Practicing interviews with AI, honestly: what helps, what employers allow

A language model makes a good practice partner. It is available at 1 a.m. the night before a recruiter screen, it will ask the same question ten times without sighing, and it can be told to be harsher than a friend would be. Used for preparation, it is one of the more useful tools a new grad has. Used during a live interview or assessment without permission, it is one of the fastest ways to lose an offer.

This guide is about staying on the right side of that line: what AI practice is good for, what it is bad at, what employers have actually written down, and how to behave in the rounds that do invite AI.

What AI practice is genuinely good for

Mock interviews. Paste in the job posting and ask the model to act as an interviewer for that role, asking one question at a time and following up on whatever you leave vague. Tell it not to accept an answer until you have said what you did. A good session feels slightly uncomfortable.

Feedback on your answers. After you answer out loud, give the model what you said and ask for criticism against specific points: was it structured, did it say "I" or only "we," was there a result, was it under two minutes, did it answer the question asked? Ask it to quote the sentence each criticism refers to, so the feedback stays tied to your words instead of drifting into general advice.

Rehearsing STAR stories. Once you have a story bank (our guide to behavioral interviews and STAR stories shows how to build one), have the model play the follow-up interviewer: "Ask me five questions about this story that would expose anything I glossed over." This is where AI is most useful, because follow-ups are where real interviews are decided and the easiest part to under-prepare.

Technical practice. After you have solved a problem yourself, ask for a review of your solution, an explanation of an approach you did not think of, or a similar problem to try next. Run the code rather than trusting the explanation; models can be confidently wrong, and catching that is a skill AI-allowed interviews now score. Our guide to technical screens covers the formats.

Reviewing transcripts after practice sessions. Recording a mock interview and reading the transcript afterwards shows you things you cannot hear in the moment: filler words, answers that wander, questions you never actually answered. Ask the model to mark where each answer lost its thread. One rule first: get consent from everyone in the recording. Federal law requires at least one party's consent, and about eleven states, including California, Florida, Illinois, Pennsylvania and Washington, require the consent of all parties (Reporters Committee for Freedom of the Press). The simple version: ask everyone before you record, and do not record a real interview unless the employer has explicitly agreed.

What it is bad at

• Knowing the employer. A model does not have the company's rubric or question bank, and anything presented as "insider" questions is a guess.
• Criticism. Models can be too agreeable. Ask for the three weakest points in every answer, not for a score.
• Writing your stories. A generated story is generic, and if it is not true it is a fabrication you will have to defend under follow-up questions. Use the model to sharpen real stories, never to invent them.
• Keeping secrets. Do not paste confidential details from an internship, material you signed an agreement about, or interview questions you were asked to keep private.

The line: preparation versus live assistance

The employers that publish candidate AI policies broadly agree on where the line sits. Before the interview, AI help is usually welcome. During it, AI help is prohibited unless you have been told otherwise.

• Anthropic invites candidates to "use Claude to research Anthropic, practice your answers, and prepare questions for us," asks that take-homes be completed "without Claude unless we indicate otherwise," and asks for "no AI assistance" in live interviews unless it says otherwise (Anthropic).
• McKinsey says practicing interview questions with AI is fine, but expects candidates not to use generative AI during assessments and interviews, asks them to disable AI software that might record or assist, and lists possible consequences that include disqualification, a rescinded offer, restrictions on future applications and legal action (McKinsey).
• Datadog permits AI help with a resume as long as it stays accurate, prohibits AI in non-technical interviews and in live coding unless explicitly stated, and warns that "misuse of AI tools, or failure to disclose their use, may lead to disqualification from consideration" (Datadog).
• Amazon tells candidates "please do not use gen AI tools during your interview unless explicitly permitted," and those who do may be disqualified, according to guidelines reported by Fortune. The same report says Goldman Sachs "prohibits the use of any external sources, including ChatGPT or Google search engine, during the interview process" (Fortune).

The pattern is consistent enough to act on: if an employer has not told you a stage allows AI, assume it does not.

What happens when people cross it

One widely reported case involved a Columbia student who built a tool that, in NBC News's description, took screenshots of coding problems and used AI "to spit out solutions in real time," and used it in his own interviews. He told NBC News that several companies "rescinded their offers after he publicly posted about his use of AI," and that a university disciplinary process "resulted in a yearlong suspension." Amazon told NBC that candidates must acknowledge they will not use unauthorized tools, including generative AI, during the interview or assessment process (NBC News).

Set aside whether anyone is caught. Interviews exist to find out whether you can do the work, and a process that hires you on borrowed answers puts you in a job you then have to do without them. Employers know this, which is one reason some have brought back in-person rounds, where nothing on a second screen will help.

For international students the stakes are higher. A rescinded offer during post-completion OPT does not stop your unemployment clock, and an integrity finding can follow you into later applications (McKinsey's policy, for one, lists restrictions on future applications). If an offer is withdrawn for any reason, talk to your DSO promptly about what you need to report.

Disclose, and follow each employer's rules

1. Read the candidate AI policy on the careers site, if there is one. Where it exists, it is the authoritative answer, not a forum post or a prep site.
2. If there is none, ask the recruiter in writing: "Is AI assistance permitted at any stage of this process, and if so, which?" Keep the reply.
3. Applications follow their own rules. Anthropic, for example, asks you to write the first draft yourself and then refine it with Claude. Whatever you use, every fact must be yours; our guide on letting an AI agent apply for you covers the attestations that stay with you.
4. Close practice tools before a real interview, including anything that listens to audio or overlays your screen. McKinsey explicitly asks candidates to disable installed AI software that might record or assist.
5. Ask for accommodations properly. If you rely on assistive technology, request it through the employer's accommodations process in advance rather than using an unapproved tool. McKinsey's policy, for example, makes an exception for approved disability accommodations.
6. Say so when asked. If you used AI in a permitted way, such as tidying a take-home README the rules allowed, disclose it plainly.

In a round that allows AI

Some employers now expect AI in specific rounds. Canva, for example, looks at whether candidates "understand when and how to leverage AI effectively" and can "identify and fix issues in AI-generated code" (Canva Engineering). In these rounds:

• Confirm the rules first: which tool, whether it is provided, and whether anything outside the environment is allowed.
• Use only the provided assistant. Permission for one tool in one round is not permission for your own tools.
• Assume it is visible. CodeSignal, for example, gives hiring teams "a full transcript of candidate-AI interactions, along with a session replay" (CodeSignal).
• Stay in charge. Break the problem down yourself, prompt with specifics, test what comes back, and say out loud why you accept or reject each suggestion.

Where Imba Interview Tool fits

Imba Interview Tool is our Mac app for coding, system-design and behavioral interviews. It runs on your own AI provider and offers practice, live suggestions from a transcript and answers from a screenshot. It is not available to download yet. Each employer's AI policy, described above, applies to it as to any other tool. When you are ready to put the practice to use, start with new-grad roles and internships.