First, a clarification. I'm not looking for a job.

What I was looking for was a way to test an idea: Could ChatGPT conduct a realistic job interview, using an actual job description, and then provide a useful, reasonably objective assessment of the candidate's performance?

If it could, I thought, this might be a remarkably simple way for job seekers to prepare for interviews. No expensive coaching sessions, no elaborate software, no rehearsing canned answers to questions that might never be asked. Just a job description, an AI interviewer, and a willingness to be challenged.

So I decided to try it myself.

Since much of my professional background is in marketing, I selected a marketing position. Not just any position, either. I chose Product Marketing Manager at Amazon Web Services — a job calling for strategic thinking, strong communication skills, substantial financial services experience, and the ability to translate complicated technology into language business executives might actually understand.

That sounded reasonably familiar. I've spent much of my career doing exactly that sort of thing.

There was one problem. The job also required six years of hands-on experience with cloud architecture. I have approximately none.

Perfect. This would be an interesting test of whether my imaginary interviewer could distinguish between someone who interviews reasonably well and someone who is actually qualified for the job.

I copied the AWS job description into ChatGPT and asked it to conduct a realistic interview, question by question, for about ten minutes. I told it to listen to my answers, follow up where appropriate, and give me a candid evaluation afterward.

I didn't give it my résumé. I didn't prepare talking points. I simply answered the questions as I might in an actual interview.

What happened was surprisingly instructive.

The interviewer wasn't an idiot

The first question was predictable enough: Tell me about your background and why you're a good candidate for this position.

I explained my career in financial services marketing, my early enthusiasm for web technology, and my experience developing strategic solutions to business problems.

So far, so good.

Then the interviewer asked me to describe a specific situation in which I'd recognized a market opportunity others had overlooked.

I recalled an initiative from the early days of the internet, when asset management companies were shipping mountains of printed sales literature to financial advisers around the country. The printing costs were enormous. So were the shipping expenses. Salespeople wanted everything immediately, which generally meant overnight delivery.

I had proposed an online resource through which asset managers could make their product literature, research, fact sheets, and other materials available to the brokerage community. It would be a shared platform, supported by subscriptions from participating companies.

Management liked the idea. I assembled a team, brought in technical expertise, developed a proposal, and received authorization to proceed.

Unfortunately, the company was sold, and the project was shelved. Other people subsequently developed a business around essentially the same concept.

ChatGPT listened to the whole story. Then it moved on.

The next question placed me in a hypothetical meeting with the chief technology officer and chief financial officer of a financial services company considering a substantial investment in generative AI.

The CTO was enthusiastic. The CFO was worried about cost, security, regulatory compliance, and whether the investment would generate a measurable return.

How would I handle the situation?

I recommended starting with a narrow pilot project — something that could demonstrate value without exposing the company to substantial financial or regulatory risk. I talked about establishing guardrails, defining the intended product, measuring adoption, and expanding only after proving the concept.

A perfectly reasonable answer, I thought.

The interviewer wasn't finished.

Then came the question I couldn't answer

Suppose, the interviewer said, that the company's technical team recommended Amazon Bedrock for the AI application.

The CFO asks why she should use AWS rather than Microsoft Azure, which her company already uses, or simply purchase an established commercial application.

How would I make the case?

I couldn't.

I knew enough to understand the question, but not enough to answer it responsibly.

So I said so.

I explained that my knowledge of AWS and Microsoft's competing offerings was limited, that I could study the differences, but that I wasn't going to speculate about capabilities I didn't understand.

The interviewer thanked me for my candor. It also pointed out, quite correctly, that the position required substantial technical fluency from the outset.

There was no congratulatory speech about transferable skills. No suggestion that enthusiasm could somehow substitute for six years of cloud architecture experience.

It simply moved to the next question.

At that point, I stopped the interview and asked for an evaluation.

A 7.2 on interviewing. A 4.5 on qualifications.

ChatGPT rated my overall interview performance at 7.2 out of 10. It gave me strong marks for strategic thinking, executive communication, identifying market opportunities, and experience in financial services.

But it rated my fit for the AWS position at 4.5.

It also identified some genuine weaknesses in my answers.

I had spent too much time explaining one project. I hadn't quantified its financial implications. In another answer, I'd talked about how I would approach a business challenge rather than describing a comparable challenge I'd actually handled.

All fair criticisms.

And it concluded that, based on the evidence presented, I probably wouldn't advance to the next round for this particular position.

I agreed.

Now, those numerical scores aren't scientific measurements. ChatGPT is not an authorized Amazon interviewer, and its ratings are subjective. Another practice session might produce somewhat different numbers.

But the distinction between interviewing ability and qualifications was useful. You can give a thoughtful, articulate interview and still be the wrong candidate for the job.

You can also be an excellent candidate and give a lousy interview.

Both problems are worth identifying before you walk into the real thing.

Try it yourself. It takes about ten minutes.

You can conduct exactly this kind of interview yourself. You don't need to purchase an interview preparation package or install specialized software. A regular ChatGPT conversation will do, subject to the usage limits of your account.

Find a job you're seriously considering. Copy the entire job description, including the qualifications, and paste it into ChatGPT along with the following prompt:


I'm going to paste a job description below. Please act as the hiring manager interviewing me for this specific position.

Conduct a realistic interview, asking one question at a time. Listen to my answers and ask follow-up questions based on what I say. Challenge me when my answers are vague, unsupported, or don't address the question.

Make the interview approximately ten minutes long, or stop whenever I ask.

Don't coach me during the interview.

At the end, evaluate my performance candidly. Rate my communication, relevant experience, quality of examples, technical knowledge, and overall fit for the job. Distinguish between how well I interviewed and how qualified I appear to be.

Tell me what I did well, what I should improve, and whether you would advance me to the next round. Don't flatter me. Be fair.

Here's the job description:

[PASTE JOB DESCRIPTION]

Then begin.


You can type your answers, but I recommend speaking them using ChatGPT's voice feature if it's available to you. An interview is a conversation, not a writing exercise. You need to hear yourself explaining things — sometimes awkwardly, sometimes at excessive length, occasionally wandering into territory you never intended to visit.

And you need practice recovering when that happens.

Don't stop after a difficult question to ask ChatGPT what you should have said. Save that for the evaluation. Otherwise, you're not practicing an interview. You're practicing an interview with a teleprompter.

When you're finished, ask what you should have done differently. Ask which answers were unconvincing. Ask where you failed to provide evidence. Then try another round.

You might even give it your résumé — after removing personal information you don't want to share — so it can examine how well your claimed experience aligns with the position.

One caution: AI can misunderstand an answer, overlook relevant experience, or place unreasonable emphasis on a particular qualification. It also doesn't know everything about the employer's actual hiring standards. Treat its judgments as informed practice feedback, not a verdict on your career.

The uncomfortable questions are the useful ones

What impressed me about this little experiment wasn't the technology's ability to produce interview questions. That's easy enough.

It was the way the questions followed my answers.

The interviewer began with my background, moved into evidence of strategic thinking, tested my approach to a business problem, and finally challenged me on the very technical knowledge I lacked.

It didn't know that I would stumble over AWS versus Azure. But the job description told it that technical expertise mattered, and it pursued the issue.

That's what a competent human interviewer is supposed to do.

Many job seekers prepare for interviews by memorizing answers to familiar questions. Tell me about yourself. What's your greatest weakness? Where do you see yourself in five years?

There's nothing wrong with preparing for those questions. But an experienced interviewer doesn't necessarily follow a script. A good answer may lead to a harder question. A questionable claim may invite scrutiny. A promising example may prompt a request for numbers.

And eventually somebody may ask you something you simply don't know.

Better to encounter that moment in your living room than across the table from a hiring manager.

There's another benefit. You may discover that a job you thought was a perfect fit isn't. Or that you have the necessary qualifications but are doing a poor job of explaining them.

Either discovery can improve your job search.

My imaginary AWS interview didn't end with an imaginary job offer. It ended with a reasonably accurate explanation of why I wasn't the person AWS was looking for.

Not exactly a triumph.

But it cost me nothing, took only a few minutes, and taught me something about both the position and my interview technique.

For a job seeker, that's a pretty good return on investment.

And unlike the real interview, you can do it again tomorrow.