Original practice question. The scoring rubric is our teaching guide. How we research and score answers
TL;DR: Use STAR-L: situation, task, action, result, and learning. Keep situation and task short, spend most time on your actions and the reasoning behind them, give a result with numbers where possible, and close with a learning that changed how you or your team work. For AI roles, choose stories that show judgment about uncertainty, evaluation, risk and trade-offs.
How to approach it
Explain the structure, how to allocate time within an answer, how to prepare a story bank, and what makes stories strong for AI PM roles specifically.
A strong answer
STAR is the structure employers themselves name. Amazon's product manager interview guidance tells candidates to "use the STAR method to frame your responses", says each interviewer typically asks two or three behavioural questions, and asks for "metrics or data where applicable" (checked November 2026). STAR-L, adding the learning, is our extension, and the time split below is our guideline for a two to three minute answer, not an employer rule.
The reason structure matters is on the employer's side too. Google's guide to structured interviewing says structured interviews "are more predictive of job performance than unstructured interviews" and describes rubrics with poor, borderline, solid and outstanding answers; a 2022 meta-analysis ranked structured interviews as the top selection procedure for predicting performance (Sackett et al., Journal of Applied Psychology). An interviewer scoring against a rubric needs to find each part of your answer quickly.
The structure and time allocation (for a 2 to 3 minute answer).
| Part | Share | Content |
|---|---|---|
| Situation | about 10% | Context in two sentences |
| Task | about 10% | Your responsibility and the stakes |
| Action | about 50% | What you did and why: decisions, trade-offs, how you worked with others |
| Result | about 15% | Outcome, with numbers if possible, including what did not work |
| Learning | about 15% | What changed in how you work afterwards |
Why the learning matters. It is the part of the story that shows you improve. A learning that became a process (pre-registered success metrics, a launch checklist, a review ritual) is far stronger than a personal insight.
A story bank. With two or three behavioural questions per interviewer across a loop, you need more than three stories. Prepare six to eight that together cover:
- A failure or project that did not work.
- A disagreement with engineering, research or leadership.
- A hard prioritisation or a decision to stop something.
- Influencing without authority.
- Handling ambiguity or a fast-changing situation.
- A safety, ethics or quality stand under pressure.
- Explaining something technical to non-technical people.
- Leading a team through change.
Each story can be adapted to several questions.
What makes stories strong for AI roles. Stories that show:
- Treating model output as probabilistic: evaluations, error analysis, staged launches.
- Making cost, latency and quality trade-offs explicit.
- Managing expectations when capability was uncertain.
- Raising risk or safety concerns and acting on them.
- Working closely with research and engineering on technical decisions.
Read diagram description
An illustrative story bank of six stories mapped to the eight themes behavioural rounds probe: a failure, a disagreement, stopping work, influence without authority, ambiguity, a safety stand, explaining technology and leading change. A launch that missed its target covers failure, ambiguity and explaining; pushing back on model readiness covers disagreement and a safety stand; killing a feature after a pilot covers stopping work and ambiguity; getting teams to adopt evaluations covers influence and leading change; holding a release for a red-team finding covers safety and disagreement; briefing executives on fine-tuning covers explaining, influence and leading change. Every theme is covered at least once.
Delivery. Speak in the first person about your actions ("I decided", not "we decided" throughout), be specific, and stop when the story is complete; let the interviewer ask follow-ups.
What interviewers probe next
"What if I don't have AI-specific stories?" Use stories that show the same judgment in other contexts: decisions under uncertainty, experiments, quality trade-offs. Then explain how you would apply it to AI products.
"How long should an answer be?" Two to three minutes, then follow-ups. Long monologues lose the panel.
How it's scored
Weak: Rambling stories with long context and little about actions or learning.
Hire: Clear STAR structure with specific actions and results.
Strong Hire: STAR-L with process-level learning, a prepared story bank covering key themes, and stories that show AI-specific judgment.
Common mistakes
Spending most of the answer on background.
Saying "we" throughout so the panel cannot tell what you did.
Ending at the result without a learning.
See STAR-L stories.