Sense-check it before you ship it.
You’re about to spend six weeks building something. Hand it to a synthetic cohort first — have them read your copy, attempt a task on your live site, or answer your interview questions — and minutes later you get the argument they had, the places they got stuck, or what they said. A tool you use, not a service you book: set it up, run it, read it.
Try it on the thing you’re building →Three ways to test
One synthetic cohort — dice-rolled lives, mood and money-pressure and patience, no two the same — pointed at whatever you need to know. Pick the shape of the question. No dashboards to learn; set up, run, read, honestly labelled the whole way.
They react
Paste copy, a page, or a screenshot. Fifteen strangers read it independently and disagree — you get where they land, where they split, and the sharpest dissent, word-for-word.
They do a task
Point the cohort at your live site with a goal — find the returns policy, buy the blue one, cancel the plan. They actually try, step by step, and you see who breezes through, who stalls, and where they give up.
They answer your questions
Write a set of questions. Each member answers in their own voice, and you get the themes question by question, the outliers who disagree, and a flag on any question whose wording nudged the answer.
What you actually get
Never a single verdict. A reaction study gives you the distribution — where the cohort agrees, where it splits, the strongest dissent quoted verbatim — plus 2–4 testable hypotheses, each citing the split that motivates it and the cheap real-world check to run next. A task study shows the run itself: who finished, who stalled, and where. An interview gives you the answers question by question, the outliers, and a flag on any question that led them. Every one is a synthetic starting point for real research, labelled as exactly that.
Don’t take our word for it — here’s a real study, labels and all. (It’s the study we used to test this very page. The cohort was brutal. We rewrote it. That’s the product.)
Why not just ask ChatGPT?
Your AI assistant is on your side. Your audience isn’t.
Ask one model to “act as a customer” and you get one agreeable voice doing impressions of a crowd — we measured it: 9 of 15 identical openers, one opinion in fifteen costumes. SpareBrain deals every simulated reader their own hand — mood, money pressure, expertise, patience — from seeded dice the model never touches. Same dice, same crowd: change one line and re-deal the identical audience to see what moved.
What it refuses to tell you
No purchase-intent scores. No willingness-to-pay. No conversion predictions. Not because we’re modest — because synthetic numbers for those questions don’t transfer to your product and your audience, and a number that looks like a forecast will be treated as one. A built-in check strips those numbers back out automatically, even when they sneak into an otherwise legitimate result. What you get instead: the comprehension, objection, and trust findings that actually decide whether people buy — plus a calibration path, once a use case earns it, for when you truly need a number. The tool that’s honest about what it can’t know is the one you can trust about what it can.
Earned trust, not asserted trust
Every use case starts labelled exploratory-only, and keeps that label until paired comparisons against real research earn it away — three matched comparisons to advance, five with distribution agreement to be called validated, stale after 90 days, demoted on failures. The thresholds are public, the comparison workspace is in the product, and the labels are baked into every export. When we can show synthetic splits matching real ones, you’ll see the receipts — not this paragraph claiming them.
Pricing
Free during beta. After that: solo makers pay per bundle of studies — no subscription, and you’ll always see a cost estimate before you run anything. Subscriptions exist only for teams who want seats. Exact numbers published before we charge anyone a penny.
Your data
The cohort is synthetic; you aren’t. SpareBrain doesn’t run analytics on you while you evaluate it, your studies are your org’s private research, and the whole lot exports in one click — a research tool that studied its visitors without saying so would deserve the dissenting voice it gets. The usage metadata we do keep (which features get used, how studies go) is aggregate and metadata-only, never your research content — see what we measure and why.
Every output is synthetic and labelled as such. SpareBrain is a research instrument, not an oracle.