Please stop interviewing robots

November 2, 2026 · Presitt ClicksAgenciesInterviewsForwards

A friend who runs research at an agency got pitched synthetic respondents three times last month. Different vendors, same deck. Slide two is always a stock photo of a smiling focus group. Slide three is the same photo with the people blurred out and a little glowing brain where their heads used to be.

The pitch: why pay for recruiting, incentives, no-shows, and a moderator who keeps saying “interesting, tell me more” when you could just ask a model what a 41-year-old HR director in Ohio thinks about benefits software?

I build Presitt. We make software that helps research get found, sent, and actually read, so I spend a lot of time with people whose whole job is listening to customers. So I’ll say it plainly: if you’re an agency and you replace your interviews with robots, you’re building your own replacement and invoicing the client for materials.

That said, I’m not here to tell you to go back to transcribing tapes with a quill. There’s a cheaper fix. It’s boring. Stick with me.

We were already cutting corners

Synthetic users didn’t invent research shortcuts. They’re just the newest one on a list every agency researcher can recite in their sleep:

  • The proposal said 24 interviews. The budget said 16. Recruiting said 11, and one of them was on mute the whole time.
  • Analysis was “thorough,” meaning someone read the transcripts at 1.5x on a train.
  • The screener asked for VPs of Finance. You got a VP of Finance, an “aspiring” VP of Finance, and a guy named Doug whose relationship to finance remains unclear.
  • You ran a survey instead of interviews because a survey makes a chart, and clients will forgive a lot for a chart.

Nobody’s proud of this list. It’s what happens when rigor meets a fixed-fee SOW.

But every item on it still involves an actual human saying an actual thing. Fewer of them, messier, fine. Synthetic respondents go one step further. They remove the part where anyone said anything.

Robots are fantastic interviewees. That’s the problem.

Credit where it’s due: synthetic users are the best participants you’ll ever have. Always on time. Camera on. Never says “sorry, can you repeat the question.” Never wanders off into a ten-minute story about their kid’s lacrosse schedule.

Which is exactly why they’re useless for the stuff that matters.

The lacrosse story is where the good stuff lives. That’s when the HR director mentions, offhand, that she runs benefits enrollment out of a Google Sheet because the expensive software takes nine clicks to do one thing. That’s your finding. That’s the slide the client screenshots and sends to their CEO. No model gives you that, because no model has ever been annoyed by nine clicks.

A synthetic respondent gives you the average of everything anyone has typed about your topic, delivered with total confidence. Ask it a leading question and it’ll follow you anywhere, like a golden retriever with a PhD. And when the client’s head of product asks in the readout “who actually said this?”, you get to answer “nobody, technically.” Fun way to end a retainer.

Are there legit uses? Sure. Warming up a discussion guide. Spitballing hypotheses before you go talk to humans. Checking whether your survey wording makes sense. Use it like spellcheck. Do not use it like a customer.

The bigger risk for agencies isn’t one bad finding. It’s that your client figures out they can rent the same robot for $40 a month. You just went from “the people who talk to our customers” to “the people who type into the thing we also have.” Tough spot to negotiate rates from.

You already did the interviews. You lost them.

What actually bugs me: most agencies I talk to aren’t short on customer conversations. They’re drowning in them. They just can’t find any.

Open your shared drive. Search “interview.” Watch 4,000 results load, mostly named “Deck,” “Deck (1)” and “Deck FINAL use this one.” Somewhere in there are 30 hours of recordings with exactly the buyer your new client wants to understand. You ran that project two years ago. It was good. You remember it was good. You do not remember where it is.

Also, the person who ran it now works brand-side and occasionally likes your LinkedIn posts.

So the new brief lands, the deadline is “yesterday, ideally,” and starting from scratch feels impossible. Somebody suggests the robots. And it almost makes sense, because compared to excavating your own drive, typing a prompt sounds relaxing.

That’s why synthetic research is selling. Not because it’s good. Because most agencies have the memory of a goldfish with a Dropbox account.

A research lead I’ll call Kate found this out the expensive way. Her team spent three weeks and a recruiting budget interviewing procurement people about switching vendors. Halfway through the readout, the client’s head of ops said, “Didn’t you already do this for us?” They had, two years earlier. The transcripts were in a folder named FINAL_v7_USE THIS. The researcher who ran that study had left for a brand job and taken the only working map of the drive with her. Kate’s shop paid to learn the same thing twice, then invoiced the client for the second copy.

Agencies get hit twice here. The knowledge leaves when the project ships to the client. Then it leaves again when the researcher leaves you.

The cheap fix nobody puts in a proposal

The best shortcut in research is still talking to people. The second best is not forgetting what they said. That’s the whole idea. Sorry it’s not a paradigm shift.

What it looks like when an agency actually does it:

  • Keep the raw stuff. The client gets the deck. You keep the transcripts, recordings and notes, wherever your contract allows (read your contract, I’m not your lawyer). The deck is the summary. The transcripts are the asset.
  • Organize by question, not by project code. Nobody has ever said “pull up ACME-Q3-PH2.” People say “didn’t we talk to procurement folks about switching vendors once?” Your archive should be able to answer that, or at least not laugh at you.
  • Spend 30 minutes looking back before you scope. Search what you already know before writing the proposal. Worst case you lose half an hour. Best case you walk into kickoff sounding like you’ve been on the account for a year.
  • Run fewer, sharper interviews. If you already know people hate onboarding, don’t spend 15 interviews rediscovering it. Spend six on what you can’t answer yet.

If a client wants faster and cheaper, this is the honest version. Not fake people. The real ones you already paid to talk to, plus a handful of new ones. You can put names on the quotes. You can defend it in the room. Doug might even be in there.

Then nobody read your deck

One more problem, and it’s the one that actually loses accounts.

Say you did everything right. Real interviews, real analysis, a deck so clean your designer teared up a little. You email it to your client contact. They reply “amazing, thanks!” and then you hear nothing for four months.

Did the VP read it? Did it get to product? Did anyone make it past the exec summary? No idea. Your contact doesn’t know either. The deck went into an inbox and then, quite possibly, into a SharePoint folder where research goes to die.

Then budget season. Someone in finance asks your contact what the agency actually delivered this year. Your contact says “some really good research,” which is the corporate version of “he has a great personality.” The line item gets cut, handed to a cheaper shop, or swapped for the robot subscription.

This is the thing I’ve been obsessing over at Presitt, because I’ve watched it take out good research teams again and again. The work was fine. Nobody could prove it got used.

So send the work in a way that tells you something back. Who opened it. Which pages they actually read. Which teams it reached, including people who never had a login. Then give that to your client contact so they can forward it to whoever signs the checks. “Your study got read across product, marketing and the exec team, and they lingered on pricing” is a renewal. “Hope you enjoyed it!” is a prayer.

So, about the robots

Synthetic respondents will get better. They’ll be useful for more things. I’m not going to pretend they’re going away.

But for an agency, the people are the product. The minute you swap them for a model, you’re reselling something your client can buy directly, with a markup and a nicer font.

Talk to real people. Keep what they told you somewhere you can find it. Make sure someone actually reads the thing. None of this is new. It’s just the stuff that got skipped when the deadline moved up.

If you want to see what proof of readership looks like on your next client deliverable, that’s Presitt. $29 a month. No login for the people who read it. If you’d rather go interview 500 robots, go for it. Save the transcripts. You’ll want them for the post-mortem.

Questions people ask

Should agencies replace customer interviews with synthetic respondents?

No. A model can warm up a discussion guide, spitball hypotheses, or check survey wording. It cannot get annoyed by nine clicks in real software, and it cannot put a name on a quote when a client asks who said it.

What is the cheaper alternative to synthetic research?

Keep the transcripts, recordings, and notes your contract allows. Organize them by the question a colleague would actually ask, then search what you already know before you scope the next proposal. Add a handful of new interviews for what you still cannot answer.

Why do good research decks still lose the account?

The deck lands in an inbox, then a SharePoint folder. At budget season nobody can prove product, marketing, or the exec team read it. Proof of who opened it, which pages they read, and which teams it reached is what a client contact can take to finance.

How much does Presitt cost?

Presitt is $29 a month. One seat for the person who sends. Readers do not need a login. It shows who opened a deliverable, which pages they read, and who forwarded it.

← All posts