TL;DR: Ask who can change the decisions that determine your success: customer scope, model releases, quality thresholds, budget and operating support. Test the answer against a recent example. A promising title is weak evidence of what you will be able to accomplish.
Start with the decision you would own
Two vacancies can both say “AI product manager” while offering very different work. One PM shapes model evaluations with researchers. Another owns the customer workflow around a purchased model. A third runs the control plane through which enterprise administrators permit and pay for usage.
Choose the work you want to do, then verify that the role contains it. As of 29 October 2026, OpenAI's Core Models posting emphasizes the connection between user needs and model development. Its separate API Infrastructure posting includes metering and enterprise controls. These are employer-stated responsibilities, not a promise about a particular interview or your eventual allocation of time.
Prepare one factual question about a decision, rather than a long speech designed to demonstrate how much AI vocabulary you know. “How did the last model regression change the release plan, and who made that call?” gives the team room to explain how it works. They may be unable to discuss a confidential incident; an anonymized example or a description of the decision process can still help.
Compare responsibilities with the means to deliver
This is a fictional comparison of two role descriptions. Neither represents a real employer.
| Responsibility | Role A: assistant product | Role B: model platform |
|---|---|---|
| Accountable outcome | Reduce the time to a correct service note | Increase successful integrations without uncontrolled spend |
| Direct product choices | Workflow scope, review experience and rollout | API behavior, admin controls and developer onboarding |
| Shared dependency | Central team chooses the model | Serving team owns capacity and reliability |
| Evidence access | Can observe approved customer workflows | Can inspect permitted usage and support patterns |
| Unresolved question | Can the PM delay a harmful model change? | Can the team reserve capacity for a contracted guarantee? |
Neither role requires sole authority over every dependency. Influence and shared ownership are normal. The problem is an outcome for which no one can explain how a blocking decision gets resolved.
Suppose Role A offers responsibility for retention but denies access to customer evidence, release timing and the review experience. Ask which decisions remain in scope and how the team resolves that mismatch. A manager who can name a recent escalation, its decision owner and resulting change offers better evidence than “we all collaborate closely.”
Follow one product decision through the organization
Follow the flow, one stage at a time.
Illustrated sequence; animation speed does not represent processing time.Read diagram description
Name the expected outcome, locate product decisions, trace dependencies, ask for a recent example, check evidence access, identify who resolves disagreement, record gaps and choose a follow-up question. Revisit the decision with the new evidence.
The sequence is a preparation aid. It is not an employer assessment or a score that predicts whether you should accept a job. Record what the team said, what remains uncertain and which uncertainty matters to your own goals.
Ask questions that reveal the operating work
| What you need to learn | A question worth asking | Useful evidence in the answer |
|---|---|---|
| Product scope | What customer outcome should this PM change in the first six months? | A defined workflow, baseline or discovery objective |
| Research dependence | When a promising capability misses its evaluation target, what ships instead? | A fallback, a decision date and a clear customer commitment |
| Release authority | Who can stop a model rollout after a severe regression? | An accountable owner and a practical stop mechanism |
| Learning access | How does the team obtain customer examples it is allowed to inspect? | Consent or access rules, sampling and a path to users |
| Operating burden | Who handles disputed actions, support escalations and evaluation maintenance? | Staffed ownership rather than an assumption that automation removes the work |
| Commercial pressure | What happened when a large customer asked for a feature outside the roadmap? | A real trade-off, acceptance terms and the work deferred |
| Your growth | Which part of this job will stretch me, and who will review that work? | Specific collaboration and feedback arrangements |
Do not interpret a missing answer as dishonesty. Early teams may still be defining the job; a clearly named unknown is useful information. The next question is whether you would have the support and authority to resolve it. For an established product, repeated uncertainty about incident ownership deserves closer investigation.
Evaluate the interview conditions too
Ask the recruiter for the format, allowed tools, expected preparation effort and any accommodation process before the relevant assessment. Tool rules can vary inside one company. OpenAI's general interview guidance, checked 29 October 2026, directs candidates to the rules for their particular interview. Anthropic's candidate policy, updated July 2025 and checked on the same date, permits preparation with AI but defaults live interviews and take-homes to no AI assistance unless otherwise indicated.
For a take-home, clarify the time budget, whether the exercise uses fictional data and what the panel expects you to defend. Use only material you have permission to share. A request to demonstrate prior impact does not require disclosing customer records or a former employer's confidential evaluation set; discuss an anonymized account or a clearly labeled reconstruction.
Compensation, working arrangements and accommodations are legitimate parts of evaluating a role. Route them to the person who can answer, without treating them as evidence that you lack ambition. The older PM interview literature is useful on company research, but advice about avoiding reasonable employment questions should not become a rule for today's candidates.
Turn the answers into a decision
Write a short private note after each conversation: what decision you would own, one dependency you can work with, one unresolved concern and the evidence needed to resolve it. Keep observations separate from your interpretation. “They have no release owner” is a stronger claim than “the interviewer could not identify the release owner”; only the latter may be supported.
If the company offers a role with substantial uncertainty, decide whether discovering and establishing the operating model is work you want. Ask for the expected scope to be confirmed with the hiring manager. A role can be attractive because the problems are difficult; it becomes a poor fit when the work differs from what you want to learn or the team cannot support the responsibility it assigns.
Practise the commercial discussion with the design-partner case, then prepare your own portfolio defense. The employer is evaluating your judgment, and you need enough evidence to exercise it too.