Aaru AI vs Artificial Societies: compared

Artificial Societies reports 86% distribution accuracy across 1,000 real-world surveys against a 91% human self-replication ceiling, according to its Survey Evaluation Report, January 2026. Distribution accuracy measures how closely simulated and human opinion distributions overlap. The human ceiling reflects how consistently people repeat their own answers. It gives you a measured reference for judging the simulation.

The same report puts biography-prompted large language models at 67% distribution accuracy. Those are personas generated by asking a model to adopt an invented biography. Our method instead builds from observations of people and their relationships, as our method and evaluation page describes.

We also test whether the answers make sense together. The January 2026 report records 89% internal coherence, measured with Cronbach’s alpha, against a human-panel range of 60–95% and an acceptable threshold of 70%. This measure checks whether questions about the same underlying attitude produce related answers. It helps assess the relationships between opinions; it does not establish whether the findings generalise to a real population.

Our September 2026 validity framework explains why that distinction matters, and how to evaluate a synthetic research vendor covers its eight tests in more detail. Matching the total for each answer can leave the combinations of beliefs wrong: the simulation might reproduce how much support a proposal receives without reproducing who supports it. We want you to inspect the responses behind the total, particularly when the decision depends on a specific audience segment.

Artificial Societies helps organisations understand people before making consequential decisions. We start with observed data, including published research, online commentary, reviews and anonymised social data, alongside public records of voting and investment. Our method page describes how we reconstruct individual belief systems, validate personas against audience benchmarks and prior behaviour, then connect them through relationships and patterns of influence.

A society contains 12 to 3,500 personas, according to Artificial Societies’ method page. You can examine a result for the population, a segment or a persona. Each survey question also asks for a verbatim reason by default, a feature described in our consumer-goods case study. That gives you the explanation behind a selection when you investigate why a message or product concept appealed to a particular group.

The network is part of the research design. Our Simulation Engine combines individual belief systems with social dynamics, exposing personas to scenarios and following opinion across the society. For a communications team, the research can therefore address the response to a proposed announcement and how opinion moves between audiences. We work from diverse observations of the audience; where those observations are missing, we would first establish the evidence needed to build it.

Aaru AI describes simulated populations used to predict what people will do before a business commits to a decision. Its homepage (opens in a new tab) sets out a sequence: define the population, change its conditions and examine the resulting behaviour. Product Innovation and Scenario Planning appear among its named applications, alongside Strategic Communications.

Aaru’s simulation page (opens in a new tab) describes public data such as census records and labour statistics, licensed behavioural data such as transactions and visits, and customer context such as internal segments and purchase histories. Aaru says it trains on population distributions and differences between people under varying conditions. It also describes modelling related traits together, using age and work status as an example of characteristics that are not independent.

Aaru distinguishes measuring performance against withheld evidence from deciding whether that evidence supports a particular business decision. Its simulation page calls the former evaluation and the latter validation. For a buyer considering Aaru’s forecasting approach, that distinction provides a useful demonstration request: ask the team to show the evidence, assumptions and uncertainty behind the outcome you need to predict. Its published method makes the decision context part of the assessment.

Aaru AI and Artificial Societies overlap in strategic communications. Aaru’s homepage (opens in a new tab) describes examining how employees, customers, investors and the media interpret an announcement, and how action, timing and spokesperson affect trust during a crisis. Those are recognisable communications questions.

We would compare the output you need to act on. Our Teneo case study describes a team examining approval scores, emotional sentiment and verbatim reactions to competing narratives. Aaru’s published description centres on population-level behavioural outcomes. These accounts give you different starting points for a demonstration: tracing a reaction back to a persona, or examining the behaviour forecast under a changed condition.

Sources: our method and evaluation and September 2026 validity framework; Aaru’s homepage (opens in a new tab) and simulation page (opens in a new tab); EY’s account, 7 October 2025 (opens in a new tab).

Our survey benchmark concerns opinion, and our validity framework recognises the gap between stated intention and behaviour. We use simulation to investigate decisions with that scope in view. A proposal for an earnings-call rehearsal should show how you will inspect investor reactions; a proposal for demand forecasting should show how the forecast relates to observed demand.

EY published a blinded recreation of its wealth research with Aaru and reported a 90% median correlation across 53 single-select questions, compared with original research involving 3,600 affluent investors in more than 30 markets (EY, 7 October 2025 (opens in a new tab)). A Big Four firm publishing its own comparison is a substantive reason to consider Aaru. You can inspect the audience and question format behind the headline.

Aaru’s wealth-research case study (opens in a new tab) dates the run to August 2025 and describes a median Spearman correlation of 0.90. Spearman correlation measures agreement in the ranking of answers. The study compared simulated and surveyed answer positions, so the result tells you how closely their ordering agreed across that question set. EY published the result within a commercial partnership; it is partner evidence rather than independent peer review.

Our distribution-accuracy measure asks how much the simulated and human answer distributions overlap. Ranking agreement and distribution overlap answer different questions, so ordering these vendors by their headline percentages would produce a misleading recommendation. The EY result supports evaluating Aaru for a comparable research task. Our Survey Evaluation Report supports examining opinion distributions and the additional response-quality measures we publish. A buyer should ask each team for evidence relevant to the intended audience and decision, with the measurement named before the percentage.

We would send you to Aaru AI for forecasts of revealed behaviour, especially when transactions or visits are central to the question. Those are among the grounding signals Aaru names on its simulation page (opens in a new tab). Its premise puts actions at the centre of the forecast, which is a reason to examine Aaru when stated preference is insufficient for your decision.

Aaru’s Scenario Planning description (opens in a new tab) addresses changes in demand, retention and price sensitivity under recession, rapid growth or sustained inflation. We would favour that stated approach for a question about how market conditions change behaviour. Ask Aaru to demonstrate the conditions it varies and the evidence used to judge the resulting forecast, using the evaluation and validation distinction its simulation page sets out.

Population scale provides another reason. Aaru’s Breakwater case study, 17 July 2026 (opens in a new tab) describes a quarterly tracker covering 71 judgements across 40,000 simulated investors. The case study reports conviction spreads, with the narrowest at one percentage point and the widest at 63 points. These are tracker findings, not an accuracy score. If a capital-markets tracker resembles the work you want to commission, Aaru has a published example at that scale to discuss with you.

Choose Artificial Societies when you need to rehearse a confidential decision and examine the audience’s reasons before changing the material. Our hyperscaler earnings-call case study describes an investor-relations team testing prepared remarks and guidance that it could not show to real investors because of selective-disclosure constraints. We built societies around the stock and tested both the expected questions and responses to alternative scripts.

The analyst society contained 14 personas representing the analysts who had asked questions on the company’s last eight earnings calls, according to the case study. The company expected its margin reset to dominate. Simulated analysts instead pressed on demand durability and financing commitments, and the study reports that the live questions followed that pattern. The result gave the team a reason to change its preparation for the live call.

The engagement delivered a pre-call briefing and platform access to rerun scenarios as the numbers changed. Artificial Societies’ case study records delivery in 48 hours, with no draft, number or script leaving the secured environment. The 48-hour figure covers this engagement from start to finish.

Teneo provides another example of research you can interrogate. In late 2025, it used Artificial Societies to compare six technology narratives for a major US company, and the Teneo case study records 189,756 unique responses and delivery of a written report plus an interactive platform. The team could explore the responses at aggregate and individual persona level, then refine messaging for each audience segment.

Aaru alternatives should be assessed against the research you intend to run. Our Aaru alternatives page sets Aaru AI beside other platforms by the outcome each one predicts or explains. For our method, we would start with the questionnaire and audience evidence. Our September 2026 validity framework describes tests that reorder answer options, rephrase questions and remove supporting evidence from a claim. These tests examine whether a simulation responds to substance or merely to presentation. Ask us to explain how that distinction applies to the message you want to test.

Artificial Societies’ peer-reviewed paper on group behaviour (He, Wallis, Gvirtz and Rathje, British Journal of Psychology (opens in a new tab), December 2024) supports the network modelling, not the survey accuracy figures; the science page explains what it found.

You should also ask what the research will let you change. Our transport-policy case study describes blind testing of competing narratives, followed by exposure to damaging coverage. The client received recommendations on positioning and proof points, with access to examine segment results and individual reactions. For that kind of decision, we would make the ability to investigate why a narrative holds up under pressure a condition of the purchase.

We publish this comparison. We used the sources listed at the end of this page: our method and evaluation material and case studies, Aaru's own pages, the EY account and third-party comparisons. We conducted no product trials for this article.

The table applies the same criteria to both products. It describes published methods and measurements without converting an undocumented feature into an absent capability. Corporate finances are outside the comparison’s scope. We recommend products by the decision and evidence described, rather than assigning an overall winner.

Information is believed accurate as of 17 September 2026. Third-party trade marks belong to their owners; naming them implies no affiliation or endorsement. Send corrections to support@societies.ai so we can amend the comparison and its publication details.