How to use AI for investment banking recruiting prep

Use AI as a practice partner, not as a source of truth or a hidden interview assistant. Give it a narrow task, answer in your own words, ask it for follow-up questions, and verify every technical correction. Before an application task or interview, read that employer’s rules: self-study may be allowed while AI assistance during assessed work is not.

By Angus, the undergraduate building IB Atlas. This is a student practice workflow, not advice from a bank or recruiter.

Separate practice from assessed work

“Can I use AI while recruiting?” is too broad. There are at least three different contexts: private self-study, application materials, and an employer’s interview or assessment. A tool that is useful in the first context may be restricted in the third.

Employer policies illustrate the distinction. HSBC says candidates may use generative AI to research roles, structure applications, and practise common questions, but may not generate responses during live or recorded interviews or submit work they cannot explain. UBS permits help preparing application documents when the content reflects the candidate, while requiring assessment tasks and video-interview answers to be completed independently. Read the exact instructions for your role and location because policies can differ and change. Sources: HSBC careers and UBS careers.

A five-step AI practice workflow

Start with one skill you can observe. “Help me get better at technicals” is vague. “Ask me one DCF question at a time and probe my assumptions” gives you a session you can evaluate.

• Choose a narrow objective. Pick one technical concept, one behavioral story, or one firm-research question. State the audience and the time limit for your spoken answer.
• Set the practice rules. Ask for one question at a time. Tell the model not to reveal an answer before you respond. For behavioral work, give it only facts that are true about your experience.
• Answer aloud without assistance. Speaking exposes unclear transitions and missing reasoning that can stay hidden when you edit a written answer. Record yourself only in your own study session, never in an employer interview that prohibits recording.
• Invite follow-ups. Ask, “Which assumption did I leave unexplained?” and “What would you ask next?” Then answer the follow-up before reading suggested wording.
• Verify and rewrite. Check technical mechanics against a reliable source or a worked example you trust. Mark each model correction as verified, corrected, or unresolved. Keep the final explanation in language you can defend without the tool.

What a useful session sounds like

For a DCF session, ask the model to begin with “Walk me through a DCF” and then challenge one step at a time. Useful follow-ups include why the cash flow and discount rate must match, which assumptions drive the valuation most, and what changes when a key assumption moves. The purpose is not to collect a polished script. It is to find the first place where you cannot explain why.

For a behavioral session, provide a true outline of one experience and ask the model where the sequence becomes hard to follow. Do not let it add achievements, numbers, or motivations. For networking, use AI to organize public research or tighten a draft, then check every fact and rewrite the note in your own voice.

AI confidence is not verification

A fluent answer can still be wrong. NIST calls this risk “confabulation”: generative systems can confidently present false or internally inconsistent content, including invented logic or citations. That is why a model’s confidence, repetition, or detailed explanation does not verify an accounting treatment, valuation convention, firm fact, or recruiting rule. Source: NIST’s Generative AI Profile.

Use a simple check: can you point to a reliable source, recompute the example, and explain the reasoning under a follow-up? If not, label the point unresolved. Do not ask AI whether you are “interview ready” or likely to receive an offer. A practice tool does not know a particular interviewer’s rubric, applicant pool, or decision, and feedback from one session cannot predict a hiring outcome.

How this applies inside IB Atlas

IB Atlas is a student-built public-beta workspace. It uses AI in parts of spoken practice, study feedback, market reading, and networking drafting. Treat those outputs as prompts for review, not verified answers or employer assessments. Check important claims, edit drafts, and keep your actual applications and interviews consistent with the employer’s instructions. Access is currently free during the public beta.