AI disclosure examples you can copy
Twelve short statements for common small-business situations. Copy them, fill in the brackets, and make sure every line is true for your business.
Four rules for any disclosure
Every sample below follows the same four rules. If you write your own, use them as a checklist.
- Say it before it matters. At the start of a chat or call, next to the image, at the top of the message. A disclosure people find after they have already relied on something does little.
- Say what the AI did. "Drafted with AI and reviewed by Dana" tells the reader more than "AI may have been used."
- Say where a person comes in. Who reviews it, and how to reach a human. Several laws, including California's health care rule, require that second part.
- Never overclaim. Do not write "every answer is reviewed by an expert" unless every answer is. A false disclosure is worse than none, because it is a false statement in its own right.
These samples are free to copy and adapt. They are starting points, not a guarantee of compliance with any particular law. The state guide and federal guide explain which rules sit behind each one. Replace everything in [brackets].
1. A short "How we use AI" statement for your website
Use it when: you use AI anywhere customers might wonder about it, and you want one honest place to point to. Footer link, About page or FAQ.
How we use AI
We use AI tools to help with [drafting product descriptions and editing photos]. A person on our team reviews that work before it is published. [Our website chat is an AI assistant. It answers common questions on its own, says so in its first message, and will connect you with a person whenever you ask.]
We do not use AI to [set your price, decide whether to approve your order, or give legal, medical or financial advice].
If you would rather deal with a person, email [address] or call [number]. Someone on our team will answer.
Last updated: [date]
Why it is worded this way: it names specific uses instead of a vague "we may use AI," states a limit, and gives a human contact. Only keep the review sentence for work that really is reviewed before it goes out; a live chat answer is not, which is why the chat gets its own line. The date matters: when your tools change, the page should change with it. No US law requires this page for most businesses. It exists so your chat bot, your staff and your website all give the same answer when someone asks. The paid pack has a fuller version with sections for each type of use.
2. Chatbot greeting
Use it when: a chat widget, messaging bot or AI agent talks to customers on your site or in an app.
Hi, I'm [Bot name], an automated AI assistant for [Business name]. I'm not a person.
I can help with [hours, order status, booking and common questions]. Type "agent" at any time to reach our team [during business hours, or leave a message and we'll reply by email].
Why it is worded this way: Maine requires a clear notice whenever a chatbot could lead a consumer to think a human is responding, and New Jersey requires disclosure at the outset when a bot is used to sell. Utah offers a safe harbor when the AI itself says it is AI at the outset and throughout the conversation. Saying it in the first message meets the Maine and New Jersey rules. For Utah's safe harbor, the notice also has to stay visible for the whole conversation, so pair the greeting with the persistent label described below. "I'm not a person" is blunt on purpose; a human-sounding name with no notice is exactly what these laws target. If you give the bot a name, keep the notice in the same message.
For Utah's safe harbor, the persistent label is required, not optional: the statute says the disclosure must come at the outset and throughout. A small "AI assistant" label next to the chat box that stays on screen for the whole conversation is the simple way to do that. Separately, Utah requires the bot to answer honestly whenever someone asks if it is a person (see example 3).
3. When a customer asks, "Am I talking to a real person?"
Use it when: setting up your chatbot's or voice agent's response to this question. Put it in the bot's instructions.
No, I'm an AI assistant, not a human. I'm happy to keep helping, or I can [connect you with someone on our team now / have a person call you back]. Which would you prefer?
Why it is worded this way: Utah requires a business using generative AI with consumers to say so when the person clearly asks. The answer should be a plain "no," never a dodge like "I'm here to help!" Offering the human option right away turns a possible complaint into a normal handoff. Test it: ask your bot the question five different ways and make sure it answers honestly every time.
4. AI-generated images and AI-edited photos
Use it when: you post AI-made images, or photos AI has meaningfully changed, on product pages, ads or social media.
Caption for a generated image: Image created with AI.
Caption for an edited photo: Photo edited with AI: background replaced. The product is shown as it ships.
Note for a product page: The lifestyle images on this page were made with AI to show the product in use. The product photos are real and unretouched except for color correction.
Why it is worded this way: the risk with product images is not the AI, it is a buyer being misled about what they will get. That is a deception problem under the FTC Act, and a label does not fix an image that misrepresents the product. Keep the product itself accurate and say which images are illustrations. If you sell through Google Merchant Center, AI-generated product images also need the right metadata, not just a caption; see platform rules. Ordinary retouching, cropping and color correction generally need no label.
5. AI-generated people in ads
Use it when: an ad uses an AI avatar, AI presenter or AI-generated "person" of any kind.
The presenter in this ad is AI-generated. [He/She/They] is not a real customer or employee.
Shorter, for tight spaces:
AI-generated presenter.
Why it is worded this way: New York's synthetic performer law (in force since June 9, 2026) requires a conspicuous disclosure when a commercial ad includes an AI-made human figure that you know is synthetic. The FTC's Endorsement Guides treat virtual personas as endorsers, and an AI figure has no real experience with your product, so it must not be presented as a satisfied customer. The second sentence of the long version handles that. Put the label on screen while the figure is visible, not only in the closing credits. Audio-only ads are exempt from the New York rule.
6. AI-assisted articles and blog posts
Use it when: you publish articles, guides or newsletters that AI helped write.
How this article was made: We used an AI tool to help outline and draft this article. [Name], our [role], checked the facts, edited it and is responsible for what it says.
Why it is worded this way: no US law generally requires labeling AI-assisted blog posts, and Google only recommends explaining how content was made where readers would expect it. The EU's Article 50 asks for a label on AI-written text published to inform the public on matters of public interest, unless a person reviewed it and someone holds editorial responsibility. This note shows both: a named reviewer who stands behind it. It also builds trust, which is the real reason to use it. If no one reviewed the article, do not publish this note, and think twice about publishing the article.
7. Customer emails written with AI
Use it when: you send automated AI replies, or want to be open that staff draft with AI.
Fully automated reply: This is an automated reply written by AI. If it doesn't answer your question, just reply to this email and a person on our team will follow up within [one business day].
Drafted with AI, sent by a person: This reply was drafted with help from an AI tool and reviewed by [Name] before sending.
Why it is worded this way: ordinary business email that a person reviews and sends needs no AI label under any law covered in these guides, outside the health care and licensed-profession settings below, so the second line is optional. Fully automated replies are different: customers reasonably assume an email signed by your business came from someone there, and the first line corrects that while giving them a human route. The exception is health care: in California, AI-generated patient messages about clinical information need a specific disclaimer (see example 10).
8. Phone and voice agents
Use it when: an AI voice answers your phone, makes calls for you, or leaves voicemails.
Inbound (answering your line): Thanks for calling [Business name]. You're speaking with an automated AI assistant. I can [book appointments and answer common questions]. Say "representative" at any time to reach a person.
Outbound (only to people who gave consent): Hi, this is an automated AI assistant calling on behalf of [Business name] about [your appointment on Tuesday]. [Message.] You can reach [Business name] at [phone number]. To stop these calls, press [9] or say "stop."
Why it is worded this way: the FCC treats AI-generated voices as artificial voices under the Telephone Consumer Protection Act. Outbound calls need the person's prior consent (written consent for telemarketing), must name the business responsible at the beginning, must give its phone number during or at the end, and telemarketing calls must offer an automated opt-out. No federal rule yet requires saying "AI," but Maine's chatbot law covers voice as well as text, and a caller who discovers mid-call that the friendly voice is software has a fair complaint. Say it up front. This script does not create consent; you need that before you dial. See the federal guide.
9. Hiring: job postings and employee notices
Use it when: AI helps screen applicants, rank resumes, analyze interviews, or make decisions about current staff.
Job posting line: We use artificial intelligence to help screen applications for this role. A member of our hiring team reviews [every application the tool screens out / all interview and hiring decisions].
Employee notice: [Company] uses AI tools in [recruiting / scheduling / performance reviews / promotion decisions]. These tools [describe in one sentence what they do]. A manager makes the final decision. If you have questions or think a decision was wrong, contact [name, email].
Candidate notice for automated tools: We will use an automated tool to help evaluate candidates for this position. It assesses [list the job qualifications and characteristics it evaluates]. If you would like an alternative selection process or an accommodation, contact [name, email] by [date].
Why it is worded this way: Ontario requires employers with 25 or more employees to say in public job postings whether AI is used to screen, assess or select applicants. Illinois requires employers to notify employees when AI is used in employment decisions, and requires notice, an explanation and consent before AI analyzes video interviews. New York City requires a bias audit and notice to NYC candidates at least 10 business days before an automated tool is used, naming what it assesses and how to request an alternative; the third sample covers the notice part only. Colorado's automated-decision law adds duties from January 1, 2027. Hiring is where AI rules have the most teeth for small employers; see the state guide.
10. Health care messages to patients
Use it when: a clinic, practice or health service uses AI to write patient messages or to help with diagnosis.
At the top of an AI-generated patient message: This message was generated by artificial intelligence. To speak with a member of our care team, call [number] or reply "STAFF" to this message.
Patient notice about diagnostic AI: Our practice uses AI tools to help review medical records and suggest possible diagnoses or treatment options. Your [doctor / provider] reviews every suggestion and makes the decision. Ask us anytime if you have questions about how these tools are used in your care.
Why it is worded this way: California's AB 3030 requires a disclaimer that the message was AI-generated plus clear instructions for reaching a human, at the start of letters and emails, throughout a chat, and spoken at the start and end of audio. Messages a licensed provider reads and reviews before sending are exempt. Texas requires practitioners to disclose AI used for diagnostic purposes, and TRAIGA requires disclosure to the patient by the date care is first provided; one notice like the second sample, given at intake, can serve both. Only say "your provider reviews every suggestion" if that is your actual process.
11. Legal, financial and other licensed advice
Use it when: a chatbot or AI tool talks with clients in a law, accounting, tax, insurance, financial or other licensed practice, especially if it gathers sensitive information.
Before we start: this chat uses generative AI. It is not a person and is not a licensed [attorney / accountant / financial advisor]. It can [collect your information and answer general questions about our services]. Advice about your situation comes from [Name], [credential], after reviewing your information. To talk with [him/her/them] directly, [book here / call number].
Why it is worded this way: Utah requires professions regulated by its Department of Commerce to disclose generative AI up front in high-risk interactions, meaning ones that collect sensitive health, financial or biometric data or give personalized legal, financial, medical or mental-health advice. The notice goes at the start of a spoken interaction and before a written one. Saying the tool is not licensed also guards against the kind of "AI can replace your lawyer" claim the FTC penalized in the DoNotPay case. Colorado's new chatbot law (from 2027) bars covered bots from falsely claiming to equal a licensed professional, and a pending federal bill, the GUARD Act, would require chatbots to say they lack professional credentials.
Where disclosures go wrong
- Buried in the terms of service. A chatbot disclosure belongs in the chat. An ad disclosure belongs in the ad.
- Only at the end. A label in the closing frame of a video, or at the bottom of a long email, reaches people after the impression has formed.
- Vague. "Content may be AI-assisted" on every page says nothing about any particular page.
- Stale. Your tools change. A disclosure that described last year's setup is now inaccurate. Date it and review it.
- Used as cover. A label does not make a fake review, a misleading product image or an invented claim acceptable. Fix the content first.
12. Social media video and posts
Use it when: you post realistic AI video, voice or imagery on YouTube, Facebook, Instagram, TikTok or LinkedIn.
Why it is worded this way: YouTube and TikTok require you to flag realistic AI content with their own tools (YouTube's "AI use" setting at upload, TikTok's AI-generated label). Meta requires its AI disclosure tool for photorealistic video and realistic audio in ordinary posts (still images are not part of that duty, though Meta may label them on its own). LinkedIn has no required label. Where a platform has a required toggle, use it first, since that is what the platform enforces. The caption is for your audience: it says what was AI-generated, which the platform label alone does not. If you clone your own voice for narration, YouTube does not require a label, but this caption still costs nothing. Details on each platform are in the platform guide.