ChatGPT Interview Prep: Prompts and Patterns

The three prep modes that work

Question generation

Prompt template: "Act as a senior engineering interviewer at [target company] hiring for a [role title]. Ask me one STAR-format behavioural question on [competency], at a difficulty level appropriate for a [seniority] hire. Do not provide the answer; wait for my response."

Iterate this 5-10 rounds per competency. After each answer, ask: "Critique my answer. What was strong? What was generic? What specific detail would make it more credible? Then ask one adversarial follow-up question." The model is much better at this when you give it a target company and seniority anchor than when you ask abstractly.

Adversarial rehearsal

Prompt template: "Act as a sceptical interviewer who has already heard a hundred answers to this question. Push back on my answer wherever it sounds generic. Probe for specifics. Surface assumptions I am making. Do not let me off easy."

Most ChatGPT critiques are too gentle by default. The explicit "do not let me off easy" framing pulls the model toward useful adversarial behaviour. Run three to five adversarial rounds per behavioural answer; the answer should land harder on the third round than it did on the first.

Concept review

Prompt template: "Explain [concept] as if to a senior engineer who will test me on it tomorrow. Include the three trade-offs they are most likely to probe, the common wrong answers, and one surprising thing about it."

The "as if to a senior engineer who will test me" framing pulls the model toward depth and trade-off articulation rather than textbook recitation. Verify every technical claim against a primary source before relying on it — see below.

The verification step

Never accept a technical claim ChatGPT makes without verifying. Cross-check against the official documentation, a recent blog post by a recognised expert, or a textbook chapter. The model will state plausibly-wrong things with full confidence; the verification habit catches them before they end up in your interview answer.

Where ChatGPT misleads — and how to catch it

Plausible-but-generic answers

Ask ChatGPT for a STAR answer about "a time you led a difficult technical decision" and it will produce something that sounds reasonable and could apply to any senior engineer at any company. Generic answers score poorly in interviews because they signal nothing specific about you. Use the model to draft structure; supply the specific project details yourself.

Confidently confirming wrong technical claims

Write a deliberately wrong technical assertion in your question ("I think Redis uses MVCC for write concurrency, right?"). ChatGPT will often agree and elaborate. The verification habit catches this; ask the model to argue both sides or to find counter-evidence before you trust a technical answer.

Inventing project experience you do not have

When you say "give me a STAR answer about migrating a monolith to microservices," ChatGPT will invent a plausible migration story complete with specific numbers. Those numbers are not yours; using them in an interview is fraud. The model is a structure generator, not a memory replacement.

When to switch tools

Switch to Claude when you want more nuanced behavioural critique — it tends to be sharper on subtext and less sycophantic by default. Switch to a primary source (documentation, textbook, recent expert blog) the moment a technical claim matters. Switch to a human (mock interviewer, mentor, knowledgeable peer) when you need calibration on whether your answer would actually fly in front of a senior interviewer.

A 1-hour prep session template

Minutes 0-20 — Question generation

Pick one named weak competency from your prep journal. Ask ChatGPT to generate five varied STAR-format prompts on it, calibrated to your target seniority and company. Answer each in writing — not aloud, written. The writing forces specificity; the speaking lets you wave hands.

Minutes 20-45 — Adversarial rehearsal

Take your two best written answers from the first segment. Switch ChatGPT into adversarial mode (the prompt template above). Run each through three rounds of follow-up. The goal is not to "win" each round; the goal is to find the question you cannot answer well yet, then revise the underlying answer so you can.

Minutes 45-55 — Concept review

Pick one technical concept the role description emphasises. Use the concept-review prompt. Verify the three trade-offs the model mentions against a primary source. Add the verified version to your prep notes.

Minutes 55-60 — Self-critique

Write three lines in the prep journal: what landed, what did not, what is the highest-priority weak area for the next session. Then close the laptop. The session is over; resist the urge to grind into hour two — diminishing returns set in fast.

Related guides