What an AI Client Simulation Tests (and How to Practice for It)

Talking to an AI that plays the role of your client can feel strange. You know it is software, you know people are watching how you handle it, and you are still expected to behave as though you are in a real meeting. If that worries you, you are in good company.

McKinsey is piloting a short AI client conversation with some entry-level candidates in select North American offices, and other consulting firms may follow, though none has confirmed it, so it is worth knowing what you could be walking into. Here is the good news. A client simulation is not a trick. It is testing a handful of habits that good consultants, advisers and account managers show every day, and every one of them can be practiced. This article covers what is being tested, what a weak and a strong response look like, and how to practice each skill before the day.

What is an AI client simulation?

It is a live, two-way conversation with an AI playing a client, customer or stakeholder. You get a situation, you talk it through, and the AI responds and sometimes changes the brief halfway. McKinsey is piloting this format with some entry-level candidates in North America, and other consulting firms may follow, although I have not seen any confirm it. It is a different thing from using an AI tool during the interview, which I cover in AI Case Interviews.

It is also not a test of how clever you are. Nobody expects a perfect answer from a client who is making things up as they go. What the format shows is how you behave in a conversation that does not follow a script.

One caution. Each employer runs its own version, so read the instructions you are sent and ask the recruiter what to expect. I can describe the skills these simulations are built to surface, but I cannot tell you how any particular firm scores them.

What is actually being tested?

Five habits show up again and again in client conversations, and they are the habits that make someone good to work with.

1. Asking clarifying questions. Do you find out what the client actually needs before you offer a solution, or do you start solving the first thing you hear?
2. Responding to change. When the client adds a constraint or changes their mind, do you adjust calmly, or defend your first idea?
3. Acknowledging concerns. When the client is frustrated, do you show you have heard them before you move to the fix?
4. Giving a clear recommendation. Do you commit to a view and say why, or list options and leave the client to choose?
5. Agreeing a next step. Do you close with who does what and by when, or let the conversation fade out?

You will see these five again in the practice steps below, because each one can be trained on its own.

What a weak and a strong response look like

Imagine the AI client opens with: "Our customers are angry about late deliveries and I need this fixed."

Weak: "I'd start by hiring more drivers and adding a tracking app. That should fix it."

This jumps to a solution, skips the client's frustration, and has no idea whether the problem is drivers, warehouse timing or something else.

Strong: "That sounds like a stressful few weeks, and I want to get this right. Can I ask two things first? Are the delays across all regions or just some, and what has changed recently? Then I can suggest where to focus."

The strong answer acknowledges the concern, asks before it recommends, and stays open to what comes back.

Now suppose the client replies, "Actually, the budget just got cut by half." A weak response defends the original plan. A strong one says, "Thanks for telling me. With half the budget I'd do one thing first: fix the biggest source of delay, and we can review the rest next month. Does that work for you?" It adapts, commits to a recommendation and asks for agreement.

How to practice each skill

You do not need special software to start, but you do need to speak out loud. Typing a tidy answer is a different skill from thinking aloud with someone who interrupts.

• Clarifying questions: Before any practice conversation, set a rule that your first two turns must be questions. Notice how much harder that is than it sounds.
• Responding to change: Ask a friend, or an AI tool, to change one fact halfway through. Practice saying "thanks, that changes things" before you adjust.
• Acknowledging concerns: Use one plain sentence that names the feeling, such as "I can see why that is frustrating." Then move on. Over-apologising reads as nervous, not caring.
• Clear recommendation: Finish every practice run with "My recommendation is X, because Y." If you cannot say it in two sentences, you have not decided yet.
• Next step: End with who, what and when. "I will send you the plan by Friday. Can you confirm the budget by Wednesday?"

Record one run and listen back. Count how many times you asked a question and how many times you jumped to a fix. Most people are surprised.

If you want a safe place to do this, the Practice Lab has client and stakeholder conversation practice for All Access members, built around these same five skills. It is practice for the habits, not a copy of any employer's assessment. For how to handle AI in other parts of the process, see How to Prepare for an AI Job Interview.

A last word

It is easy to feel that an AI client has the upper hand because you cannot read it the way you read a person. It helps to remember the AI is not looking for a flawless performance. It is giving you room to show how you think when the brief shifts, and calm, curious people tend to do that well.

If a conversation like this goes badly in practice, treat it as information, not a verdict. A rough practice run is the cheapest place to find out what to work on.

If you would like a human to practice with, I offer 1:1 coaching in 45, 60 and 90 minute sessions.

Once a month I share what job seekers are struggling with and what recruiters want to hear instead: The Practice Lab Edit.