Candor helps you decide what to build, ship, and price, using synthetic users grounded in real evidence. Use it for problem discovery, problem validation, concept testing (including name, value proposition and messaging tests) and price testing on anything from a brand-new idea to a refinement of a live product. The fastest way to use it well is to learn the five things you’ll touch.
A project is the unit of work. One project = one audience you want to understand + one set of synthetic participants built for it + the interviews and reports that come from it. Naming the project and describing the audience happen on the same screen: you name it, describe who you want to speak with, pick a region, and upload any evidence you already have. Everything else flows from there.
Once your project is set up, Candor researches the audience for you. It reads your uploaded documents, proposes a set of candidate segments, then searches the web for evidence about them, extracting signals as it goes (behaviors, pain points, beliefs, constraints). A critic pass flags weak evidence at the end. You review the segments and decide which ones to include. That selection becomes the foundation for the participants.
Candor generates groups (the behavioral shapes that exist in your audience) and then materialises individual participants inside each group. Each participant has a personality (sampled from peer-reviewed Big Five distributions), cognitive biases with calibrated intensities, memories, and decision-making patterns. Each field in their profile and memory is labelled Grounded, Derived or Assumed (the personality and bias scores and the demographics summary carry no label), so you can see what is behind it: something you can check, something that follows from the evidence, or a gap Candor filled and says so.
Two ways to interview. Live: you type questions, participants respond in real time, and a critic validates each response for consistency before delivery. Auto: Candor runs the interview for you against a guide you configure, with built-in stopping criteria, coverage tracking, and signal extraction. Participants remember everything across sessions within the same project. What gets asked comes from the study type you chose: problem discovery, problem validation, concept testing or price testing. Concept testing itself comes in four kinds, a standard test on whole concepts plus name, value proposition and messaging tests that compare options for one concept you already have. See the use cases for what each one asks and returns.
An auto-interview run synthesises itself into a report when the last session finishes. It opens with What Matters Most, which holds the headline, the learnings and the single highest-priority move, then works down through the findings, the next steps, and an evidence appendix. In problem validation, concept testing and price testing each learning is paired with the step it implies. Problem discovery deliberately leaves the per-learning step off, because discovery is for understanding; its top move and its next steps are still there. What sits inside the findings is tuned to the kind of interview you ran, so a problem-validation report gives a verdict per hypothesis while a price-testing report gives willingness-to-pay ranges. Today’s reports paraphrase the evidence rather than quoting it, the full interview transcripts sit in the evidence appendix, and you can export the whole thing.
Candor Bot sits in the bottom-right corner of every screen in the app. It explains what you’re looking at, interprets a result, and, on the screens where you’re typing, will tell you whether what you’ve written is specific enough before you commit to a run. That last one is worth the habit: the quality of an audience description drives the quality of everything downstream.
Candor isn’t a faster way to invent customers. It’s a faster way to test whether what you think you know is supported by evidence. The participants push back. The critic catches drift. The synthesis surfaces tensions you’d miss with a single interview. Use it before you commit, not after.
No. Candor isn't an AI persona generator. It's a research platform that builds synthetic participants from real evidence (your uploaded research plus targeted web search) and lets you interview them. If you just need a persona slide for a deck, ChatGPT does that for free. If you need to find out whether the audience actually exists, what they care about, and whether your idea lands with them before you commit, that's what Candor is for. The output is decision-grade signal, not a stock photo with a name and a job title.
Most studies run in 1 to 2 hours of elapsed time, with maybe half an hour of your actual attention spread across that window. Your audience and its participants generate together in one run of 25 to 35 minutes, and it is queued, so you can close the tab. An interview guide takes a couple of minutes. Auto-interviews run in parallel and take 25 to 35 minutes. Synthesis adds 3 to 5 minutes once the last interview finishes. The bottleneck is your time spent reading and reviewing, not Candor's processing time.
No, but you'll get sharper synthetic participants if you do. Candor can build an audience from a description alone, using web search to find evidence. If you've already done research (PDFs, transcripts, market reports, survey data), uploading it makes the audience tighter, since the facts and quotes your documents state get weighted higher than web evidence (a document author's own interpretation of their evidence is kept as a hypothesis, not a finding). Even a few well-chosen documents move the audience from generic to specific. Start with what you have and add more later if the audience doesn't feel like yours.
Yes. A project is one audience plus its synthetic participants plus everything that's been interviewed inside it. You can run problem discovery first, then concept testing on the same participants a week later. They remember the prior conversation, so the second study builds on what they already told you. That's actually the workflow most teams converge on: open-ended discovery to find what matters, then targeted concept or price testing to pressure-test ideas against it. Concept testing comes in four kinds, so once an idea lands you can run a name, value proposition or messaging test on options for it.
ChatGPT will happily roleplay any persona you describe, but it has no evidence behind the character. That character agrees with whatever you push on. Candor's synthetic participants are built from real research, label each field in their profile and memory with one of three words (Grounded, Derived, Assumed) telling you what is behind it, hold positions, disagree with each other, and get checked by a critic agent before each response is delivered. The output is something you can audit and trust enough to make a real decision. You won't get that from a roleplay prompt.
Start with a free project, or walk through Candor with us first.