What are Artificial Societies? Company and method explained

Artificial Societies’ distribution accuracy is 86% across 1,000 real-world surveys, against a 91% human self-replication ceiling, according to its Survey Evaluation Report, January 2026. Distribution accuracy measures the overlap between simulated and human opinion distributions: the proportions choosing each answer. The human ceiling reflects how consistently people reproduce their own answers when asked again.

Source: Artificial Societies, Survey Evaluation Report, January 2026, presented on our evaluation page.

Our evaluation also examines the answers beneath the distribution. Artificial Societies’ personas contradicted themselves in under 2% of cases, compared with about 9% for human panels, according to the same January 2026 report. This measure counts inconsistent answers, rather than factual errors about the world. For a researcher examining audience reactions, agreement with human opinion and consistency within the responses answer different questions about the data.

The evaluation measures internal coherence too: whether answers to questions about the same underlying attitude align. That matters when you want to examine relationships between opinions within an audience. We publish these measures together because the population result and the beliefs represented within it both matter to a study of how a decision shifts opinion.

Our audience simulations begin with a definition of the people whose reactions matter. You can specify demographics, organisations and job titles, or use your own audience data. The construction process has named stages: Persona Genesis, Society Assembly and the Simulation Engine.

Persona Genesis gathers observations from sources such as published research, reviews and anonymised social data. Public records of voting and investment also contribute. Our method page names LinkedIn, Reddit and Amazon among the sources. Psychometric methods reconstruct how individuals reason, using those observations to model their belief systems. We check every persona against audience benchmarks and prior behaviour before admitting it to a society.

Society Assembly groups personas by shared characteristics and connects them through relationships and patterns of influence. The Simulation Engine then exposes them to material or scenarios in the context of the audience’s current situation. You can examine the resulting opinions for the population, a segment or an individual persona.

Our guide to how AI persona simulation works explains the construction process. In client work, those levels of analysis take a practical form: Teneo could inspect approval scores, emotional sentiment and verbatim reactions, alongside a written report identifying the strongest narrative for its client.

A prompted AI persona, one kind of synthetic persona, starts with an instruction to play a role. Our method page gives the example “you are a 34-year-old teacher from Ohio”. That instruction supplies a demographic description. Our approach starts with observations of people’s behaviour and builds a model of their beliefs, then connects personas so the simulation can examine social influence as well as individual responses.

We validate the personas because plausible speech is insufficient evidence of a stable belief system. A model can sound like the person it represents without holding consistent attitudes. Checking prior behaviour and audience benchmarks gives us something against which to test the representation before using it to explore reactions to new information.

Within synthetic market research, our societies model both individual responses and the influence people have on one another. For a strategy involving policymakers and industry peers, we consider those relationships part of the research design. Personas are connected through observed relationships and patterns of influence so the simulation can examine how opinions form and spread through the audience; why opinion spreads through networks explains what those connections add.

Epstein and Axtell’s artificial societies are abstract agents following local rules, built to show how social patterns emerge; Artificial Societies’ personas are built from observations of real people, to model how a particular audience will respond to a decision.

Epstein and Axtell’s Growing Artificial Societies: Social Science from the Bottom Up (opens in a new tab), published in 1996, studies social behaviour emerging from individual interactions. Their Sugarscape agents seek sugar and have attributes such as vision and metabolism. Artificial Societies’ personas draw on observed human behaviour and undergo validation before entering a society. The connection to that research tradition is the study of collective behaviour through interactions among individuals.

Artificial Societies’ method grew out of years of research originating with the first large-scale AI society paper (opens in a new tab). He, Wallis, Gvirtz and Rathje reported a simulation of 33,299 chatbots in the British Journal of Psychology in December 2024. Communities formed around shared language; among the English-speaking chatbots, communities also formed around similar content.

The paper concerns homophily, the tendency to form communities with similar others. Our January 2026 survey evaluation measures opinion accuracy separately. The peer-reviewed finding establishes a specific collective behaviour reproduced by chatbot societies, giving our work on networks of personas a research foundation in how communities form.

Teneo used Artificial Societies’ simulations to advise a major US company on its technology narrative before a public launch. During the late-2025 engagement, Artificial Societies built three societies containing over 5,000 AI personas to represent consumers, industry peers and policymakers, according to our Teneo case study. The strategy was confidential and the audiences were hard to reach.

The Washington D.C. society represented policymakers, lobbyists, think tanks and political influencers. The tech-leader society included founders, academics and engineering talent, while the general-population society matched US census distributions. We grounded the personas in real-world profiles, social listening and qualitative research.

Testing six competing narratives through 30 deep-dive questions produced 189,756 unique responses, according to the late-2025 Teneo case study. Surveys and experiments captured reactions within each society, at aggregate and individual persona level. We identified the strongest overall narrative and adjusted the messaging for audience segments.

Teneo received a written report and access to an interactive enterprise platform to examine approval scores, emotional sentiment and verbatim reactions. Artificial Societies delivered results in less than three weeks, including building the three societies from scratch, according to the late-2025 case study.

“What we were able to accomplish with Artificial Societies would simply have been impossible with traditional market research.”

Artificial Societies is used by Fortune 100 organisations to test how strategies shift opinions, as our method page describes. Consider us when your decision involves audiences whose relationships affect how opinion develops. Our work covers strategic communications, government affairs and investor communications, including sensitive material you need to examine before release.

Teneo’s technology-strategy work shows the value of examining audiences separately. The engagement identified an overall narrative and then adjusted messaging by segment. We would put that distinction at the centre of a communications study: the overall choice and the wording for a particular audience are decisions you can test together.

Our method requires diverse observations of the audience, whether public or client-provided. Where those observations do not exist, there is nothing to ground personas in. When defining a society, bring the audience specifications and existing research you hold so we can discuss the observations available for that population.

Our September 2026 article on simulation validity states the scope we work within: “Synthetic audiences will never be a true substitute for the people they seek to represent. But we can test them more thoroughly than any panel of respondents.” Teneo used that testing to compare alternative narratives before its client committed to a public strategy.