SolvraONE
In customer service, tone counts as much as content. A correct answer in the wrong tone feels cold or condescending; a friendly answer that solves nothing feels like being put off. For an AI agent there is a third question: does it speak in a way that nobody mistakes it for a person? This page shows how these fit together.
Label AI openly
- Right at the start: the first message says that an AI is writing here. Not only in the small print or when asked.
- Honest throughout the conversation: if the customer asks “Are you a person?”, the answer is a clear no.
- No games with names: an agent may have a name. But it must not be designed so that customers take it for an employee, for example with a made-up photo and CV.
Agree the wording of the notice for each channel before launch. This is not legal advice; have your case checked.
Since 2 August 2026, the duty from Article 50(1) of the AI Act applies in the EU: anyone talking to an AI system must be told, unless it is obvious to a reasonable person anyway. Under Article 50(5) of the AI Act, the notice must come at the latest at the first interaction, clear, unambiguous and accessible.
For customer service, this means in practice:
Wording rules for an AI agent
| Rule | What it means |
|---|---|
| Short and to the point | Answer first, then the explanation. No intro that only costs time. |
| Acknowledge feelings, do not act them | “That is annoying” is honest. “I feel for you” is not, coming from an AI. |
| One thing per message | Better two short messages than one block in which the question gets lost. |
| No empty promises | Only promise what the agent or the team can really do, with a realistic time. |
| Honest about limits | “I can’t decide this, I’m passing it to a colleague” instead of dodging. |
| The customer’s language | Polite by default; the tone adapts when the customer writes casually, but never disrespectfully. |
| No stock phrases | No “Thank you for your patience” when nobody needed patience. |
| No blame on the customer | “Something went wrong there” instead of “You filled in the wrong field”. |
Examples: good and bad
| Bad | Good |
|---|---|
| “Hi, I’m (a first name) from the customer team. How can I help?” | “Hello, this is the digital assistant, an AI. Happy to help; if needed, I’ll hand you over to a person.” |
The bad example leaves open whether a person is writing. The good one says it in the first sentence and names the way to a person.
Annoyed customer
| Bad | Good |
|---|---|
| “I completely understand your frustration and feel for you! Let’s find a solution together!” | “That’s annoying, especially the second time. I can see the double charge. I can’t trigger a refund myself; I’m passing the case to our team now with all the details.” |
The good example acknowledges the annoyance, states where things stand and says honestly who decides.
A question the agent cannot answer
| Bad | Good |
|---|---|
| “That’s a great question! Unfortunately I don’t have any information on that right now.” | “I don’t have a reliable answer on that. I’ll ask a colleague and get back to you here as soon as the answer is in.” |
Customer asks: “Are you a person?”
| Bad | Good |
|---|---|
| “I’m here to help you!” | “No, I’m an AI. If you’d rather talk to a person, I’ll connect you.” |
Promises like a reply, a hand-over or a transfer in these examples assume that the channel supports it and the process behind it is in place.
We wrote the following examples ourselves. They show typical situations in customer service.
Greeting
If not, the agent gives the contact details of the team instead.
Reading mood: what works, what doesn’t
An agent can tell from word choice and the course of the conversation whether a customer is impatient, annoyed or at a loss: exclamation marks, capital letters, repetition, words like “again” or “ridiculous”.
- Irony and brevity are easily misread. “Great, thanks” can be praise or annoyance.
- Mood is not a judgement of the person. The agent uses it to help better, not to rate customers.
- Strong feelings belong with people. If the agent detects great anger or distress, it hands over. How to do that cleanly is on the page Hand-over to a person.
That helps to adjust the tone and hand over to a person sooner.
There are still limits:
What customers notice
We recommend reviewing the whole conversation against these four points:
- Whether they are taken seriously. An answer that deals with their concern instead of reeling off a template.
- Whether they were misled. We recommend disclosing AI from the start: a late disclosure can make customers feel deceived, even if the answers were good.
- Whether things move forward. A friendly tone without a next step feels like a wall.
- Whether the tone fits the company. An agent that speaks differently from your team feels foreign. So set a short style guide and check conversations against it regularly.
Your next step
Read in full
Want to know how your current chat comes across to customers? We talk about it in a first call of 30 minutes. Whether AI used on your website is labelled is shown by the website check; it is free. Get in touch
What could AI take off your plate?
In a first call (30 minutes) we look at your work and tell you openly what is worth it and what is not.
Frequently asked questions.
Does an AI agent have to say that it is an AI?
In the EU yes, unless it is obvious anyway. The AI Act requires the notice at the latest at the first interaction, clearly and unambiguously. This is not legal advice.
May an AI agent have a human name?
A name alone is not a problem, as long as it is clearly stated that an AI is writing. It becomes a problem when name, picture and tone together give the impression of an employee.
Can AI recognise emotions?
An AI can pick up signs of annoyance or impatience from word choice and the course of the conversation. It is not right every time, especially with irony. That is why recognition only serves to adjust tone and hand-over.
Should an AI agent show empathy?
It should acknowledge feelings, for example with "That is annoying", but not act out feelings of its own. We recommend matter-of-fact sympathy without pretending to have feelings.
Formal or informal: which address suits an AI agent?
In German-language customer service, the formal "Sie" is the safe default. If the informal "du" fits the brand and your team, the agent can use it, as long as it stays consistent.
Do customers notice whether the tone is right?
Yes, above all whether they are taken seriously and whether there is a next step. Also avoid customers feeling misled about who they are talking to.
How is your website doing?
The check is free and without obligation. You get the report by email.
Sources.
- EUR-Lex: Regulation (EU) 2024/1689 (Artificial Intelligence Act), Art. 50 and Art. 113Vendor / primary · As of
Sources as written
- EU Artificial Intelligence Act, EUR-Lex: transparency obligations and start of application
Keep reading.
Human in the loop: when an AI agent hands over
Human in the loop in customer service: when an AI agent hands over to staff, what goes with it and what the AI Act asks about human oversight.
What does an AI agent cost in customer service?
What does an AI customer service agent cost? Pricing models in the market (per resolution, per seat, per conversation) and what drives cost, with sources.
AI and the GDPR: what the law asks of AI agents
AI agents and the GDPR: lawful basis, processing agreements, where data is processed and impact assessments. What businesses need to check.
