Skip to content
Subconscious

Synthetic Responses vs. Causal Experiments: Choosing a Market Research Tool

On this page

Synthetic Users and a causal experiment platform such as Subconscious both put questions to AI-generated respondents, and they publish different kinds of evidence. Synthetic Users runs interviews and surveys with simulated participants to explore a problem, a message or a concept, and describes itself as "a discovery co-pilot, not a replacement for real research" (syntheticusers.com, read 2026-11-12). Subconscious runs a designed, randomized choice experiment on simulated buyers and reports the estimated effect of each change on modeled choice with a confidence interval; its published benchmark grades such runs against published human experiments, study by study. Use Synthetic Users for the discovery uses it names: exploration, concept and messaging testing, and continuous insight between research phases. Use a causal experiment when you need candidate estimates of which option leads and in which direction. In the public registry on 2026-11-12, direction held in 127 of 308 detailed comparisons and rank in 103, and effect sizes are not yet validated, so human research on the leading options still measures the outcome for the options you ship.

What each one is, from first-party sources

Synthetic UsersSubconscious
OutputInterviews, surveys, concept tests, card sorts and website runs with simulated participants, all drawn from one token pool ([pricing](https://www.syntheticusers.com/pricing)): "Test a product, a message or a policy on synthetic respondents and see what they would do, and why, in minutes" ([home](https://www.syntheticusers.com/), read 2026-11-12)Estimated effect of each price, feature or message level on modeled choice, with a confidence interval, from a randomized factorial design
Respondent model"a multi-agent architecture where AI participants develop individual personality profiles based on the OCEAN model" ([home](https://www.syntheticusers.com/)); "Every synthetic user is a distinct, brainscan-powered individual" ([pricing](https://www.syntheticusers.com/pricing))Simulated respondents matched to census statistics from the American Community Survey and the National Household Travel Survey; a discrete choice model estimates parameters with confidence intervals ([fidelity.subconscious.ai](https://fidelity.subconscious.ai/))
Stated use cases"Early exploration & problem discovery", "Concept & messaging testing", "Continuous insight between research phases"Pricing, product and message decisions where several options compete
Validation it publishesThe vendor reports synthetic-organic parity of 85% to 92% "in independent comparison studies", "Measured across thematic overlap, depth & qualitative alignment", depending on audience type, and "21+ Peer-reviewed papers supporting the synthetic research thesis" (vendor-reported support for the thesis) ([pricing](https://www.syntheticusers.com/pricing) and [home](https://www.syntheticusers.com/)). A separate undated methodology post (read 2026-11-12; it references February 2024) compares 8 synthetic interviews with 8 organic interviews and says "Quantitatively calculating a parity number in this context involves subjective judgment" ([how we measure success](https://www.syntheticusers.com/science-posts/how-we-measure-success), read 2026-11-12). A science post reports election calls published before the vote: vote shares "within 0.3 points of each candidate's share" in Brazil's presidential vote, and "the winner in 60 of 70 races decided so far" in the 2026 US midterms ([we called it](https://www.syntheticusers.com/science-posts/we-called-it-brazil-us-midterms-2026), read 2026-11-12)Mean Spearman rank correlation of estimated choice parameters 0.55 across roughly 300 replications of published studies, 0.73 across 43 design-filtered studies (July 2026 working paper, not peer reviewed, summarized on [fidelity.subconscious.ai](https://fidelity.subconscious.ai/)); a public [study registry](https://fidelity.subconscious.ai/study) (read 2026-11-12) with per-study direction, rank and size verdicts
Role of human research"Organic user research stays essential for validation and edge-case work" (home FAQ)Human research measures the outcome for the tested options; in the 27 registry studies with a stated size ratio, simulated effects ran from 0.01 to 3.8 times human, 20 graded inconclusive and 7 not equivalent ([registry](https://fidelity.subconscious.ai/study), read 2026-11-12)
Pricing"Plans start at $12,500/year", token-based; a per-interview table lists Synthetic Users at "$40-63" against a recruitment agency at "$80-120" ([pricing](https://www.syntheticusers.com/pricing))Studies are scoped through a decision review

Synthetic Users' pages show no publication date; every fact above was read on 2026-11-07, rechecked on 2026-11-12, and should be re-read before you rely on it.

What each kind of evidence proves

Synthetic Users' published use cases are exploration, concept and messaging testing, and continuous insight between research phases. Its parity figures measure thematic overlap, depth and qualitative alignment against organic interviews, and the undated methodology post (read 2026-11-12; it references February 2024) says that calculating its parity number "involves subjective judgment". That is the evidence to weigh when you want words, objections and hypotheses from a simulated interview. Its election post compares simulated vote shares with official counts. The cited Synthetic Users evidence reports no matched comparison for product price or feature effects, and no source here compares the two vendors on the same study, so nothing on this page ranks them on accuracy.

A causal experiment publishes a different kind of result. Because the design randomizes every attribute jointly and estimates each effect from designed contrasts, the result is a comparison: option B moves modeled choice by this much relative to option A, with this interval. The registry compares modeled and human effects by direction, rank and size. In the 308 detailed registry comparisons on 2026-11-12, direction held in 127, was mixed in 92 and failed in 89; rank held in 103, was mixed in 93 and failed in 112; 73 studies failed on both (registry). In the Mintel moisturizer conjoint, simulated respondents built from demographics alone reached 0.79 to 0.88 Jensen-Shannon similarity on attribute distributions against two human waves of 1,000 and missed the sharp human demand drop between $34 and $48 (Mintel case). In River City Federal Credit Union's slogan study, the simulated experiment identified the weakest slogan before spend, and that slogan also ranked worst in the live Facebook and Instagram campaign (River City case). The validation page carries the registry's study list and the misses.

Choose Synthetic Users when

Choose a causal experiment platform when

Using both

Discovery first, then the experiment. Bring the options and the decision deadline to a decision review to scope the experiment step.

How this page was made

By: Avi Yashchin, CEO and Co-Founder. Rebuilt with AI assistance on 2026-11-07, replacing an earlier version that was paraphrased from a third-party article, and updated on 2026-11-12. Synthetic Users facts were read on syntheticusers.com (home, pricing and the how-we-measure-success post) on 2026-11-07 and rechecked on 2026-11-12; the pages show no publication date. Subconscious figures come from the fidelity.subconscious.ai summary of the July 2026 working paper, the public study registry (read 2026-11-12) and the public case pages.