Brox and Candor both answer customer questions with AI participants, but they start from different places. Brox builds a digital twin of each real person it models. Candor builds participants from evidence about a type of customer, so none of them is a copy of anyone. That one choice shapes accuracy checks, consent, scale and price.
| Dimension | Brox | Candor |
|---|---|---|
| What a participant is | A one-to-one digital twin of a specific real person | A participant built from evidence about a type of customer. It matches no real person. |
| What it's built from | Hours of real and AI-led interviews with each person, with up to 300 pages of text behind some twins. People come from Brox's recruited panels, custom audiences or your own customers. | Your uploaded research first, then web research. Personality traits come from published population data for the participant's region and occupation. |
| Size | 30,000 twins | Up to 24 participants per audience, built for each study |
| How you ask questions | Type a question or scenario and get a dashboard back in under 15 minutes, with each number traced to the twins behind it | One-to-one interviews, run by you or by Candor's automated interviewer, followed by a written report |
| Where it runs | Live in the US, UK, Japan and Turkey | The countries you choose for each study, from a list of about 50 |
| Checking a result against a real person | Possible, because each twin models one person. Brox reports an F1 score of about 85%. | Not possible, because no participant is a real person. Candor hasn't yet published a test of its output against real people. |
| Pricing | Subscriptions from $100,000 to $1.5 million a year, with unlimited use | Free: $0 a month, $10 per automated interview. Standard: $75 a month, $7 per automated interview. Interviews you run yourself are free. |
Brox facts come from brox.ai, its platform page and VentureBeat’s report of May 6, 2026, checked on September 26, 2026. If a row is out of date when you read this, tell us and we’ll fix it.
The difference is who each participant stands for. A Brox digital twin stands for one real person. Brox describes its twins as behavioral replicas of real people, built to predict the decisions those people would make.
A Candor participant stands for a type of customer. Candor starts from a description of your audience, gathers evidence about it and builds each participant from that evidence. The participant has an invented name and matches no one.
Neither design is better in general, because they answer different questions. A twin suits the question “what will these real, recruited people decide?” A Candor participant suits the question “how does this kind of customer reason about my problem, and why?”
Checks against real people. Every twin is built from one real person, so Brox can compare a twin’s answer with that person’s. Its platform page reports an F1 score of about 85%. F1 is a standard measure of how often predictions are right, balancing misses against false alarms. The page doesn’t describe how the test was run, so ask Brox for the method if the number matters to your decision.
Depth on each person. According to VentureBeat, Brox runs hours of real and AI-led interviews with each person it twins, covering drivers as personal as upbringing and relationships. Some twins have up to 300 pages of text behind them.
Fast answers you can trace. Brox says you can type a question or scenario and get a dashboard back in under 15 minutes. Each number on it traces back to the interview moment behind it.
An enterprise track record. Brox is live in the US, UK, Japan and Turkey, and its site names clients including Google, Amazon, Pfizer and Chase. Candor is newer. If an established client list is part of your procurement case, Brox has one and we don’t yet.
Candor gathers evidence about your audience before anyone exists to interview: your own research first, then web research. It groups the audience from that evidence and builds each participant from it. Personality traits are drawn from published population data for the participant’s region and occupation.
The details about who a participant is, how they behave, what they believe and how they talk each carry a label. Personality and bias scores don’t. Grounded means there’s something you can check behind it. Derived means it follows from the evidence without a source stating it outright. Assumed means Candor filled it in and says so. How it works walks through the full process, and why we don’t build digital twins of real people explains the reasoning behind the choice.
Brox can check a twin against the person it models. Candor can’t, because there’s no real person to check against. Candor also hasn’t yet published a test of its output against real human answers. We make no accuracy claim for it, and its findings aren’t a forecast of what any named person will do.
What Candor gets in return is less to govern. No participant is built from a real person’s data, so there’s no one whose consent has to be recorded, or honored when they withdraw. The files you upload are the exception. They hold real people’s data, and Candor reads them to pull out findings about a type of customer.
Price is the other difference. Brox is an enterprise subscription. Candor has a free plan and charges per automated interview, with no contract, so a small team can run a study this week. The pricing page has the details.
Choose by the question you need answered, and by what you’re willing to govern.
This comparison is one of several. See Candor vs Synthetic Users for another synthetic research platform, Candor vs UserTesting and Traditional vs synthetic user research for research with real people, or the full comparison hub. For an overview of the category, see what is synthetic user research.
Yes, for some jobs. Both answer customer questions with AI participants, so teams weigh one against the other. They're built for different questions, though. Brox models specific real people and predicts what they'll decide. Candor models types of customers and runs one-to-one interviews to learn how they reason and why. If you need a forecast for a recruited population that can be checked against real answers, Brox fits. If you want to explore a problem, test a concept or pressure-test a price before real research, and you'd rather no participant be a copy of a real person, Candor fits.
Brox sells enterprise subscriptions. VentureBeat reported in May 2026 that they start at $100,000 a year and reach $1.5 million for larger deployments, with unlimited use during the contract. Brox's own site doesn't list prices. Candor publishes its pricing. The Free plan is $0 a month with $10 automated interviews. Standard is $75 a month with $7 automated interviews. Interviews you run yourself are free on both plans. A 24-participant study on Standard costs $243 in its first month, monthly fee included.
No. Candor doesn't build a twin of any real person. Each participant is built from evidence about a type of customer: your uploaded research, web research and published population data. Participants get invented names, and there's no feature for building one from a named person. So Candor can't predict what a specific individual would say, and it doesn't claim to. It also means no participant carries one person's record or needs one person's consent.
They measure different things, so there's no direct comparison. Brox reports an F1 score of about 85% for its twins, which can be checked against the real people they model. Its platform page doesn't say how that figure was measured. Candor has no comparable figure. Its participants aren't copies of real people, and it hasn't yet published a test of its output against real human answers. Candor aims to show how a type of customer reasons, with most profile details labeled by what they rest on. Treat its findings as a map of what to test with real customers next, not a forecast.
Yes, in different ways. Brox says it can build twins from your own customers, as well as from its recruited panels or custom audiences. Candor treats your uploaded research, such as interview transcripts, survey exports and CRM notes, as its strongest evidence. It pulls findings about a type of customer out of those files and builds participants from the findings, not from any one person in them. You can remove names and identifiers before you upload, and the research still works as evidence.
Start with a free project, or walk through Candor with us first.