A Reminder of What UX Research Actually Is, Now That It’s Easy to Fake

I keep running into some version of the same pitch over and over and over. Skip the recruiting, skip the scheduling, skip the eight-week wait for a readout. Feed a model some personas, ask it what “users” think, and you’ve got a synthetic panel. It’s fast. It’s genuinely fast I am not discounting that. It also produces something that looks exactly like research without any of the actual research having happened or really validated, and I don’t think enough people are stopping to say that part out loud.

A model can hand you a clean set of “user quotes,” a theme breakdown, a recommendation that sounds confident, in about ten minutes. It has the shape of a finding. Quotes, themes, a takeaway, formatted the way research usually gets formatted. Nothing about it signals that no actual human being said any of it, because nothing has to. That’s what’s new here. Faking a study used to take real effort. Now the fake version is often quicker to produce than the honest one, and on a slide, they can be hard to tell apart.

So it’s worth going back to basics on what this work was even doing in the first place. Underneath the guides and the recruiting screens and the affinity mapping, research has always come down to one thing: put a real person in front of something and see what they say when you haven’t scripted the answer. Everything else exists to protect that one moment from bias, from leading questions, from your own assumptions.

A synthetic panel gives you a fast, plausible average opinion, and I won’t pretend that’s worthless by any means. It’s a decent gut check and if used correctly can help to unlock faster heuristics. It can catch an obviously confusing question before it goes out the door.

What it can’t do is surprise you, and the surprise was always the point. A real person tells you about the workaround they built because your product didn’t do what they needed. They get quietly annoyed by a screen you assumed was fine. They use your tool in some context you never designed for and never would have guessed. A model trained on the average of everything already written about a topic gives you, unsurprisingly, the average. It flattens exactly the edge cases good research is supposed to catch.

None of this shows up anymore when it goes wrong, either. Nobody sends an email saying “we’ve quietly started treating model guesses as real findings.” It just happens, gradually, because the synthetic version arrived faster and looked fine on the readout. You don’t find out until the thing ships and the actual reaction doesn’t match what got forecasted, and by then it’s a much more expensive problem to trace back to its source.

I’m not making a case against using AI in this work, to be clear. The parts that speed up synthesis, draft a rough first pass at a guide, or cluster themes across a messy transcript pile are genuinely useful, and I use them. A researcher who’s good with these tools will move faster than one who isn’t, full stop.

What I’m getting at is narrower. Know which part of the job just got easy to fake, and hold that part to a higher bar instead of a lower one. If a stakeholder can get something plausible-sounding out of a model in ten minutes, then “we talked to some users” stops being a high enough bar to call something real research. The new bar is closer to: did we find out something a model could not have guessed on its own. That’s checkable. It’s also a lot higher than most research decks have historically had to clear.

The actual skill this rewards is knowing where synthetic data is fine and where it’s a problem. Fine for pressure-testing wording on a survey nobody’s sent yet. Not fine for figuring out whether people will trust a new AI feature with their data, because trust is exactly the kind of thing people describe one way and feel another, and a model has no mechanism for catching that gap. It just doesn’t have access to it.

Real research was never really defined by having quotes and themes in a slide deck. It was defined by contact with an actual person who could tell you something you hadn’t already assumed. That’s always been true. It just didn’t need saying out loud until it became this cheap to fake the look of it without any of the contact underneath.

A Reminder of What UX Research Actually Is, Now That It’s Easy to Fake was originally published in Bootcamp on Medium, where people are continuing the conversation by highlighting and responding to this story.