If a bot answers your DMs, your website chat, or your phone, there's a date you should know about: August 2. That's when Article 50 of the EU AI Act — the transparency rules — starts applying.
The core of it is almost embarrassingly simple. If a person is talking to an AI system, they have to be told they're talking to an AI system, before or at the very start of the conversation. Not in your terms and conditions. Not implied by naming it "assistant." The European Commission's own guidance says the disclosure has to be perceivable in the interaction itself — a sentence buried in a footer doesn't count.
The second piece covers content. AI-generated audio, images, and video need to carry machine-readable marks so they can be detected as generated — that one mostly lands on the tool makers, not you. Deepfakes must be disclosed, full stop. And AI-written text published to inform the public has to be labeled — unless a human reviewed it and someone takes editorial responsibility for it. That exemption matters, and we'll come back to it.
"I'm not in Europe." Read this anyway.
The rules apply based on where your users are, not where your business is. If your website chat can talk to someone in the EU — and unless you're geo-blocking a continent, it can — you're in scope. The penalty ceiling for transparency violations is €15 million or 3% of worldwide annual turnover, whichever is greater. Nobody is fining a ten-person business the maximum on day one; regulators go after the big, flagrant cases first. But "probably not me" is a strange thing to bet on when the fix is one sentence.
There's a second reason to care: this is the direction everywhere. Several US states already have their own bot-disclosure laws on the books, and more are drafting them. The businesses that make honesty the default now don't have to scramble later.
Disclosure isn't a compliance tax on your automation. It's the thing that makes customers trust your automation.
What to actually do this weekend
Make a list of every place AI talks to a customer. Website chat, Instagram DM automation, the AI phone agent, automated email replies. Most small businesses have two or three of these and have half-forgotten one of them.
Give each one an opening line. Something plain: "Hi — I'm an AI assistant for [business]. I can help with most questions, or connect you to a person." That single sentence satisfies the chatbot rule. While you're in there, make sure the human handoff actually works — a bot that admits it's a bot and then traps you anyway is worse than no bot.
Decide who owns your AI-written content. Here's the part most coverage skips: if a human reviews AI-drafted text and takes editorial responsibility for it, the labeling requirement for published text doesn't apply. Which means the workflow you should already have — AI drafts, you review, you publish under your name — is also the compliant one. The rule isn't "never publish AI text." It's "either a person stands behind it, or you say a machine wrote it."
And if you'd rather just say a machine wrote it? That works too. You're reading a post that discloses exactly that, in the byline, on every article. It hasn't scared anyone off. If anything, people trust the rest of the site more because nothing here pretends to be something it isn't.
Customers were never going to be angry that your bot is a bot. Most of them already assumed it. What erodes trust is the pretending — and as of Monday, the pretending is illegal in 27 countries. One honest sentence. Set it up before the weekend's over.