Time fo writing: 20 August 2026
If you are using an AI chatbot, AI receptionist, Voice AI or an AI agent to communicate with customers, there is an important new rule you need to be aware of.
Article 50 of the EU AI Act became applicable on 2 August 2026.
And one of the main requirements is actually pretty straightforward:
If someone is interacting directly with AI, they should know they are interacting with AI.
You can read the actual legislation here:
EU AI Act – Article 50 on EUR-Lex
The European Commission has also published detailed guidance explaining how Article 50 should work in practice:
European Commission – Article 50 Transparency Guidelines
So what does this actually mean for a normal business using AI?
The days of pretending your AI bot is a human may be over
This is probably the biggest practical takeaway.
A lot of effort has gone into making AI agents sound as human as possible.
That’s understandable.
We don’t want customers talking to something that sounds like an old-fashioned automated phone system.
But there is a big difference between making an AI sound natural and making someone believe it is a human being.
For example, imagine you have an AI chatbot called Sarah.
A customer starts a conversation and Sarah says:
“Hi, I’m Sarah. How can I help?”
The customer could quite reasonably assume Sarah is sitting at a desk somewhere.
Instead, something as simple as this makes the position clear:
“Hi, I’m Sarah, ABC Ltd’s AI assistant. How can I help?”
You haven’t ruined the conversation.
You’ve simply told the customer what they are interacting with.
This could be particularly important with Voice AI
This is where I think businesses need to pay particular attention.
AI voices have become incredibly realistic.
With platforms such as HighLevel you can now have an AI receptionist answer the phone, qualify enquiries, answer questions, book appointments and even transfer calls.
Imagine phoning a heating company and hearing:
“Good morning, ABC Heating, Sarah speaking. How can I help?”
If Sarah isn’t actually a person, the caller could understandably believe they are speaking to a receptionist.
A simple change fixes that:
“Good morning, ABC Heating. You’re speaking with our AI receptionist. How can I help?”
That’s hardly going to destroy the customer experience.
In fact, I think businesses could turn it into a positive.
“I’m ABC Heating’s AI receptionist. I’m available 24/7 and can answer questions, book appointments or get one of the team to call you back.”
Now AI is part of the benefit rather than something you’re trying to hide.
What about website chat?
The same principle applies.
If someone opens a website chat and an AI immediately starts answering their questions, make it clear that it is an AI assistant.
You don’t need a frightening legal disclaimer.
Something like:
“Hi, I’m our AI assistant. How can I help?”
may be all that’s needed to make the nature of the interaction clear.
The important thing is that the customer shouldn’t have to spend ten minutes chatting before eventually asking:
“Hang on. Are you a bot?”
Don’t rely on the customer asking
I’ve seen AI prompts which include instructions along the lines of:
“If someone asks whether you’re an AI, tell them you are.”
That’s sensible.
I would definitely have that instruction in the prompt.
But Article 50 isn’t really about admitting you’re AI when somebody catches you out.
The person should be informed that they are interacting with AI from the start of the AI interaction.
So I’d do both.
Make the AI status clear at the beginning and make sure your prompt prevents the AI from subsequently pretending to be human.
SMS and WhatsApp get more interesting
This is an area I think businesses using AI for lead follow-up and database reactivation need to look at.
Imagine you send a normal automated SMS:
“Hi John, are you still looking for a new boiler?”
John replies:
“Yes.”
An AI agent then takes over the conversation.
The important point isn’t necessarily whether AI wrote the original SMS.
The question is when John starts interacting directly with AI.
At that point I’d want the AI to make its position clear.
For example:
“Thanks John. I’m ABC Heating’s AI assistant. I can ask you a few questions and arrange for someone from the team to contact you.”
That’s transparent without making the conversation clunky.
What about HighLevel users?
This is one reason Article 50 caught my attention.
GoHighLevel has been rapidly expanding its AI capabilities.
Businesses can now potentially have AI involved with:
- Website chat
- SMS
- Lead qualification
- Appointment booking
- Database reactivation
- Voice AI
- AI receptionists
- Automated follow-up
So if you’re using HighLevel AI, I’d spend ten minutes looking at your setup from the customer’s perspective.
At what point does AI take over the conversation?
Does the customer know?
Could they reasonably think they’re talking to one of your employees?
And try this one:
Ask your AI agent if it is a real person.
You might be surprised by the answer.
If you’ve previously prompted the AI to stay in character at all costs, this is definitely worth testing.
Does this affect UK businesses?
Potentially.
Don’t simply assume:
“We’re not in the EU anymore, so this doesn’t apply to us.”
The EU AI Act can have application beyond businesses physically established within the EU.
If your AI systems are being used in the EU or interacting with EU customers, you should establish whether the legislation applies to your particular circumstances.
This could therefore be relevant to UK businesses selling into Europe and UK agencies building AI systems for European clients.
What about AI-written blog posts?
This is another area where I think we’re going to see some misleading headlines.
Article 50 does not simply mean that every piece of content created with the help of ChatGPT now needs a big:
“THIS WAS WRITTEN BY AI”
warning at the top.
There are separate rules covering AI-generated and manipulated content, including certain text published to inform the public about matters of public interest.
But there is an important exception where the AI-generated text has been subject to human review or editorial control and somebody takes editorial responsibility for it.
That’s an important distinction.
If you ask AI to automatically create 100 news articles and publish them without anybody checking them, that’s one thing.
If you use AI to help research and draft an article, then read it, fact-check it, change it, add your own experience and decide what gets published, that’s very different.
The European Commission’s Article 50 guidance specifically addresses this distinction.
Article 50 isn’t just about chatbots
There is more to Article 50 than conversational AI.
It also contains transparency requirements relating to things including:
- AI-generated and manipulated content
- Deepfake images, audio and video
- Certain AI-generated text on matters of public interest
- Emotion recognition
- Biometric categorisation
There are also requirements concerning machine-readable identification of AI-generated or manipulated content.
So this isn’t simply the “AI chatbot law.”
But for the average business experimenting with AI right now, customer-facing conversational AI is probably the obvious place to start checking.
Five things I would do now
If you’re already using AI to communicate with customers, I’d do these five things.
1. Find every place AI talks directly to customers
Check your website, SMS, WhatsApp, social messaging and telephone systems.
2. Check the start of the AI conversation
Is it clear that the customer is now interacting with AI?
3. Review your AI prompts
Look for instructions telling the AI to pretend to be human, stay in character or avoid revealing that it is AI.
4. Test your AI
Talk to it yourself.
Ask:
- “Are you a real person?”
- “Are you a bot?”
- “Are you AI?”
See what happens.
5. Check whether you have EU customers
Particularly if you’re a UK business, don’t dismiss this simply because Brexit happened.
I don’t think this is necessarily bad news for AI
Some businesses are going to worry that telling customers they’re interacting with AI will reduce conversions.
I’m not convinced.
If I contact a company at 10pm and an AI assistant immediately answers my question, I don’t particularly care that it’s AI.
What I don’t want is an AI deliberately trying to convince me that somebody called Sarah is sitting in the office at 10pm replying to me.
There is an opportunity here to make AI a benefit:
“I’m our AI assistant, so I’m available 24/7.”
That’s a pretty good customer proposition.
Human-like is fine. Pretending to be human isn’t.
That’s probably my biggest takeaway from Article 50.
AI conversations should absolutely feel natural.
They should understand context, ask intelligent questions and communicate in normal language.
But there is no reason they need to deceive someone into believing they’re human.
For businesses using HighLevel Conversation AI, Voice AI or other AI agents, now would be a very good time to test exactly what your customers experience.
The change required could be as simple as adding three words:
“I’m an AI assistant.”
And that seems a pretty small price to pay for being transparent with your customers.
Further reading
Read Article 50 of the EU AI Act on EUR-Lex
Read the European Commission’s Article 50 transparency guidance
This article is intended as a practical overview rather than legal advice. If AI is a significant part of your business or you operate extensively within the EU, you should take appropriate professional advice.
About the Author
Julian Mills
HighLevel Consultant & Marketing Automation Strategist
Julian Mills helps business owners automate lead generation, sales follow-up, customer communication and business processes using HighLevel and MarketerM8. Since 2009 he has helped businesses implement CRM, marketing automation and AI-powered systems that save time, improve customer experience and generate more sales.
About Julian
Book a Discovery Call
Join a Free HighLevel Workshop
Learn About HighLevel with MarketerM8
Published: 26 June 2026