By Scott Sumner | AI at Work | Published October 23, 2026

Two conversation chairs face each other on a rehearsal floor. One is solid black, one is purple outline. Practice the awkward part before the client call.

AI role-play gives you somewhere to practice a difficult client conversation before a real customer is on the other end. Use it to work on your questions and explanations, not to predict what that customer will say.

Answer: To practice a client call with AI, describe a fictional situation, give the assistant the customer’s role and ask it to respond one turn at a time. Rehearse your own answers, then request feedback tied to your exact words. Treat the simulation as practice, not customer research.

The useful part is not getting the AI to agree with you. It is discovering which explanation falls apart when someone asks a reasonable follow-up.

Jump ahead

Rehearse the awkward part, not the whole sales script

Imagine a client asks why a website change costs extra. You know the answer: the new request changes the agreed work. But your first explanation sounds like a contract being read aloud by someone who would rather be elsewhere.

That is a fictional scene. It is also a useful rehearsal target.

You do not need an AI-generated script for a thirty-minute meeting. Practice the moment where you explain the change, ask what the customer needs most and offer a next step without promising work you cannot deliver.

My preference is a narrow exercise: one situation, one conversation skill, five minutes. For example, explain a revised scope without blaming the customer. Or ask enough questions to understand a complaint before offering a fix.

There is room for positive practice too. Rehearse explaining a recommendation to a nontechnical owner, inviting a quiet stakeholder into the discussion or asking why a project matters now. Not every useful role-play needs an angry customer.

Give the practice partner a useful brief

OpenAI’s prompting guidance recommends clear, specific instructions, enough context and refinement based on the response. For rehearsal, that means describing the situation instead of saying only, “Act like a difficult client.”

A good brief states what each person knows. The fictional customer might know the original project price but not understand why adding a booking workflow changes the work. They can be concerned and direct without being abusive.

Give the assistant boundaries too. Tell it not to invent past promises, change the agreed facts or concede immediately. Those instructions guide the exercise; they do not guarantee the model will follow every rule. Reset the scene if it introduces a new history you never supplied.

Use invented names and details. You rarely need a real client email, recording or contract to practice explaining a scope change. Keep confidential information out of the rehearsal, and use a tool your business has approved.

If explaining the service is consistently difficult, check the underlying offer. A clearer service page with explicit scope can prevent some of the confusion before a call starts. Rehearsal cannot repair a promise that was unclear from the beginning.

A prompt for a five-minute rehearsal

This is a suggested exercise, not a tested training product or a claim about improved sales results. Copy it into an approved conversational AI tool and adapt the fictional facts.

Help me rehearse a client conversation. You play a fictional business owner. I play the website provider.

Facts: The agreed project is a five-page informational website. You now want online appointment booking. We have not agreed on the extra work, price or timeline. You thought booking would be a small addition.

Your goal is to understand the options and avoid unexpected costs. Be reasonable but unconvinced. Do not invent earlier promises or make up a price. Do not become hostile.

Start with one short customer question. Wait for my answer before continuing. Stay in the customer role until I say “pause and review.” Do not write my replies for me.

When I say “pause and review,” quote one thing I explained clearly and one phrase that was vague or defensive. Identify any commitment I made that the supplied facts do not support. Suggest one better question I could ask. Do not give a numerical score or predict whether a real customer would buy.

Answer in your own words. If you ask AI to write both sides, you are reading a demonstration rather than practicing the conversation.

Text is enough to begin. If speaking is the skill you want to practice, ChatGPT Voice supports spoken conversation. Available options depend on the account, region and app version. Check your workspace’s settings before using it for business material.

For this fictional scenario, a useful response might be: “Let’s look at what you need booking to do. Is the goal for customers to request a time, or to choose a confirmed slot from your calendar?” That question explores the actual requirement before discussing a solution or price.

Ask for evidence, not a confidence score

A score of 92 out of 100 looks official. Without a defined, validated assessment, it is just a number the assistant produced.

I would rather see the sentence that caused trouble.

Ask the assistant to quote your words and explain its criticism. Did you answer the customer’s question? Did you use a term they would need explained? Did you offer to check something, or accidentally guarantee it?

You can disagree with the feedback. The model may misread a sensible boundary as unhelpful, or suggest a polished sentence you would never say. Keep the substance that helps and reject the rest.

Then repeat only that section. Try a shorter explanation or a better question. For an important conversation, ask a colleague who knows the work to review the wording as well.

What a simulation cannot tell you

The fictional customer is not a sample of your market. Their agreement does not validate your pricing, prove demand or predict the outcome of the real call.

This is where rehearsal differs from customer research grounded in real evidence. Use real customer conversations to learn what buyers believe. Use the simulation to practice how you listen and explain.

Nor should the exercise become a way to pressure someone into accepting a bad fit. A good outcome may be a clearer boundary, a smaller project or an honest decision not to proceed.

For sensitive employment, legal or crisis conversations, seek the appropriate qualified support rather than treating a chatbot as your advisor. This exercise is for ordinary business communication, not specialist judgment.

Make the next call easier to have

Pick one sentence you are putting off saying. Build a fictional version of the situation, rehearse it and listen for the part you keep over-explaining.

The point is not to memorize a perfect answer. It is to reach the real conversation with a clearer question and less need to improvise the basics.

If repeated customer confusion starts on your website, bring that question to Sumner Digital. We can discuss whether the service explanation, scope or inquiry path needs attention before adding another AI tool.

Sources

Sources checked October 23, 2026. The rehearsal method, sample prompt and business recommendations are Scott Sumner’s analysis, not findings from a training-effectiveness study.