacy law.

I'm not a lawyer, accountant or other licensed adviser, and this page is general information, not advice. Rules and prices differ by country and change often, so treat it as a direction to think along, and please check with a lawyer or the relevant regulator before you act on it.

Questions people ask

Do I have to label AI-written blog posts or website copy?

There is no general rule that ordinary marketing text must carry an AI label. The EU AI Act's text duty is narrow: AI text published to inform the public about public-interest matters, unless a person reviewed it and holds editorial responsibility (Article 50). Human review matters for quality and for that exception. Check local law, and add an author or reviewer name as a trust signal.

Does my website chatbot have to say it is AI?

In the EU, Article 50 says people must be told they are interacting with an AI system unless that is obvious to a reasonably well-informed person. Some US states have narrower rules, as reported by law-firm summaries, so read the current statute or ask a lawyer. In practice, a one-line "automated assistant" message plus a human option is cheap and removes the doubt.

Should I disclose AI in cold emails or follow-ups?

The Article 50 text does not ask for a label on ordinary marketing email, and email laws (consent, sender identity, opt-out) apply as usual, so check local law. What matters more: do not fake a personal relationship, do not claim a person wrote something by hand, and keep the unsubscribe working. If an AI replies to people, tell them it is an assistant and give a way to reach a person.

Do I need to label AI images?

Realistic images, audio or video of real or realistic people or events are the case to label: the EU deepfake duty, YouTube's disclosure rule and several ad platforms point the same way. Clearly stylised art generally does not need a label, but a short caption is harmless. Check each platform's current page before you publish.

Does Google penalize AI content or require a label?

Google's generative AI guidance does not require a label. It warns that AI output can be inaccurate and that mass-producing pages can break its spam policy on scaled content. See does Google penalize AI-written pages.

Does the EU AI Act apply to a small US business?

It can, if your AI system is used by people in the EU. Article 50 says it applies from 2 August 2026 (the chatbot duty is not part of the marking grace period), and the implementation timeline gives pre-existing generative providers until 2 December 2026 for the marking duty. Whether it covers your setup is a legal question. A chat-start disclosure is cheap protection either way.

Can I name my AI assistant with a human first name?

You can, but do not use it to suggest a person, and never use a fake human photo or backstory. The first message should say it is an automated assistant, and it should answer "are you real?" truthfully. California's bot law, as summarised by law firms, targets hidden bots used to sell, so read the current text if you serve Californians.

What should an AI disclosure policy include?

The tools you use, who reviews AI-drafted content, how chatbots introduce themselves, when images and video are labelled, a ban on AI-written reviews, a platform-rules check before publishing, rules for customer data, a log and a review date. A copy-ready template is in the policy section above. Pair it with your contract wording if you work for clients.

More in this topic: Websites: legal and trust basics (7 guides) →

Sources

Ready for the next step?

A 30 minute call. If you do not need me, I will tell you.

Ranjit Bhinge You talk to me, Ranjit. I reply within a few hours.

Not sure what you need? Tell me what is going on. I will point you the right way, even if it is not me.

Blur Studio · 30 minute call

Book a 30 minute call

You talk to me, Ranjit. If you do not need me, I will tell you.

  • We talk about your business and what is going on.
  • You leave with a clear next step, even if it is not me.
  • I reply to every message within a few hours.
Loading the calendar…