Episode 2: The Sparring Partner Model for Using AI in Research

Most advice about AI in research lands in one of two camps. Either it's coming for researchers' jobs, or it's a toy you can't trust with anything real. Deanna Sim works somewhere more useful than either. She's a user researcher at Polarsteps, she uses AI most days, and she almost never ships what it gives her.

Her word for it is sparring partner. On the second episode of BlockSurvey's podcast, Research Secrets, she described a way of working with AI that treats it as something to think against, not something to think for you. It's the most practical framing we've heard, and it's worth stealing.

Using AI as a research sparring partner

A sparring partner doesn't fight your fights. They give you something to react to, expose the weak spots in your form, and push you to try a different angle, then step back so you do the real work. That's how Deanna uses AI.

She reaches for it when she's writing a screener and wants to test her thinking. Could she come at this from another angle? She uses it to find holes in her survey logic before real respondents find them for her. When a recruitment email isn't landing, she asks for other ways to phrase it. Every time, the AI produces something, and every time she treats that something as a prompt for her own judgment rather than an answer.

Hear Deanna explain it in her own words → watch the full conversation.

"It's rare that I would directly use something from AI. It's more of a sparring partner."

Where AI actually helps in user research

Deanna is specific about the tasks she'll hand over. The green light is for low-risk, context-light work: understanding survey responses, or pulling the top themes out of a pile of open-text answers as a starting point for analysis.

The payoff is that the quick, shallow passes get quicker, freeing up her time for the harder problems a person has to solve. Fast learnings on the small stuff buy room for the work that needs a researcher.

The limits of AI in research

This is the part most AI-in-research takes skip. Deanna doesn't trust the output. When AI summarizes survey responses, she double-checks the numbers, because it isn't always reliable. When she uses it to draft a recruitment email, she edits it heavily before it goes anywhere. She's blunt that taking what AI hands you and running with it is a mistake.

That caution is the point. It comes from experience with the tools rather than a gap in her skill with them. Knowing that a clean-looking summary can quietly miss the one thing that mattered is what the job is actually about.

What AI is good at in research, and what it isn't

The sparring-partner model matches AI to what AI is good at. It generates options fast: angles you hadn't considered, phrasings you wouldn't have reached, a rough first cut of themes. It's weak at the things that make research worth doing, like judging which insight matters, getting the numbers right, and knowing what a good insight even looks like.

Deanna draws that same line elsewhere in the conversation. Getting user insights is easy, she says. Getting quality ones is hard, and telling the two apart is harder still. AI sits on the easy side of that line. Treat it as a starting point and it speeds you up. Treat it as the finish line and it'll hand you something that looks like an insight but isn't.

How to use AI in your own research

If you want to work this way, the change is mostly about where you place AI in your process. Put it early, not late. Use it to react to, to poke holes, to generate a first draft you fully expect to rewrite. Keep it away from the version that ships.

Give it the small, reversible tasks and hold on to the consequential ones. A misfired brainstorm costs you nothing. A survey summary you trusted without checking can push a whole decision the wrong way.

And edit heavily. Nobody getting real value from AI found a magic prompt. They found a habit: treat every output as a rough draft with a mistake hidden in it somewhere, then go looking for it.

None of this means using AI less. Deanna uses it every day. The discipline is staying clear on which job you're handing it, the sparring, and which job stays yours.

About the guest

Deanna Sim is a user researcher at Polarsteps. She was the second guest on Research Secrets, BlockSurvey's podcast about uncovering real user insights to build better products.