McKinsey Tests AI Clients: What Applicants Need to Show in Conversation

How does an applicant respond when a client introduces a new priority halfway through a conversation? McKinsey wants to simulate situations like that with AI. The Wall Street Journal reports a new component in selected entry-level interviews this fall. The interesting idea is that a simulation can show how someone responds to follow-up questions, rather than simply delivering prepared answers.

Key takeaways

• An AI avatar plays the client in selected McKinsey interviews. This is a limited trial, not a uniform worldwide process.
• According to t3n’s report, the conversations last about ten minutes, with humans observing and evaluating them. The client simulation itself does not grade candidates.
• McKinsey separately describes another chatbot that checks objective eligibility requirements. Its AI assessment is reviewed by humans before a decision.
• The company welcomes AI for preparation. During an assessment, it is prohibited unless specifically authorized.

A role-play with a counterpart who can change course

A traditional case interview tests how someone breaks down a problem and explains their thinking. The new client simulation adds another perspective: can the person adapt a recommendation to the needs of the person across from them? According to t3n, the initial trial targets students seeking business analyst positions in North American offices. The original Wall Street Journal report describes an introduction for some entry-level candidates this fall.

This shifts attention from a finished argument to the course of a conversation. A plausible recommendation can become inappropriate when a client adds constraints. Someone who continues working through a prepared solution demonstrates something different from someone who asks questions and explains a revised recommendation. The AI supplies the counterpart; humans still assess the person.

As an illustrative example, consider advising a client on a new location. Initially, low rent might be the priority. During the discussion, the client adds that employees need to reach the site by public transportation. A helpful response would bring both goals together and explain the trade-off. That takes more than producing an answer as quickly as possible.

Role-play like this could provide employers with another work sample. That does not establish that it leads to better hiring. The accessible reports provide no independent evidence that the simulation predicts later job performance more reliably than existing interviews. The technical ability to hold a conversation is not enough to answer that question.

Why an AI interview can mean different things

McKinsey’s own privacy notice for AI-assisted interviews describes a different use case: a chatbot comparing information with predefined eligibility requirements, such as experience or language skills. The company says it does not assess personality or soft skills. Recruiters review its assessment before a decision, and no one is rejected solely because of the chatbot’s output.

That description should not be read as scoring instructions for the new AI client. In one process, software collects and checks information against a job’s requirements. In the other, it creates a conversation for humans to observe. Combining both under the label of AI recruiting obscures the essential question: what job does the machine actually perform?

For applicants, a specific explanation in the invitation would therefore be more useful than a general AI notice. Relevant questions would include the system’s role, the skills being observed, and how the subsequent evaluation works. Even a conversation graded by humans can be recorded or processed technically. The information for that particular format should explain what data it involves, rather than requiring applicants to infer that from another process.

Preparation is allowed; a hidden answer generator is not

The official careers page draws a clear line: AI can help polish a resume, explain concepts, or practice interview questions. McKinsey rejects invented experience and automatically generated real-time answers during an interview. Its assessment rules generally prohibit AI use unless it has been specifically permitted.

A useful practice prompt might therefore be: play a client, ask follow-up questions one at a time, and change one constraint after I explain my recommendation. Candidates can then check for themselves whether their answer addressed the new information. This is a preparation suggestion, not a reproduction of an official McKinsey test or a guarantee of how it will be evaluated.

The published integrity rules also require candidates to disable installed AI software that could unintentionally record or assist them. This has become a practical issue: as ChatGPT’s new access to Chrome tabs illustrates, assistants may already be available where a conversation or task takes place. A helper that is useful in everyday work can become unauthorized assistance during an assessment.

The invitation and the recruiting team’s explanation therefore govern the specific application process. A company’s own use of AI does not automatically authorize applicants to use their own assistants. Conversely, an explicitly designated AI exercise should not be confused with a general prohibition on tools. Unclear instructions can be clarified before the appointment.

The real addition is another work sample

The client simulation creates an understandable opportunity: communication can be observed in a specific situation, instead of only being discussed. Its value would be especially convincing if applicants received a clear explanation of what makes a conversation good. Listening, asking questions, and explaining a changed recommendation can be discussed much more concretely than a broad attribution of personality.

As the trial continues, a key question would be whether different people encounter comparably demanding situations and whether human evaluations remain consistent. An avatar alone does not solve that problem. McKinsey’s experiment should be judged on whether the additional work sample actually reveals skills relevant to the job and makes the application process easier to understand.