Sometimes you have to, and the rule depends on where the AI shows up. Realistic AI video and audio can need a label on Meta, YouTube and TikTok, an AI voice on an outbound call in the US needs the person's consent first, a chatbot that serves people in the EU has to say it's AI, and a fake review written by AI, posted as if a customer wrote it, is banned in the US.
Where no rule forces it, we'd still tell people in most of the places customers meet the AI, because customers say they want to know. We read each rule on the platform's or the regulator's own page on 2026-10-20 and 2026-10-21. It isn't legal advice, and the rules differ by country, so check the ones where your customers are.
Social posts: label realistic AI video and audio#
All three big platforms can ask for a label when AI makes a video or audio look or sound real. None asks you to label obvious cartoons, and YouTube exempts minor edits like filters.
| Platform | Needs a label | Doesn't | |---|---|---| | Facebook, Instagram, Threads | Meta "may require" one for photorealistic video or realistic-sounding audio made or altered digitally, including with AI. A reel with a realistic AI voiceover is its example. | Images (Meta may still label one it detects). A cartoon-style video. | | YouTube | Realistic content that makes a real person say or do something they didn't, alters footage of a real event or place, shows a realistic scene that never happened, or uses AI music as the main focus. | Beauty filters, captions, cloning your own voice for a voiceover. | | TikTok | AI-generated or significantly edited content that shows realistic-looking scenes or people. | Nothing stated for unrealistic content. |
Meta's help page says "there may be penalties" for content shared without a required label, and that it labels content "AI content" on its own when it finds industry signals that AI made it (Meta Help Center, checked 2026-10-20). YouTube can add a label itself, may penalise creators who keep skipping it, and says disclosing "alone won't limit a video's audience or impact its eligibility to earn money" (YouTube Help, checked 2026-10-20). TikTok's guidelines, in effect since 2026-09-24, say unlabelled content "may be removed, restricted, or labeled" by its team (TikTok Community Guidelines, checked 2026-10-20).
The gap worth knowing: an AI still image on Instagram doesn't need Meta's label, but the same picture in a TikTok video might, if it looks real. Paid ads are separate again: Meta's page says labelling works differently for ads, and Meta and Google each set their own ad rules, so check those before you boost a post.
US calls and bots: consent to call out, disclosure in some states#
The US Federal Communications Commission ruled in February 2024 that AI-generated voices, voice clones included, count as an "artificial or prerecorded voice" under the Telephone Consumer Protection Act. So a call placed to a consumer with an AI voice needs the called person's prior express consent, unless it's an emergency or exempt, and a telemarketing call needs that consent in writing. Every such message has to identify the business responsible for the call, and a sales call has to offer a way to opt out (FCC Declaratory Ruling 24-17, released 2024-02-08, checked 2026-10-20).
Read the scope closely. The ruling covers calls the AI places to consumers: its requirements apply to AI "that initiates any outbound telephone call". It says nothing about an AI answering calls customers make to you, and we found no FCC statement on that. Some states do cover it, and Utah and California are two examples. Utah requires a business using generative AI with a consumer to say so when the person clearly asks whether they're dealing with a human or AI, and requires people in regulated occupations to say it up front in high-risk interactions: out loud at the start of a call, or in writing before a chat (Utah Code 13-77-103, checked 2026-10-21). California makes it unlawful to use a bot online to mislead someone about its artificial identity in order to make a sale, and a "clear, conspicuous" disclosure that it's a bot removes that liability (Business and Professions Code 17941, checked 2026-10-21). The law's word is "bot", which section 17940 defines as "an automated online account" whose actions or posts are all or almost all not a person's (17940). It doesn't say whether a chatbot on your own site counts.
Chatbots: the EU says tell people they're talking to AI#
The EU AI Act's Article 50 says an AI system built to talk directly with people has to tell them they're interacting with AI, unless that's obvious to "a natural person who is reasonably well-informed, observant and circumspect". The same article says anyone using AI to make a deepfake image, audio or video has to disclose it. The notice has to be "clear and distinguishable" and come at the latest at the first interaction (Article 50, AI Act Service Desk, checked 2026-10-20).
The chatbot duty in Article 50 sits with the provider. Article 3 defines that as whoever develops an AI system, or "has an AI system ... developed", and puts it into service "under its own name or trademark" (Article 3, checked 2026-10-21). Use an off-the-shelf bot under the vendor's name and the vendor carries it. Have one built, or run one under your own brand, and you may be the provider. The deepfake duty sits with whoever uses the tool, which can be you either way.
On timing: Article 113 says the Act applies from 2 August 2026 and names no exception for Article 50 (Article 113), and the Commission says the transparency rules "come into effect in August 2026" (European Commission, checked 2026-10-20). Being outside the EU doesn't settle it: Article 2 also covers providers and deployers in other countries when the AI's output is used in the EU (Article 2).
One date did move. The AI Omnibus amendments, in force since 27 July 2026 (same Commission page), give providers of generative AI systems already on the market before 2 August 2026 until 2 December 2026 to mark their outputs in a machine-readable format under Article 50(2) (Article 111, checked 2026-10-21). Article 111 gives no such delay for the chatbot or deepfake duties.
Reviews: fake AI reviews are banned in the US#
The US Federal Trade Commission's rule on reviews and testimonials, in effect since 2024-10-21, bans reviews by someone who doesn't exist, from someone with no real experience of the business, or misrepresent that experience, and the FTC names "AI-generated fake reviews" as an example. A business can't write them, buy them, or post ones it knew or should have known were fake, and the FTC can seek civil penalties against knowing violators (FTC announcement, 2024-08-14, checked 2026-10-20). The same rule bans paying for reviews of a set sentiment and staff reviews that hide the connection. Our reading: a real customer using AI to word an honest review isn't what the rule describes.
The FTC's questions-and-answers page says you "can respond publicly" to a negative review, and that the rule bars falsely accusing the reviewer. It also says the rule has "no blanket prohibition on the use of AI-generated avatars in marketing", but an avatar's testimonial is prohibited if the testimonial underneath is fake or false (FTC Q&A, checked 2026-10-21). It's silent on using AI to draft your replies.
Canada: no general federal AI disclosure rule yet#
As of 2026-10-21, the federal government's own consultation page says "Canada does not currently require developers or deployers of generative AI systems to disclose AI-generated content." Innovation, Science and Economic Development Canada is asking whether to act on labelling AI-generated content and on telling people when they're dealing with an AI system, and says the consultation "does not predetermine the outcomes". It also names Bill C-34, a proposed bill that would require labelling of certain synthetic content on regulated social media services (ISED, checked 2026-10-21).
Today's law is the misleading-advertising rules. The Competition Bureau's submission to that consultation, dated 2026-09-23, says "there is no 'AI exception' to the law" and that labels on AI content "alone may not be sufficient" (Competition Bureau Canada, checked 2026-10-20). So an ad made with AI is judged the same way as one made without it.
What we'd do (our take, not a rule)#
Our line: disclose when the AI speaks to a customer as itself, or shows something that didn't happen. Don't bother when it helps you write words you then read, fix and sign.
The reason is what customers say. In a May 2026 survey of 6,000 adults in the US, UK and Canada, 85% said it should be clearly indicated when they're dealing with AI rather than a person, 81% called AI pretending to be a real person an ethical concern, and 31% said they'd hang up if a business put them through to AI (AnswerConnect, survey by OnePoll, checked 2026-10-20). AnswerConnect sells human answering services, so read it as one vendor's survey. It didn't compare disclosed AI with hidden AI, and a label won't win back the people who hang up on any AI. What a label does avoid is the objection 81% raised: AI pretending to be a person.
Where we'd put a line, and what it could say:
- The phone, first sentence. "Hi, you've reached Northside Plumbing. I'm the AI assistant: I can book a visit or take a message, or ask for a person." (A sample business.) No human first name for the bot, and no fake café noise behind it.
- The chat window, before the first reply. "You're chatting with our AI assistant. Type 'person' any time and the team picks it up." Only promise the handoff if it's real.
- A realistic AI image, even where Meta doesn't ask. "Image made with AI. The cake is real; the garden party isn't."
- A newsletter or blog post AI helped draft. Nothing. You read it, you fixed it, it's yours.
- Review replies. Draft them with AI if it helps, have a person read each one, and sign it as the business. Never post a review for a customer, in any wording.
Before you switch on any customer-facing AI this week, call it or message it the way a customer would. If it doesn't say it's AI in its first line, change the greeting before anyone else hears it.