I used to be terrible at interviews

Not because I was unqualified. Because I'd walk in having done about 15 minutes of googling the company's "About" page, with zero prepared answers, and try to wing it. The technical bits I could handle. The "tell me about a time when..." questions? I'd freeze, ramble, lose my thread, and then spend the tube ride home composing the perfect answer I should have given.

Then I started using AI to prep, and the difference was immediate. Not because the AI is smarter than me. Because it forced me to actually prepare, think through my answers, and practise saying them out loud. The AI is just a very patient, very available practice partner that doesn't judge you for asking the same question six times.

Here are the exact prompts I used for my last three interview processes. Two resulted in offers. The third I turned down before the final round because the AI-assisted company research uncovered some red flags. So really, three for three.

Step 1: Company research with Perplexity

perplexity

Before anything else, you need to understand the company beyond their careers page. Perplexity is perfect for this because it searches the current web and cites sources, so you're getting actual recent information rather than a generic company profile.

The prompts I use:

"What are [company name]'s biggest business challenges in 2026? Include any recent news, leadership changes, market pressures, or strategic shifts."

"What do current and former employees say about working at [company name]? Look at Glassdoor, Blind, and recent LinkedIn posts."

"Who are [company name]'s main competitors, and how is [company name] positioned differently?"

"Has [company name] had any layoffs, restructuring, or significant changes in the past 12 months?"

Why this matters for interviews: When the interviewer asks "Why do you want to work here?" you can reference specific things. "I read about your expansion into the European market and the challenges around localisation. That's exactly the kind of problem I've spent my career solving." That answer beats "I'm really passionate about your mission" every single time.

The red flag I mentioned: During research for my third interview process, Perplexity surfaced several Glassdoor reviews mentioning a toxic engineering culture and a Blind post from a current employee describing 70-hour weeks as "normal." I brought this up diplomatically in the interview ("I've heard the team works quite intensively, can you tell me about work-life balance?") and the interviewer's uncomfortable response told me everything I needed to know.

Step 2: Generate likely interview questions

claude

This is where the prep gets specific. Paste the entire job description into Claude and ask it to generate likely questions.

The prompt:

"You're an experienced hiring manager. I'm interviewing for this role: [paste the full job description]. Based on this job description, generate 20 likely interview questions across these categories: (1) behavioural questions about teamwork and conflict, (2) technical questions about the specific skills mentioned, (3) situational questions about handling challenges, (4) culture fit questions, (5) questions I should ask the interviewer. Be specific to this role, not generic."

Why Claude and not ChatGPT: Claude tends to generate more specific, nuanced questions. ChatGPT often defaults to generic interview questions that could apply to any role. Claude reads the job description more carefully and picks up on specific requirements.

The output from this prompt is usually eerily accurate. In my last three interviews, roughly 60-70% of the questions I was actually asked were close matches to what Claude predicted. You won't get exact matches, but you'll walk in having thought about the right topics.

Step 3: Practise STAR answers

The STAR method (Situation, Task, Action, Result) is the standard format for behavioural interview answers, and most people are terrible at it because they've never actually practised structuring their experiences this way.

The prompt:

"I'm preparing for a behavioural interview question about [topic, e.g., 'resolving conflict with a colleague']. Here's the situation I want to talk about: [describe your experience in 2-3 sentences]. Help me structure this as a compelling STAR answer for a [role title] interview. The answer should take about 90 seconds to deliver out loud. Make it specific, include a concrete result with numbers if possible, and end with what I learned."

The follow-up prompt that makes it better:

"That's good but it sounds too polished. Make it more conversational, like I'm telling a story to a friend. Add a moment of honesty about what was difficult or what I got wrong initially."

This second prompt is crucial. STAR answers that sound rehearsed are almost as bad as no preparation. The AI can help you find the balance between structured and natural.

My tip: Prepare STAR stories for these five situations and you'll cover about 80% of behavioural questions: A time you disagreed with a colleague or manager A time you failed or made a significant mistake A time you had to learn something quickly A time you led a project or initiative A time you dealt with an unhappy customer/stakeholder

Step 4: Salary research

Don't skip this. Going into a negotiation without data is like playing poker without looking at your cards.

The prompt (use Perplexity for this one):

"What is the current market rate for a [job title] in [city] with [X] years of experience? Include data from Glassdoor, Levels.fyi, LinkedIn Salary, and Payscale. Separate base salary from total compensation. Note any differences between company sizes."

Then follow up with Claude:

"I'm interviewing for [role] at [company] in [city]. The market rate for this role is [range from Perplexity]. The company has [size/funding/revenue context]. I'm currently earning [current salary]. What salary range should I target, and how should I frame my expectation if asked about salary early in the process?"

Claude is genuinely good at helping you think through negotiation framing. It'll suggest specific phrases like "Based on my research and the scope of this role, I'm targeting the [X-Y] range" which sound much more confident than "I'm flexible" (which means "please lowball me").

Step 5: The night-before mock interview

This is the one that makes the biggest difference, and it's the one most people feel too silly to try.

The prompt:

"Act as a tough but fair interviewer for the role of [role title] at [company name]. Here's the job description: [paste JD]. Ask me one question at a time. Wait for my answer. Then give me honest, constructive feedback on: (1) whether I actually answered the question, (2) the structure of my answer, (3) what was compelling and what was weak, (4) a suggestion for improvement. Then ask the next question. Start with a warm-up question, then get progressively harder. Do about 8 questions total."

Important: Answer out loud. I know it feels ridiculous talking to your laptop. Do it anyway. The difference between thinking through an answer and saying it out loud is enormous. You'll discover that the perfect answer in your head comes out as a rambling mess when you actually have to speak it. Better to discover this at 10pm the night before than at 10am in the interview.

chatgpt with voice mode is actually excellent for this, since you can have a genuine back-and-forth conversation rather than typing your answers.

The feedback is surprisingly useful. The AI will point out when you're being vague ("you mentioned 'significant improvement' but didn't quantify it"), when you're not answering the actual question asked, and when your answer runs too long. These are the exact things a real interviewer notices but never tells you.

Bonus: Post-interview follow-up

After the interview, while your memory is fresh, dump your recollections into Claude.

The prompt:

"I just finished an interview for [role] at [company]. Here are the questions I was asked and a rough summary of my answers: [list them]. Help me: (1) identify any answers I should clarify or expand on in a thank-you email, (2) draft a concise follow-up email that references specific conversation points, (3) flag any questions where my answer was weak so I can prepare better if there's a second round."

The honest caveat

AI prep won't compensate for a lack of genuine skills or experience. If you can't do the job, no amount of polished interview answers will hide that (and even if it did, you'd be miserable once you started).

What AI prep does is close the gap between your actual abilities and your ability to communicate them under pressure. Most interview failures aren't competence failures. They're communication failures. The person could do the job brilliantly but couldn't articulate that in a 45-minute conversation with a stranger.

That's the gap AI closes. It gives you the structure, the practice, and the confidence to show up as the best version of yourself. The rest is still up to you.

And yes, if the interviewer asks "How did you prepare for this interview?" you're allowed to say you used AI. In 2026, that's not a weakness. That's resourcefulness.