Your chatbot now has to say it is AI
Since 2 August 2026 every chatbot has to identify itself as AI. What that means for your business, what the fines are, and how to comply without hurting conversion.
On 2 August 2026 a significant part of the European AI Act started to apply. For big tech that is old news. For small and medium businesses it suddenly became very concrete, because the rule now in force touches something thousands of companies have sitting on their website: the chatbot.
The core idea is simple. If someone is talking to a machine, they should know it. This article explains what that means, what you need to arrange and what you do not, and why for most businesses this is an afternoon of work.
What exactly changed?
Article 50 of the AI Act covers transparency. Since 2 August 2026 you must inform users when they are interacting with an AI system, unless it is obvious to a reasonably observant person. In practice: a short notice at the start of the conversation.
Note a distinction many articles blur. The stricter high-risk rules were actually postponed through the Digital Omnibus to December 2027 and August 2028. What applies now is the transparency layer, and that applies immediately.
Does this apply to my business?
Almost certainly yes. The transparency obligation applies to any company deploying an AI system toward users in the EU, regardless of size. There is no small-business exemption, although lower maximum fines do apply to SMEs.
A regular customer service chatbot on a small business website falls under limited risk. That is the lightest category that still carries obligations, and this notice duty is exactly what it involves. You do not need a conformity assessment or a technical documentation file the way high-risk systems do.
What do you actually need to do?
For the vast majority of small businesses it comes down to four points.
- Your chatbot makes clear at the start of the conversation that it is AI. One sentence is enough.
- Your phone voice agent does the same in its greeting.
- AI-generated audio, images and video are marked in a machine-readable way as artificially created.
- If you produce deepfakes or realistic AI images of people, you label them visibly.
If your chatbot comes from an external vendor, you as the deployer are still the one who has to guarantee this. So ask your vendor explicitly whether the notice is switched on and how it is worded.
What if you ignore it?
Fines for breaching the transparency obligations run up to 15 million euro or 3 percent of global annual turnover. That sounds absurd for a painting company, and in practice a regulator will not immediately max out a fine for a small business. Lower ceilings apply to SMEs as well.
The realistic risk is different: a customer complaint, an unpleasant letter from the regulator, or a competitor making something of it. It takes half an hour to fix, so it is an unnecessary risk to carry.
Will an AI notice cost me customers?
That is the question business owners ask us most. The honest answer: no, provided you word it well. People talking to a chatbot usually already know it is not a human. Pretending otherwise is precisely what destroys trust the moment they find out.
A good notice is short, comes first and promises something. Compare these two:
Bad: This is an automated system. Better: Hi, I am the digital assistant at Jansen Installations. I can help you right away, and if it gets too specific I will put you through to a colleague.
The second version complies with the law and increases the chance of a real conversation, because it sets expectations and offers a route to a human.
Checklist for this week
- Open your own chatbot as a visitor and read the first line. Does it say it is AI?
- Call your own number if you run a voice agent and listen to the greeting.
- Check that there is a clear route to a human.
- Ask your vendor in writing whether the notice requirement is handled.
- Add a line to your privacy statement about which AI you use and what for.
What this is really about
The AI Act enforces what good companies were doing anyway. Be open about what a machine does, always let people step out, and make sure the assistant genuinely helps rather than stalls. Businesses that take that seriously find customers are perfectly happy with an AI assistant, as long as it is fast and honest.
Not sure whether your chatbot or voice agent complies? Send us a message and we will take a look with you, no strings attached.
Frequently asked questions
Does my chatbot really have to say it is AI?
Yes. Since 2 August 2026, Article 50 of the AI Act requires you to inform users when they are interacting with an AI system, unless it is obvious. A short notice at the start of the conversation is the simplest way to comply.
Does this apply to small businesses too?
Yes. The transparency obligation applies to every company deploying AI toward users in the EU, regardless of size. There is no exemption for small businesses, although lower maximum fines apply to SMEs.
My chatbot comes from a vendor. Am I responsible?
If you deploy the chatbot toward your customers, you are the deployer and you have to ensure the notice is there. Ask your vendor explicitly whether this is configured and how the notice is worded.
Does the notice requirement apply to an AI phone agent?
Yes. A voice assistant on the phone is just as much an AI system communicating with people. Have it state this in the greeting and always offer a route to a human colleague.
Do I now need a risk assessment or documentation file?
Not for a regular customer service chatbot. That falls under limited risk, where only transparency obligations apply. The heavy documentation and assessment duties apply to high-risk systems, and those were postponed to December 2027 and August 2028.