Why 'ask me interview questions' gives you a useless mock

I've sat on both sides of the table. I've run hiring loops for my own startup since 2021, and before that I spent a brutal three months back in March 2020 interviewing my way out of a job I'd outgrown. So I've watched the same mistake from both chairs, and the worst one happens before the interview even starts: people prep with a prompt like 'ask me some interview questions for a product manager role.' That's it. Then ChatGPT spits out 'What is your greatest weakness?' and 'Where do you see yourself in five years?' and they feel prepared. They are not prepared. They've rehearsed answers to questions no good interviewer asks anymore. The model didn't fail you. You gave it nothing to aim at, so it reached for the most generic template in its training data. It doesn't know the company, the level, the team, or what you're actually weak at. An interview prompt generator's job is to pull those out of you before the mock starts. What's the exact title? Is this a panel or a hiring-manager screen? What's the one question you're dreading? Feed those in and the same model becomes a sharp, specific interviewer instead of a quiz machine reading from a 2015 listicle.

The context that turns a chatbot into a real interviewer

Here's the minimum I put into an interview prep prompt. Skip any of these and the mock goes soft: - The exact role and level (Senior PM, not 'PM'; L5, not 'mid') - The company and, if you can find it, its interview style (Glassdoor reviews are gold here) - The interview type: recruiter screen, hiring-manager call, panel, take-home debrief - Your real background — the messy version, not the resume version - The two or three questions you're scared of - How hard you want it to push: gentle warm-up or stress test That's six inputs. The interview prompt generator asks for them, then builds a prompt that tells the model to stay in character, ask one question at a time, wait for your answer, and critique it before moving on. Do you need all six every time? No. But the company and the level are non-negotiable — a mock for 'an interview' is worthless, while a mock for 'a hiring-manager screen at a 50-person fintech for a senior role' actually rehearses the conversation you'll have.

STAR-method drilling is where most prep falls apart

Behavioral questions are won or lost on structure, and the structure everyone knows by name and almost nobody executes is STAR: Situation, Task, Action, Result. People can recite the acronym. Then they open their mouth in a real interview and ramble for four minutes about Situation, never get to the Result, and the interviewer quietly marks them down for 'doesn't quantify impact.' I did exactly this in a 2020 onsite and got dinged for it in the feedback, which I only saw because the recruiter was unusually kind. The interview prompt generator has a mode built for this. You give it a story — 'the time I shipped a feature that tanked engagement and had to roll it back' — and it generates a prompt that asks the model to interview you through the STAR frame one beat at a time. What was the situation, in one sentence? What was your specific task, not the team's? What did you personally do? What was the measurable result? It won't let you skip the Result, because skipping the Result is the whole problem. After a few reps your stories compress from a rambling four minutes to a tight ninety seconds, which is roughly the length a good interviewer actually wants. Is this a substitute for having real stories worth telling? No — the tool can't manufacture experience you don't have. But if you've done the work and just can't tell it well, this is the fastest fix I know.

Company research prompts: stop walking in blind

The candidates who stand out aren't the ones with the slickest answers. They're the ones who clearly did their homework on the company. I've ended interviews early — in a good way — because a candidate asked a question so specific about our roadmap that I knew they'd actually read our changelog. You can prompt your way to that. The generator builds research prompts that go past the 'tell me about the company' surface: what does this company likely value in this role based on the job description, what are the three hardest questions someone interviewing here might face, what recent news or product launch should I be able to speak to, and what smart questions could I ask that show I understand their actual problems? Pair that with real sources — Glassdoor for interview reports, Levels.fyi for comp benchmarks, the company's own blog and LinkedIn posts for what they're shipping — and you walk in sounding like someone who wants this job, not any job. One caveat I'll be honest about: the model's knowledge has a cutoff, so it can be stale or flat wrong on recent company news. Use it to structure your research, then verify the facts yourself before you quote them in the room.

ChatGPT vs Claude as a mock interviewer — they behave differently

I've run mock loops through both for a while now, and they don't play the interviewer the same way. ChatGPT is more willing to stay tough and in-character; it'll fire a cold follow-up like 'that's a team result, what did you do?' without softening it. Claude tends to be warmer and more coaching by default, which is comforting and sometimes too easy on you — though you can fix that with one line in the prompt telling it to be a skeptical interviewer who doesn't hand out reassurance. Which do I prefer? For raw stress-testing, I lean ChatGPT. For the debrief afterward — 'here's why that answer was weak and how to restructure it' — I think Claude's feedback is more useful and less generic. The interview prompt generator doesn't pick for you. You copy the prompt into whichever you want, and it shapes the instructions toward the behavior you asked for. Want a brutal panel? It writes the prompt for a brutal panel. Want a gentle first-round warm-up? Same tool, different dial.

A worked example: from 'PM interview at a fintech' to a real mock

Let me show the actual difference, because abstract advice is cheap. The lazy version: 'Act as an interviewer and ask me product manager questions.' The model asks five generic questions, you give five rehearsed answers, nobody learns anything, and you close the tab feeling falsely ready. The generated version starts by asking you for specifics, then produces a prompt like this: 'You are a hiring manager at a 200-person fintech interviewing for a Senior PM on the payments team. Conduct a 30-minute behavioral and product-sense interview. Ask one question at a time. After each answer, rate it 1-5 on structure and specificity, point out the single biggest weakness, then ask a sharp follow-up before moving on. Start with a behavioral question about handling a failed launch.' Drop that into ChatGPT and you get a mock that feels uncomfortably like the real thing — the follow-ups land, the ratings sting, and the weak spots surface in private instead of on the day that matters. The first time I ran a version of this on myself, the model caught that I never quantified outcomes in any of my answers. That single piece of feedback was worth more than the previous week of unstructured prep, and it cost me about four minutes to set up.

What an interview prompt generator can't do for you

I'd rather be straight with you than oversell this. A generated prompt makes the model a better interviewer and a sharper coach, but it can't do three things, and pretending otherwise would waste your time. First, it can't give you experience you don't have. If the question is 'tell me about a time you led a team' and you've never led a team, no prompt fixes that — at best it helps you find a smaller, honest version of leadership in your past. Second, it can't read the room on the actual day; a human interviewer reacts to your tone, your nerves, your eye contact, none of which a text mock rehearses. So do at least one real mock with a person if you can — Pramp pairs you with strangers for free, and a friend over video is better than nothing. Third, it can't promise the questions it generates will be the ones you get; it's pattern-matching on the role, not reading the interviewer's mind. What it does do well is volume and structure: it'll run you through twenty hard reps while a human friend gets tired after three, and it never sighs when you ask to redo an answer for the fifth time. Use it for that, verify the company facts yourself, and book the human mock for the parts a chatbot genuinely can't touch.