AI in Consulting Interviews: What MBB Firms Do in 2026

Do consulting firms use AI in case interviews in 2026?

Mostly no. As of 18 October 2026, the standard case interview at McKinsey, BCG and Bain is still a conversation between a candidate and a human interviewer, with AI banned during the interview itself. Three AI exercises have been reported in selected offices and rounds, but none appears on an official careers page. Treat AI exercises as possible extras on top of a normal case.

CaseCoach reported in September 2026 that none of the hundreds of candidates it surveyed that summer had been tested on AI skills. That distinction matters for preparation. Your core skills stay the same as in the complete case interview guide: structure, math, exhibit reading and a clear recommendation. AI exercises add one layer, which is judging machine output under time pressure.

What each firm has confirmed, reported and not said

Every claim below carries one of three labels. Official means the firm states it on its own website or channel. Reported means a named news outlet published it. Candidate report means it rests on candidates' accounts relayed by a prep provider. All entries were checked on 18 October 2026.

McKinsey: Lilli, the practice tool and the client avatar

McKinsey's official position is simple. Its interview preparation page welcomes AI for polishing a resume, practicing questions and explaining frameworks. It forbids using AI to generate answers during an interview, using it in assessments unless permitted, and inventing achievements. It also asks virtual candidates to switch off AI note-takers.

The Lilli exercise is reported. Fortune wrote in January 2026 that McKinsey is piloting a final-round task with Lilli for US candidates, based on the prep firm CaseBasix; McKinsey did not respond. CaseBasix describes the test as instructing the tool, reviewing its output, applying judgment and giving a structured answer, with no advanced prompting required.

Two newer developments are better documented. Business Insider reported in May 2026 that McKinsey launched a free AI practice tool for quantitative cases, open worldwide to entry-level applicants after initial screening. In October 2026, the Wall Street Journal reported a pilot in which some North American undergraduates speak with an AI client avatar for about 10 minutes while human assessors score them. For the rest of the McKinsey process, see the McKinsey case interview guide and the McKinsey Solve guide.

BCG: what its GenAI rules allow and forbid

BCG publishes the most detailed candidate rules of the three firms. GenAI may help you proofread a resume, practice sample prompts and reflect on past experiences. It may not be used during the online case, during any live interview, or to invent or exaggerate experiences. BCG names Zoom's AI Companion as an example of a banned meeting assistant.

BCG also states that AI never makes hiring decisions and that recruiters review every application. CaseCoach reports that BCG is testing AI approaches in selected offices; BCG's own pages describe no AI exercise. The BCG case interview guide covers the standard format.

Bain: what its talent head says and what candidates report

Bain's official signal comes from an August 2026 episode of its Beyond the Bio podcast, where global head of talent acquisition Ron Kermisch said Bain is hiring more people, including engineers and data scientists, and values technical curiosity. The episode does not describe AI in interviews.

The interview detail comes from candidates. StrategyCase reports that candidates in a European partner round received instructions for a written case followed by a discussion that included about 10 minutes of ChatGPT on a Bain laptop. Office coverage, weighting and the exact task remain unknown. The Bain case interview guide explains the rest of the process.

What interviewers assess when you work with AI

An interviewer watching you use a tool can see three things: what you ask for, what you accept and what you reject. The prompt shows whether you framed the problem before typing. The acceptance shows whether you checked the output. The rejection shows judgment, and judgment is the scarce skill.

I interviewed candidates from the other side of the table, and the question I asked myself was always whether I could put this person in front of a client. AI changes nothing about that test. A consultant who forwards a client's spreadsheet without checking it fails in the same way as one who forwards a chatbot's answer unchecked.

A five-step drill for AI-assisted cases

Practice with any chatbot. Ask it to estimate annual e-bike sales in the Netherlands, then audit the answer. Suppose it replies: 18 million people, 40% use an e-bike, so 7.2 million e-bikes, replaced every 3 years, so 2.4 million sales a year.

1. Restate the question. Annual unit sales of new e-bikes, one country, one year.
2. List every assumption. Population, share of users, bikes per user, replacement cycle.
3. Attack the weakest two. A 40% share of all residents includes infants. A 3-year replacement cycle is short for a durable, expensive product.
4. Rebuild with your numbers. Assume 30% of residents ride an e-bike (5.4 million bikes) and an 8-year life: about 0.7 million sales a year. These are illustrative assumptions; the point is the method.
5. Present the corrected answer. "The tool overstated demand more than threefold, mainly through the replacement cycle. My estimate is roughly 0.7 million, and I would validate the cycle with dealer data."

Step 5 is the same discipline as any case close: answer first, then reasons.

Rules to follow so AI use never costs you an offer

• Use AI freely before the interview, for practice cases and feedback on your stories.
• Never run AI during a live case or online test unless the firm sets up the exercise.
• Close every AI tool, plugin and meeting assistant before a virtual interview.
• Never let AI write experiences you did not have. Both McKinsey and BCG call this out.
• In an AI exercise, narrate your checks out loud so the interviewer can score your judgment.

The judgment skills in this drill are the same ones I teach in Cracking the Case, which includes three fully worked candidate cases with the synthesis written out. For timed practice at catching wrong numbers in dense material, The Key to the PST offers 26 written-case questions with a full answer guide.