Reverb gives you a fast, honest read on a product concept — first from Echoes, synthetic respondents that react like your target customer, then from verified humans when you need ground truth. Here's how it actually works.
Most research forces a choice: fast and cheap, or rigorous and slow. Reverb makes it a sequence. Screen broadly with Echoes in minutes; validate the survivors with real, verified people. The same brief, the same questions, one continuous flow.
Synthetic respondents, each briefed to react as a specific slice of your market. Test twenty concepts, sweep a price, interrogate any reaction — for the cost of a few cents and a couple of minutes. A directional read that finds the risks fast.
When a decision is high-stakes, confirm the read with real, identity-verified people sourced to ESOMAR-grade standards. You get a side-by-side delta against the Echo prediction — and a calibration score that says exactly how close the synthetic layer was.
Ask a language model to rate a concept “1 to 5” and it does something useless: it hugs the middle, almost never commits to a strong yes or no, and produces ratings no real market would ever give. So we don't ask for a number.
Each Echo is conditioned on a real slice of your market — defined by category usage and buying behaviour, the attributes that actually shape a purchase decision. It reacts in character: its priorities, its scepticism, its budget.
Instead of a rating, we collect what the respondent would actually say — an honest, in-their- words reaction to your concept, plus what draws them in and what holds them back. No scale to game, no middle to hide in.
We then translate each reaction into a purchase-intent distribution by measuring how closely it resembles a set of calibrated reference points — the meaning of the words, not a number the model invented. Aggregate across the panel and you get a realistic spread of intent, not a flat “everyone says 3.”
The approach behind Echoes — eliciting language and mapping it to ratings by semantic similarity — was validated by researchers at PyMC Labs and Colgate-Palmolive against thousands of real consumer responses. On head-to-head tests it reproduced human purchase intent strikingly well.
Maier et al. (2025), “LLMs Reproduce Human Purchase Intent via Semantic Similarity Elicitation of Likert Ratings.” arXiv:2510.08338. We build on this method and extend it with our own calibration layer.
Synthetic respondents are a powerful instrument with real limits. The fastest way to lose your trust would be to pretend otherwise — so every Echo report says exactly how far to trust it.
Most synthetic-research tools ask you to take their accuracy on faith. We'd rather show you.
When you validate a finding with verified humans, we measure exactly how close the Echo prediction was and feed that back in. Over time the synthetic layer gets measurably sharper in your categories — and you see the score, per category, on every report.
When you go to humans, they're sourced to professional (ESOMAR-grade) standards and can be confirmed as real, unique people through privacy-preserving identity verification — no personal data stored, each person counted once. The quiet fix for an industry with a real panel-fraud problem.
Run your first Echo panel free. No fieldwork, no waiting, no black box.
Start a study