How the providers compare
Six providers were checked, alongside tools built on prompting general-purpose models. Five train models on data about people, and one uses general-purpose language models. Every published accuracy figure comes from the provider or its founders, and they use different metrics at the audience or the individual level.
| Provider | To get started | Use it in | Technology | Published accuracy |
|---|---|---|---|---|
| Aaru | Not published A pilot first, scoped with Aaru’s team | Web app | Trained on data about people Simulated individuals. A population model, not a language model, built from census, transaction and behavioural data Aaru, 2026 | 3.53 points Mean absolute error, audience level Average over answer shares on 2,993 questions Aaru, 2026 |
| Artificial Societies | Not published A demo first | Web app | Trained on data about people Twins of real people. Its own model, Rothko. Personas grounded in real people’s public online data, connected in a network Artificial Societies, 2026; Artificial Societies, Rothko | 86% Distribution overlap, audience level 1,000 surveys, against a 91% human ceiling Artificial Societies |
| Electric Twin | Not published A walkthrough first. A licence priced by audiences and seats | Web app | General-purpose language models Twins of real people: not described. Personas constrained by survey data Electric Twin, white paper | 4 points Mean absolute error, audience level Best case, published as 96% (1 − mean absolute error). 92% on its stricter NDAM measure, against a 94% human ceiling Electric Twin |
| Listen LabsSalesforce agreed to acquire Listen Labs in September 2026 | Not published A demo first for Listen Twins. Sign-up with a work email covers its interview research only | Web app | Trained on data about people Twins of real people. Listen Twins, one model per real interviewee, built from AI-led interviews Listen Labs | Not published Individual level Tested on held-out interview answers Listen Labs |
| Semilattice | $139 / month Sign up on the site. Simulations on public user models start in November 2026; until then, Custom models only, through a call | Web appMCPAPI | Trained on data about people Simulated individuals. Anacreon, a fine-tuned model per cluster of similar real people Dahiya (Semilattice), 2026 | 0.775 Ordinal alignment, individual level Median cluster, one customer audience; 0.679 exact match Dahiya (Semilattice), 2026 |
| Simile | Not published A request form first | Web app | Trained on data about people Twins of real people. A fine-tune of Qwen, grounded in interviews and its own studies Simile; Park et al. (Simile), 2024 | 86% Ratio to human retest, individual level Interview and survey agents; 83% from interviews alone Park et al. (Simile), 2024 |
| Tools built on prompting | ||||
| Prompting toolsSuch as Minds and Synthetic Users | From $20 / month (Ask Rally); Synthetic Users from $12,500 / year; Evidenza by proposal | Varies Minds has an MCP server and an API | General-purpose language models Simulated individuals. Persona prompts Argyle et al., 2023; Synthetic Users; Minds | Method not published Undefined percentage Figures such as “88% similarity” without a stated method Evidenza |
Based on each provider’s public website and published papers, October 2026. Listed alphabetically. An MCP server lets an AI agent such as Claude run simulations as part of its own work; Semilattice has one, and among the tools built on prompting, Minds does. Semilattice, which publishes this primer, is one of the providers. Accuracy figures are self-reported and use different metrics, so they do not compare directly. Audience-level figures compare answer shares and individual-level figures compare each simulated person with a real one; see Two levels of accuracy. For the published critiques of trained and general-purpose models, see Which critiques apply to which models.