AI Case Interviews: How to Prep When an AI Tool Joins the Room

There are now three quite different ways AI turns up in a job interview. It can be the interviewer — an automated screener asking the questions. It can be the question — “how do you use AI in your work?” And, increasingly, it can be a tool sitting on the desk in front of you, which you are expected to use while someone watches.

This post is about the third one. It is the newest, it gets the least sensible advice, and the candidates who handle it well are usually not the most technical people in the room. (If you're facing the first kind, read our guide to AI interview screeners.)

What is actually happening

The best-known example is McKinsey. In January 2026, reporting traced to the consulting-prep firm CaseBasix said some US final-round candidates were being asked to complete case-style tasks using Lilli, McKinsey’s internal generative AI tool. Former McKinsey consultants who coach candidates report the trial spread to select European offices later that month, that it appears in in-person final rounds rather than virtual ones, and — importantly — that McKinsey has told candidates it is not yet scored.

Bain is reported to be introducing its own AI-enabled interview component from mid-2026, with less detail public so far. In tech, Meta began replacing one of its two onsite coding rounds with an “AI-enabled” round from October 2025, where candidates work in a codebase with an AI assistant in the next panel.

A fair caveat: almost everything known about these formats comes from candidates and prep companies, not from the firms themselves. The details will shift. The underlying idea is stable, though, and it is worth understanding.

First, find out which room you’re in

Before you prepare for an AI-allowed interview, check that you are actually in one. The rules run in both directions, and in Australia the default is often the opposite.

The Department of Foreign Affairs and Trade, for example, tells graduate applicants plainly that AI tools are not to be used during the interview or any assessment task — and links that to integrity and security clearance. Business Insider has reported that Amazon told its recruiters to disqualify candidates found using AI in interviews. Using a tool where it isn’t allowed costs you the job. Refusing to touch one where it is expected can make you look slow.

So ask. It is a normal, professional question, and a recruiter with a real process will have a real answer:

“Could you confirm how AI tools are handled in the next stage — will one be provided, is it optional, or should I treat the round as AI-off?”

If you’re applying for public sector roles, assume AI-off for every live interview and assessment unless you are told otherwise in writing.

What is being scored

When an employer puts an AI tool in front of you, they are not testing your prompting. They are testing whether your judgement holds up when a fast, confident and occasionally wrong assistant is sitting next to you. Across the reports, four things come up again and again:

• Structure: did you frame the problem yourself before asking the tool anything, or did you let its first answer set the shape of your thinking?
• Judgement: did you check the output, catch what was wrong or generic, and decide what to keep?
• Iteration: when the answer was vague, did you redirect the tool with a sharper question, or accept it and move on?
• Communication: could you explain, out loud, why you trusted one part and discarded another — and land a recommendation in your own words?

That is the same list a good case interview has always scored. The tool is the medium. You are still the subject.

The loop that works

Candidates who do well tend to treat the AI like a capable but junior team member: quick, useful, and in need of supervision. In practice that looks like a short loop you repeat:

1. Frame first. Spend the first minute saying how you would break the problem down — before you type anything. Your structure, not the tool’s.
2. Ask narrow questions. “What are the main cost drivers for a regional grocery chain?” beats “Solve this case.”
3. Read it critically. Look for numbers you can’t trace, assumptions that don’t fit the client, and answers that would apply to any business.
4. Decide out loud. “I’ll keep the first two points, but the third assumes a national footprint and this client is in two states, so I’m setting it aside.”
5. Synthesise yourself. The final recommendation should be in your words, with your reasoning. Reading the tool’s summary aloud is the fastest way to look junior.

What it sounds like

Imagine the case: a Queensland retail chain has seen profit fall 18% over two years, and you have an AI tool to help. First, the weak version:

“Okay, I’ll ask the AI what could cause falling profit… It says rising costs, more competition and changing customer behaviour. So I think the recommendation is to cut costs and invest in digital.”

Why it fails: no structure of the candidate’s own, no checking, no link to this client. The tool did the thinking, and the answer would fit almost any retailer in the country. Now the strong version:

“Profit is revenue minus cost, so I want to know which side moved. I’ll start by asking the tool to split the two-year change into revenue and cost movements using the exhibit… It’s saying costs are flat and revenue is down, but it’s used the total store count — two stores closed last year, so I want like-for-like. Let me re-ask on the same-store figures… That changes things: same-store revenue is roughly steady, so most of the drop is the closures. My working hypothesis is that this is a footprint decision, not a demand problem, and the next thing I’d test is what those two stores were contributing.”

Why it lands: the candidate set the structure, used the tool for the legwork, caught a real error, and explained every decision. That is exactly what the interviewer is there to watch.

The mistakes that cost the most

• Treating it as an answer engine. Vague prompts get generic output. If the first answer is weak, change the question rather than accept it.
• Going silent. The interviewer can’t read your screen as fast as you can. Narrate what you’re asking and why, especially while you wait.
• Avoiding the tool to look independent. If it’s provided, use it. Ignoring it signals discomfort, not brilliance.
• Letting the tool pick the framework. Once its structure is on screen, it is hard to think past it. Say yours first.
• Forgetting the basics. At McKinsey, the case interviews and the Personal Experience Interview still decide the offer. The AI round sits on top of them, not instead of them.

A two-week prep plan

This should be a light layer on top of your normal preparation, not a replacement for it.

• Days 1–5: keep doing your usual cases or practice questions, but open a public AI tool (ChatGPT, Claude or Gemini) in a second window. Before each prompt, say your structure out loud.
• Days 6–10: on every practice case, find at least one thing the tool got wrong or too generic, and say how you’d correct it. Keep a short list — those catches are your best material.
• Days 11–13: run full timed cases with the tool, narrating throughout. Then run one with the tool switched off, so you’re ready if the round turns out to be AI-off.
• Day 14: rest, confirm the format with your recruiter, and check your setup if it’s virtual.

Keep it in proportion

If you’re interviewing at a firm running one of these formats, it is worth a few hours of practice. It is not worth panicking over, and it is not a reason to abandon what already works. The skills being watched are the ones that have always got people hired: a clear structure, sound judgement, and the ability to explain your thinking under a bit of pressure.

Practice thinking out loud

The hardest part of an AI-enabled interview isn’t the tool. It’s keeping up a clear running commentary while you work. That is a spoken skill, and it only improves with spoken practice.

The AI interview coach in the Practice Lab lets you answer out loud and get a scored report on structure and specificity — a good way to hear whether your reasoning comes across as clearly as it sounds in your head. For the wider picture, read How to Build Genuine AI Fluency Before Your Interview and How to Talk About Using AI Tools in a Job Interview.