The European Commission has published new guidelines on the transparency requirements under the AI Act. Starting August 2, 2026, organizations must make it clear when someone is communicating directly with an AI system. For customer interactions, this has direct implications for the use of chatbots, voicebots, and digital assistants.
AI is increasingly becoming part of customer interactions. Chatbots answer questions, voicebots identify the reason for a phone call, and generative AI helps draft responses. However, it’s not always clear to customers whether they’re communicating with a human agent or an AI system.
The European AI Act is intended to change that. On July 20, 2026, the European Commission published guidelines for the application of the transparency obligations set forth in Article 50 of the AI Act. These obligations take effect on August 2, 2026.
What will change on August 2, 2026?
Organizations that offer or use AI systems must ensure that people can recognize AI. The rules focus, among other things, on:
– direct interactions with AI systems;
– AI-generated or manipulated text, images, audio, and video;
– deepfakes;
– systems for emotion recognition or biometric categorization.
When it comes to customer contact, the first requirement is particularly important: when an AI system communicates directly with a person, that person must be informed of this fact.
This notification must be clear, recognizable, and accessible from the start of the first interaction. Only when it is abundantly clear to an averagely informed and observant person that he or she is communicating with AI may a separate notification be omitted. The European Commission states that this exception must be interpreted narrowly.
A chatbot or voicebot must not pretend to be human
As of August 2, a chatbot may not present itself as if there were a human employee on the other end of the conversation. The same applies to a voicebot with a particularly natural-sounding voice.
A name like “Sophie from Customer Service,” for example, without further explanation, might give the impression that the customer is speaking with an employee. A clearer introduction would be:
Hello, I’m Sophie, our customer service team’s digital AI assistant. I’d be happy to help you with your question.
For a voicebot, the introduction could be, for example:
Welcome. You’re speaking with our digital AI assistant. I’ll ask you a few questions so I can help you as best as possible.
The customer should not have to visit an information page or privacy policy first to find out that AI is being used. The disclosure must be part of the interaction itself and provided immediately at the outset.
Not every automated system is automatically an AI system
The guidelines distinguish between genuine, direct interaction with AI and systems that merely collect data or provide standard responses.
For example, a traditional phone menu where a customer hears “press 1 for sales” is not automatically an AI interaction. A voicebot that understands open-ended questions, generates answers, and carries on a conversation would likely qualify as one.
The obligation applies when four conditions are met:
1. An AI system is involved.
2. The system is designed for a genuine two-way exchange.
3. The AI system communicates directly with the person.
4. The interaction takes place with a natural person.
AI that solely supports processes in the background is not subject to this specific reporting requirement. Examples include AI that automatically categorizes incoming messages without communicating directly with the customer.
Who is responsible: the supplier or the organization?
The AI Act distinguishes between the provider and the deployer of an AI system.
The provider is the party that develops the AI system or markets it under its own name. The deployer is the organization that uses the system as part of its own services.
The responsibility for designing an interactive AI system in such a way that users are informed lies primarily with the provider. In practice, the organization using the chatbot or voicebot must also verify that the notification is actually configured correctly.
A vendor may provide the functionality, but the organization often decides for itself:
– which introductory text to use;
– what name to give the bot;
– how the bot is displayed;
– when the notification appears;
– how a conversation is transferred to an agent.
Transparency must therefore be an integral part of both the technical infrastructure and the governance and management of customer contact channels.
What applies to AI-generated content?
Article 50 also contains rules for AI-generated or manipulated text, images, audio, and video.
Providers of generative AI systems must ensure that such output is technically identifiable, for example, through machine-readable markers. Organizations that publish deepfakes must also clearly label them for human viewers. An invisible technical marker alone is not sufficient.
AI-generated texts on topics of public interest may also be subject to a labeling requirement. This requirement does not apply if the text has been reviewed for content by an expert and is published under human editorial responsibility.
A spell-check or grammar check alone is not sufficient. There must be a substantive review, in which a person can approve, edit, or reject the text.
That doesn’t mean that every email drafted with the help of AI must automatically be labeled “created with AI.” However, organizations do need to consider how AI-generated content is reviewed, approved, and published.
Emotion recognition also requires transparency
Some customer contact platforms use AI to analyze emotions or behavior. The AI Act includes a separate information requirement for people who are exposed to emotion recognition or biometric categorization systems.
Not every form of sentiment analysis automatically constitutes legal emotion recognition. The precise functioning of the system is the determining factor. For example, when biometric characteristics derived from the voice, face, or behavior are used to infer emotions, further legal and technical investigation is necessary.
Organizations that use such features must therefore know exactly:
– what data is being analyzed;
– what conclusions the system draws from that data;
– whether biometric data is used in the process;
– how customers are informed about this;
– what human oversight is in place.
What should customer service organizations do now?
The transparency rules take effect on August 2, 2026. There is no general transition period for the reporting requirement regarding chatbots and voicebots. The limited transition period, which runs through December 2, 2026, applies only to certain technical markings of AI-generated content by systems that were already on the market before August 2.
Organizations can take the following steps:
1. Map out all AI interactions
Identify where customers interact directly with AI. Don’t just look at the website; also consider:
– WhatsApp and other messaging channels;
– mobile apps;
– social media channels;
– telephone voicebots;
– virtual assistants;
– digital avatars;
– AI agents within self-service environments.
2. Review the introduction at
. Make sure it is clearly stated at the start of the interaction that the customer is communicating with AI. Use easy-to-understand terms such as:
– AI assistant;
– digital assistant;
– chatbot;
– voicebot.
A product name that is known only internally is usually not clear enough.
3. Avoid creating a false human identity
Check the bot’s name, profile picture, voice, and writing style. These should not give the impression that the customer is talking to a human employee when that is not the case.
A friendly and human conversational experience is still possible, but the system’s identity must be transparent.
4. Set up a smooth handoff to staff at
. The AI Act does not require that every chatbot be able to transfer calls directly to a staff member at all times. However, from a good customer service perspective, it is wise to make it clear when and how human support is available.
Especially when dealing with complex, emotional, or sensitive issues, a customer should not get stuck in an automated process.
5. Make arrangements with suppliers
Ask suppliers how their solution complies with Article 50 of the AI Act. Discuss, among other things:
– how the transparency notice is configured;
– who is responsible for the text;
– how AI-generated content is technically flagged;
– what logging is available;
– how changes are documented;
– whether emotion recognition or biometric analysis is used.
6. Document decisions and reviews
Document why a system is or is not subject to transparency requirements. Keep records of introductory texts, test results, versions, and approvals.
This not only aids in oversight but also ensures that marketing, customer service, IT, privacy, and compliance teams follow the same guidelines.
Transparency doesn’t have to negatively impact the customer experience
Some organizations fear that customers will trust a chatbot less once they’re explicitly told it’s powered by AI. In fact, the opposite may be true.
A customer who thinks they’re talking to an employee and later discovers that’s not the case may feel misled. A clear introduction sets realistic expectations. The customer knows what the digital assistant can do, what its limitations are, and when an employee is available.
Transparency is therefore not just a legal obligation. It is also a design principle for reliable and human-centered customer interactions.
From Obligation to Trust
The new guidelines make it clear that organizations should not wait for customers to discover on their own that they are interacting with AI. Transparency must be built into the experience from the very first point of contact.
For organizations that use chatbots, voicebots, or generative AI, now is the perfect time to reassess the customer journey. Is it clear what the AI does? Does the customer know when a human agent takes over? And have the text and processes used been verifiably reviewed?
By answering these questions now, AI will not only be used in a compliant manner, but will also become more understandable, more reliable, and more effective.
Pegamento helps organizations design and implement AI responsibly in customer interactions. We do this by integrating technology, processes, employees, and the customer experience. The result is an AI solution that not only works intelligently but is also transparent and reliable.
