Episode 1: Are Synthetic Users Worth It? AI in Customer Research

You can spin up a synthetic user in seconds. Feed an AI what you already know about your customers, and it will roleplay one of them, answering your survey or interview questions without ever booking a slot on your calendar. No recruitment, no scheduling, and the cost rounds to almost nothing. For a lot of teams, that sounds like the end of expensive research.

Ariadna Roman isn't sold on that story.

She's a senior researcher and service designer in Barcelona who has spent seven years running interviews, focus groups, and workshops across banking, healthcare, mobility, and tech. On the first episode of BlockSurvey's podcast, Research Secrets, she gave us a straight answer about where synthetic users help and where they'll lead you astray.

What a synthetic user actually is

Her definition is plain: a synthetic user is a person who doesn't exist, built from the data you already collected about real ones. Say you interviewed a group of Gen Z customers last quarter, and now you want to test whether a new feature lands with that same audience. You have two options: recruit and talk to real Gen Z users again, or feed everything you learned last time into an AI and put your questions to the synthetic version instead.

How close the answers come to reality depends on how much you feed the model. Give it enough context and the responses start to resemble what real people might have said. That resemblance is also where the trouble starts.

Cheaper, faster, and lower-risk to try

The appeal isn't a mystery. Synthetic users cost far less than recruiting, scheduling, and analyzing interviews with real humans. They're faster, too. There's no waiting on recruitment, no chasing appointments, no sitting through the interview before you can start the analysis. You ask, you get answers, you move on.

Ariadna sees a third reason, and it's the more interesting one. Companies new to research often don't trust the process yet, because they've never run it. Synthetic users let them dip in without betting much. If it goes wrong, the loss is small. That low-stakes on-ramp gets hesitant teams to try listening to their customers at all, which she counts as a win.

Real people contradict themselves; synthetic users don't

The catch is that people are inconsistent, and not because we lie. We say one thing matters to us and then act against it. We hand over information nobody asked for. A synthetic user, built from tidy past data, smooths all of that away. As Ariadna put it:

"Sometimes we say we do things we don't do. And not because we are lying, it's how we work."

Real research surfaces that gap. You watch someone call a feature critical, then describe a workaround that shows they'd never touch it. A synthetic user won't contradict itself that way, because it only knows what you already knew. You get a clean answer and miss the messier truth sitting under it.

She unpacks this in the episode. Watch the full conversation.

Use them, but never let them make the call

So is she against synthetic users? No.

"I'm not against it. I think it's a very good idea, but I don't think it's a valid idea for everything."

Her rule is to combine. Use synthetic users where speed and cost matter and the stakes are low, but never let them make the call alone. She wouldn't launch a product on synthetic feedback by itself. She'd back it with real data and real people before deciding anything that counts.

Do the real thing first

For anyone starting out, Ariadna's advice runs against the efficiency pitch: talk to real users first. She calls herself a little old school about it, and her reasoning holds up. The skill of running a live interview, thinking on your feet, and noticing what a person actually means is what teaches you to ask a synthetic user good questions later. Skip that apprenticeship and you won't recognize what's missing when the AI hands you something plausible but hollow.

She does expect part of this to shift. She thinks synthetic users will grow more accurate, insights will arrive faster, and real-time research will happen less often than it does today. But she has doubts about whether the data will ever be as trustworthy as a real conversation, and her reason is specific: the unplanned moment.

In a live interview, you mention something she never asked about. That triggers a question she hadn't planned, and that question uncovers an insight neither of you saw coming. A synthetic user, answering only what you thought to ask, rarely gets you there.

So, are they worth it?

Sometimes. Synthetic users are a real tool with a real place: fast checks, early exploration, and teams testing whether research is worth their time at all. They're a poor stand-in for the moment a real person tells you something you didn't think to ask. The teams getting the most out of AI in research tend to be the ones who know which of those two situations they're in.

About the guest

Ariadna Roman Porro is a senior researcher and service designer based in Barcelona, Spain. You can connect with her on LinkedIn.