AI disclosure questions, answered plainly

In the US, generally no. No federal or state law we track requires a label on AI-assisted blog posts or web pages, and Google does not require one for Search; it only suggests explaining how content was made where readers would expect it. Two cautions. First, if the post makes claims, they have to be true, whoever wrote them. Second, if you publish for EU readers, Article 50 asks for a label on AI-written text on matters of public interest unless a person reviewed it and takes editorial responsibility. A short "drafted with AI, checked by [name]" note handles both. See the sample wording and the EU rule.

It depends on what the image shows and where it runs. If an AI image could mislead buyers about the product, the problem is the image, and a label will not fix it under the FTC Act. If an ad shows an AI-made person, New York has required a conspicuous disclosure since June 9, 2026. Political ads have their own labels in many states. Platforms add rules: Google Merchant Center requires AI image metadata, TikTok requires a label or your own disclaimer in ads, and Meta requires disclosure in political and social-issue ads. In the EU, realistic AI images of real people, places or events count as deepfakes and need a visible label. See platform rules.

No. The FTC's fake reviews rule, in force since October 21, 2024, bans creating, buying or spreading fake reviews and testimonials, including reviews from people who do not exist, and the FTC named AI-generated fake reviews when it announced the rule. Knowing violations can cost up to $53,088 each. You can ask real customers for honest reviews. You cannot reward only positive ones, and reviews from staff or relatives must say so. The FTC's decision in December 2025 to drop its order against an AI writing tool did not change any of this for businesses that post reviews. Details in the federal guide.

Yes, if you are honest about it. Do not present the avatar as a real customer with real experience of your product; the FTC's Endorsement Guides treat virtual personas as endorsers, and endorsements must reflect actual experience. New York's synthetic performer law (in force since June 9, 2026) requires a conspicuous disclosure when you know a commercial ad includes an AI-made human performer; its reach outside New York is unclear, so label regardless. And never make an avatar that looks or sounds like a real celebrity or public figure without permission; Google Ads suspends accounts for that without warning, and the FTC's ban on deception applies. A simple on-screen line, "AI-generated presenter," shown while the figure is on screen, is the kind of disclosure these rules point to, as long as the ad does not also present the avatar as a real customer. See example 5.

Not for being AI-written. Google's guidance says AI content is fine if it is helpful and accurate. What Google penalizes is scaled content abuse, meaning many low-value pages produced to game rankings, whether by AI or by hand. That policy has been in effect since March 5, 2024. Have a person fact-check AI drafts, including titles, meta descriptions and image alt text, and publish pages that answer real questions. The exception is Google Merchant Center, where AI-generated product images, titles and descriptions must be labeled in your product data. See Google's rules.

In some places, yes, and it is a good idea everywhere. Maine requires a clear notice whenever a chatbot could lead a consumer to believe a human is responding. New Jersey requires disclosure at the outset when a bot is used to sell or advertise goods or real estate. California bans using a bot to sell while misleading people about its artificial identity. Utah requires an honest answer when a customer asks, and proactive notice for regulated professions in high-risk conversations. In the EU, chatbot providers must make their bots disclose they are AI from August 2, 2026. There is no federal requirement. An opening line such as "I'm an automated assistant, not a person" meets the Maine, New Jersey and California bot rules. Also keep a small "AI assistant" label visible for the whole chat, which fits Utah's safe harbor, and make sure the bot answers honestly whenever it is asked, as Utah requires. Health care, mental-health and companion-style bots have extra rules, such as reminders every 3 hours for companion bots in New York (and Colorado from 2027). Check the state guide for the places you serve.

Only with the right consent. Since February 8, 2024, the FCC has treated AI-generated and cloned voices as "artificial" voices under the Telephone Consumer Protection Act. Telemarketing and advertising calls with an AI voice need the person's prior express written consent. The call must name your business at the start, give your phone number during or at the end, and offer an automated opt-out. People can sue for $500 per call, up to $1,500 if willful. No federal rule yet requires you to say the voice is AI, though the FCC has proposed it, and Maine's chatbot law covers voice. See the federal guide and a sample script.

The truth, plainly: "No, I'm an AI assistant, not a person," followed by an offer to connect them with someone. Utah law requires a business using generative AI with consumers to disclose it when the person clearly asks, and an evasive answer can also be a deceptive practice anywhere. Build the answer into the bot's instructions and test it by asking the question several different ways. See example 3.

Treat it that way. Under the FTC Act, a statement your chatbot makes to a customer is a statement your business made. If it invents a refund policy or promises something you do not offer, that is your claim. Utah's law states it outright: using AI is never a defense to a consumer protection violation. Give the bot your real policies, limit what it can promise, and make it easy to reach a person.

In several places, yes. Illinois requires employers, even very small ones, to notify employees when AI is used in hiring, promotion, discipline and similar decisions, since January 1, 2026, and requires notice and consent before AI analyzes video interviews. New York City requires a bias audit and notice at least 10 business days before an automated hiring tool is used on NYC candidates. Ontario requires employers with 25 or more employees to say in job postings whether AI screens applicants. Colorado's automated-decision law adds notice and explanation duties from January 1, 2027. Sample wording is in example 9.

Yes, stricter. In California, AI-generated patient messages about clinical information need an AI disclaimer and instructions for reaching a human, unless a licensed provider reviewed them first. Texas requires practitioners to disclose AI used for diagnosis, and requires AI disclosure to patients in health care services. Utah requires professions regulated by its Department of Commerce to disclose generative AI up front when a conversation collects sensitive health, financial or biometric data or gives personalized advice. And claiming an AI tool can replace a professional is exactly what the FTC penalized in the DoNotPay case. See examples 10 and 11.

Not as of October 15, 2026. No federal law requires general labeling of AI content or chatbot disclosure. Bills are moving: the GUARD Act, which would make chatbots disclose they are not human, and the NO FAKES Act, on unauthorized AI replicas of a person's voice or likeness, both cleared the Senate Judiciary Committee in 2026. Neither has passed. Meanwhile the FTC's ban on deceptive practices, its fake reviews rule and its Impersonation Rule all apply to AI, and the FCC's robocall rules apply to AI voices. See the federal guide.

Not so far. A December 2025 executive order directs federal agencies to challenge state AI laws, but the order itself preempts nothing. The Senate voted 99-1 in July 2025 to strip a proposed 10-year moratorium on state AI laws, and Congress has passed no preemption statute since. A March 2026 White House framework asks Congress to act but would leave general consumer protection laws in place. Plan as if state laws apply, because they do. See where the fight stands.

Only if you have an EU connection. If you build an AI system that is placed on the market or used in the EU, the Act can reach you wherever you are based. If you sell to EU customers, run a chatbot for them, or publish realistic AI images or video aimed at EU audiences, the transparency duties in Article 50, which apply from August 2, 2026, are worth a careful look. Fines can reach EUR 15 million or 3% of turnover, with the lower figure applying to small businesses. If you only sell in the US and do not target EU audiences, it is background. See the Article 50 guide.

We found no Minnesota statute requiring businesses to tell customers they are dealing with a chatbot or AI. Minnesota does ban election deepfakes in set windows before elections, bans offering or advertising AI "nudification" tools (since August 1, 2026, with penalties up to $500,000 per use), and allows suits over sexual deepfakes. A Minnesota business still has to follow other states' rules when it serves customers there, plus federal law and platform rules. See the Minnesota section.

Each law has its own wording, and the FTC's 2023 Endorsement Guides define the term for endorsements. Across the rules on this site, the practical test is the same: would an ordinary person actually notice and understand it before they rely on the content? That means plain words, placed where the person is looking (in the chat, on the image, at the start of the call), at the moment it matters, readable on a phone, and not tucked into fine print or your terms of service. The EU's version adds that it must come at the first interaction or exposure and be accessible. When in doubt, make it bigger and earlier.

It ranges widely. Platforms act fastest: rejected ads, removed listings, lost monetization, or a suspended ad account. Regulators vary. New Jersey's bot law runs from $2,500 to $10,000 per offense. Utah's runs up to $2,500 per violation. Texas's TRAIGA starts with a written notice and 60 days to fix the problem; curable violations that are not fixed run $10,000 to $12,000 each, and violations that cannot be cured run $80,000 to $200,000 each. FTC rule violations can reach $53,088 each. TCPA robocall suits are $500 to $1,500 per call. Some state laws give a cure period or start with a warning letter, but do not count on it. The cheapest fix is to disclose up front.

For the EU, no. The European Commission says content created before August 2, 2026 does not need to be labeled retroactively, though it encourages it. The US laws in these guides mostly govern conversations, ads and messages as they happen, so the practical question is what is still running: a chatbot, a live ad, an evergreen video. Label what is live and new, and fix older content when you next touch it.

No law we track requires one for a typical small business. It is useful anyway. When a customer, a client or a journalist asks how you use AI, a dated page means your staff, your chatbot and your website all give the same answer. It also forces you to decide where a person reviews AI work, which is the question that matters most. A short version is free on the examples page. The paid pack has a fuller fill-in version plus a human review policy, team AI rules and an update log.