When Do You Have to Disclose AI? US Rules for Chatbots, Calls and Ads

Few questions come up more often from businesses adopting AI than this one: do we have to tell people? The answer in the United States is "sometimes, and more often than you might think." There is no universal federal labeling law, but a set of specific federal rules, state laws and long-standing consumer protection principles require disclosure in particular situations. Knowing where those lines are prevents expensive surprises, and choosing to be open even where no law requires it tends to build trust. This is general information, not legal advice.

The baseline: don't deceive

The broadest rule is the oldest. Section 5 of the FTC Act prohibits unfair or deceptive acts or practices. It does not mention AI, but it applies fully to it. The Federal Trade Commission has repeatedly warned businesses that it is deceptive to:

• Mislead consumers about whether they are interacting with a person or a machine, where that matters to their decisions.
• Make unsupported claims about what an AI product can do.
• Use AI to create fake reviews, testimonials or endorsements.
• Use realistic synthetic images, voices or video to misrepresent a product, its results or who stands behind it.

In other words, even where no AI-specific disclosure law applies, hiding AI in a way that misleads people can be illegal on its own.

Situations where disclosure is specifically required

State rules are multiplying quickly, and several states have considered or passed broader chatbot disclosure requirements, including rules aimed at AI companion apps. Check the current law in each state where your customers are.

Customer service chatbots

Many companies now put an AI assistant on their website or in their app. Good practice, and in some places a legal requirement:

• Say it is AI at the start of the conversation, in plain language: "You're chatting with our AI assistant."
• Do not give the bot a human name and photo that implies it is a person, unless it is clearly labeled.
• Offer a path to a human, especially for complaints, billing disputes and anything involving safety or health.
• Stand behind what it says. Courts and regulators are unlikely to accept "the chatbot said it, not us." If your bot misstates a refund policy, you may be held to it.

Phone calls and voice

The FCC's 2024 ruling that AI-generated voices are "artificial" under the TCPA has major consequences for outbound calling. Using a cloned or synthetic voice in marketing calls without prior express consent exposes a business to the same liability as unlawful robocalls, including statutory damages per call. Calls must also identify the business and, for telemarketing, provide opt-out options. Voice cloning of real people without permission raises additional right-of-publicity and fraud concerns.

Marketing content and images

No federal law requires a label on every AI-written blog post or AI-generated image. The legal question is whether the content misleads. Watch for:

• Product images that show features, results or quality the product does not have.
• Before-and-after images generated or enhanced by AI for health, beauty or fitness claims.
• Synthetic people presented as real customers or employees.
• Imitations of real people's voices or faces without their consent.

Platforms add their own layer. Major social media and ad platforms have policies requiring labels on realistic AI-generated content in some contexts, especially political and social-issue ads. Violating them can get ads rejected or accounts restricted.

When disclosure is wise even if not required

• Journalism and expert content, where readers assume a human wrote and checked the material.
• Professional services, such as legal, financial, medical and tax work, where clients rely on professional judgment. Some professional bodies and courts have issued their own AI rules.
• Creative commissions, where a client is paying for human artistry.
• Education and training materials.
• Contracts that ask. Many enterprise customers now include AI-use questions in vendor contracts and security questionnaires. Answer truthfully.

What good disclosure looks like

A disclosure only counts if people notice and understand it. Practical guidelines:

• Put it where the interaction starts, such as the first chat message or the opening of a call, not in the terms of service.
• Use plain words: "AI assistant," "automated message" or "AI-generated image," not "powered by advanced technology."
• Keep it near the content it describes, and make it readable on phones.
• Repeat it where the context changes, for example when a chat moves from a person back to a bot.

A simple disclosure checklist

1. Could a reasonable person be misled about whether they are dealing with a human? Disclose.
2. Is AI generating a voice for outbound calls? Confirm TCPA consent and identification.
3. Are reviews, testimonials or endorsements involved? Make sure they are real and connections are disclosed.
4. Is this political or issue advertising? Check state laws and platform policies.
5. Does AI affect decisions about people, such as hiring or eligibility? Check notice requirements where those people are.
6. Would a customer feel misled if they found out later? If so, tell them now.

Openness as a default

Disclosure rules in the US are a patchwork, and keeping up with every one is hard. A simple, consistent practice, labeling bots, never faking people or reviews, getting consent for synthetic voices and answering honestly when asked, covers most of them at once. It also tends to be good business: customers are generally more forgiving of a clearly labeled AI assistant than of discovering one they were not told about.