Skip to content

AI in sales and customer service: offers, replies, rules

By the KI können editorial team8 min read

Written with AI and checked against the sources by a second AI. As of November 16, 2026. More in the notes on AI.

Sales and customer service involve a lot of writing: offers, follow-ups, replies to complaints. An AI gives you quick drafts for this. For them to suit your customers and promise nothing that isn’t true, it needs your facts and clear limits. Here is how that works and which rules apply.

A customer service employee is on the phone at her desk in front of a laptop, a colleague with a tablet stands beside her, a shelf of boxes behind.

Key points

  • The AI drafts replies, follow-up e-mails and meeting guides from your facts. You make the promises.
  • Customer names, contract data and notes from the CRM only go into an AI your employer has approved.
  • Reply templates for frequent questions save the most time. They need fixed facts and placeholders.
  • No artificial time pressure and no invented figures: that is unfair and often prohibited.
  • An AI chat on a website has had to identify itself as AI since August 2026.

Which tasks fit

In sales, an AI like ChatGPT, Claude, Gemini or Copilot is above all a help with phrasing and preparing. But it knows neither your prices nor a customer’s history. Good to hand over are:

  • Reply templates for frequent questions
  • Follow-up e-mails after an offer
  • A guide for a customer meeting
  • Collecting objections and preparing answers
  • Answering complaints calmly and factually
  • Translating product information into plain language

Prices, discounts, delivery times and every promise stay with you. An AI likes to write “of course we will refund” if you don’t stop it. That is why every example below states what the AI must not offer or promise.

The rule of thumb: whatever you wouldn’t promise a customer on the phone without checking first, the AI doesn’t write either. If you are missing a detail, ask your team before the reply goes out.

In the trial lesson you put your first instruction together and get feedback. No account, in a few minutes. Start the trial lesson

Reply templates for frequent questions

Selin works in customer service at a municipal utility. Many e-mails revolve around the same topics: changing the monthly installment, reporting a meter reading, moving house. She has the AI draft a template for each topic and gives it her company’s fixed rules.

Sample instruction
Write a reply template for customers who want to change their monthly installment.
Facts: changes can be made in the customer portal under [menu item] or by phone at [number]. The new installment applies from the next month. One change per month is possible.
Tone: friendly, clear, formal. 100 words at most.
Placeholders in square brackets for name and customer number.
No statements about prices, tariffs or refunds.

Selin’s team lead reads the templates before the team uses them. If rules change, the template is updated and the date noted at the top. That way everyone knows which version applies.

Answering complaints

Jasper runs customer service at an online store for garden furniture. With angry e-mails, the AI helps him strike the right tone. He puts the e-mail into the chat without the name, order number and address.

Sample instruction
Here is a complaint, without names: [text].
What we know: the delivery arrived five days later than announced, and a chair leg was damaged. We will send a replacement part within three working days.
Write a reply, formal, 120 words at most. Apologize once, without excuses.
Offer only the replacement part as a solution. No vouchers, no refund; I decide on those.

Which solution is offered is stated in the instruction. The AI only phrases it. If the solution is missing, it may come up with one that your company doesn’t offer.

Following up on offers and preparing meetings

Owen is a field sales rep for a plumbing supplies wholesaler and looks after installers in his region. Before a meeting, he has the AI make a short guide, and after it, an e-mail with the points discussed.

Sample instruction
Tomorrow I have a meeting at an installation business with eight people, an existing customer for three years.
Goal: I want to present our new range of heat pump accessories.
Write me a guide for 30 minutes: three open questions to start, the key points, two possible objections with answers, a clear next step.
No prices, no statements about subsidy programs. I will check those myself.

After the meeting, Owen dictates bullet points and has an e-mail made from them. He adds the name and the terms only afterwards, in his e-mail program.

Preparing for objections

An AI can collect objections you might not have thought of. But don’t let it write answers that rely on pressure. Good answers take the objection seriously and explain instead of persuading.

Making product information understandable

Data sheets and technical descriptions are often hard to read. An AI can turn them into a short explanation for customers. Give it the public data sheet for this, not internal cost calculations.

Sample instruction
Here is the data sheet of a dehumidifier: [text].
Explain in five sentences at most which rooms it is suitable for and what to keep in mind when setting it up.
Audience: private customers without technical knowledge. Formal.
Take all figures exactly from the data sheet and don’t name any that are not in it.
No comparisons with devices from other manufacturers.

Afterwards, put the explanation next to the data sheet and check every figure. When simplifying, “up to 20 liters a day” quickly becomes “20 liters a day”, and that is a promise.

Customer data and CRM

Notes from the CRM, contract data and call logs are data about people. They don’t belong in a personal AI account. A code instead of the name isn’t enough as long as you know who is meant, and a company in a small industry is quickly recognizable too.

  • For customer data, use only the AI your employer has approved, with a work account.
  • If an AI is built into the CRM, your employer’s rules apply. Ask what it may analyze.
  • For drafts, placeholders like [company] and [contact person] are enough. You add the real details afterwards.
  • Price lists, margins and terms are trade secrets. Don’t put them into a personal account.

Which data may go into which account is explained in detail in the guide Data protection and AI tools.

Selling fairly, even with AI

On request, an AI writes very persuasive texts. But some tricks are unfair and often prohibited, whether a person or an AI phrases them:

  • False time pressure: write “today only” only if it is true. Falsely claiming that an offer is available only for a short time is prohibited toward consumers.
  • Invented figures and quotes: “over 500 satisfied customers” or customer quotes only if you can prove them.
  • Advertising by e-mail: to all recipients, including companies, only with prior express consent. There is a narrow exception for existing customers.
  • Sales calls: to consumers only with prior express consent, to companies with at least presumed consent. If an AI voice calls without a person, that can count as an automated calling machine. Then prior express consent is needed from everyone.

AI chat and AI on the phone

Many businesses are considering having inquiries answered by an AI chat on their website or by a phone assistant. Since August 2, 2026, the EU AI Act has required such a system to identify itself as AI, at the latest with the first message. According to the EU Commission’s guidelines, a note only in the terms and conditions is not enough.

  1. The chat says right at the start that an AI is answering.
  2. There is an easy way to reach a person.
  3. The chat knows only checked facts and says when it doesn’t know something.
  4. Someone regularly reads samples of the answers.

An error in an AI answer is often only noticed when a customer relies on it. Hence the samples. More on the rules in the guide The EU AI Act explained for working people.

Check before sending

  1. Do prices, dates and delivery times match your records?
  2. Does the text promise something you are not allowed to promise?
  3. Is there a figure in it that you can’t prove?
  4. Does the tone fit this customer?
  5. Is the name spelled correctly and added only now?

Having conversations summarized

Some phone systems and video call programs summarize conversations with AI. Use this only if your employer has approved it and the other people on the call have agreed. Recording a conversation without permission can even be a criminal offense (Section 201 of the German Criminal Code). Check every summary before it goes into the CRM: it may contain a promise that was never made.

How to practice this

You practice replies from facts, templates with placeholders and careful checking in the learning app “KI können” in short lessons, including examples from retail and the office, and put the instructions to work in your own AI. The trial lesson needs no account.

Sources

Information as of November 16, 2026. Laws and offers can change; this article is not legal advice.