Thanks for reaching out. Be sure to check us out on LinkedIn for all of our current news and announcements.
Oops! Something went wrong while submitting the form.
When ChatGPT first launched, we did what everyone did.
We told it to roleplay a customer. "You're a 38-year-old product manager at a mid-market SaaS company. I'm going to ask you some questions."
It was amazing. For about ten minutes.
Then the cracks showed. The answers sounded plausible but stayed shallow. The model was averaging out everything it knew about "product managers" and handing it back to us. No edge. No memory. No disagreement. No real person underneath.
So we waited.
The models got better. Other synthetic research platforms showed up, and we tried those too. Some were genuinely fast. Most weren't good enough to make a real decision on. A few were shockingly expensive. And almost none of them did the one thing we actually cared about: ground the personas in real evidence instead of inventing them from a prompt.
So we decided to build it ourselves. It's called Candor.
Building got easier. Building the right thing didn't.
Candor runs problem discovery, concept testing, price testing, and value-prop validation on synthetic users. The interview part isn't the interesting bit. Plenty of tools can stage a chat. What matters is what's underneath the persona.
A few things we built in on purpose:
1) Grounded in your evidence, not a prompt. Upload your research, transcripts, and reports. Candor pulls signals, retrieves corroborating web evidence, and builds the audience from the ground up. Every trait carries a provenance tag, so you can trace it back to a source you can audit.
2) Personas that push back. Real people don't agree with everything you say. Candor's personas hold positions, disagree with each other, and stay in character. No generic AI "yes, great point" drift.
3) Memory across sessions. Interview the same persona next week with a new concept. They remember the last conversation. You get longitudinal signal, not a reset every time.
4) A critic checks every answer. Before a response reaches you, a separate model checks it against the persona's prior statements and the underlying evidence. If it contradicts itself, it gets caught.
Synthetic research doesn't replace talking to real customers. It complements it.
The bottleneck was never the research method. It's how much research you can actually run. Your team can run one study at a time. Product, brand, pricing, and innovation all need ten. So the urgent studies jump the line, most never run, and people end up building on instinct or a ChatGPT prompt they're quietly calling "research."
A panel round costs tens of thousands of dollars and takes weeks. That means you only test the questions you can absolutely justify. Everything else ships untested.
Candor is for everything else. Evidence-grounded signal you can act on the same day you ask, so the expensive, slow research goes to the few decisions that truly warrant it.
We're in development now and opening up the waitlist. If you own more research demand than your team can absorb, we'd love for you to try it and tell me where it breaks.
Join the waitlist at runcandor.com.
