Hard to reach audiences: research with AI personas explained

Professional audiences become hard to reach when recruitment costs, confidentiality or timing prevent the research you need. Spending more on recruitment may address access; it cannot resolve the exposure problem described in our transport-policy case study, where showing draft messages to opinion leaders would have changed the debate under examination.

Elite research can achieve substantial participation. Matias López’s May 2022 study in Quality & Quantity (opens in a new tab) analysed 282 elite samples collected between 1959 and 2020. It reports average response rates of 80% for random sampling and 50% for purposive or availability sampling. López attributes high response rates partly to the time and resources researchers spend securing each person’s participation. We would distinguish a recruitment problem from a decision that cannot safely be tested with its intended audience.

The earnings-call engagement illustrates the latter. An investor relations team at a leading hyperscaler needed to test prepared remarks and a guidance range, but selective-disclosure constraints prevented it from showing the material to real investors. Simulated audiences let us examine reactions to those unreleased numbers and scripts inside a private environment. The study concerned preparation for the call; the responses came from personas representing the audience.

Our Survey Evaluation Report, January 2026 records 86% distribution accuracy across 1,000 real-world surveys. That 86% distribution accuracy is 95% of the 91% human self-replication ceiling reported in the same evaluation. Distribution accuracy concerns the pattern of answers across an audience; the human ceiling reflects how consistently people reproduce their own answers when asked again.

We would use that survey benchmark to assess the architecture, then examine the proposed audience and decision separately. A policy narrative tested with Washington D.C. opinion leaders and an earnings script tested with sell-side analysts ask different questions. The transport-policy and earnings-call case studies describe how we defined those audiences and used their simulated responses. The benchmark gives the survey comparison; the case studies give you the work to inspect before commissioning a similar engagement.

For the earnings-call team, the relevant check was whether the simulation anticipated the concerns that emerged on the live call. For Teneo, the task was to compare technology narratives across distinct audiences. Those are concrete ways to judge whether our research fits your decision, without treating a distribution score as a promise about a particular person’s next answer.

An audience definition starts the work. You can specify demographics, organisations and job titles, or contribute first-party material such as customer relationship management records and existing research. In our transport-policy study, the definition covered lawmakers, regulators, journalists, advocates and industry voices. The society represented more than 360 organisations and 800 distinct job titles, according to the case study. That scope gives a public affairs team something concrete to examine before interpreting the results.

Our method page names social data, public record and first-party data as inputs. Published research, online commentary and reviews sit alongside anonymised social data and public records of voting and investment. Sources include LinkedIn, X, Reddit, Facebook, YouTube, Instagram and Amazon. These observations concern how people communicate and behave in context. Each persona passes a validation gate before entering a society, and we connect personas into networks of influence.

Construction follows the audience. The transport-policy and earnings-call studies matched personas one-to-one with real individuals using public data and social activity. For the consumer-goods study, we combined profiles with similar demographic and psychographic traits into enriched lookalike consumers and de-identified their behavioural traces. These are different ways to ground a persona, with different implications for what the persona represents. For a professional audience, we would settle that representation question before discussing the number of responses to collect.

Our published engagements include policy audiences, financial analysts and technology leaders. The table compares defined audiences with the access problem each engagement addressed. The persona counts describe the simulated groups; they are not counts of people recruited for interviews.

Source: Artificial Societies, Teneo, transport-policy and earnings-call case studies. Teneo describes a late-2025 engagement.

The analyst society had a precise membership rule: participation in the company’s last eight earnings calls, as the earnings-call case study explains. Its 14 personas represented that roster. A public affairs audience needed a different definition, spanning the people who draft, interpret and debate policy. We would ask you to define membership with that degree of care, rather than start with a desired sample size.

Senior executives belong among the professional audiences this method addresses, subject to observed data. The delivered examples above give you evidence involving analysts and policy audiences; the buy-side group also includes chief investment officers. If your decision concerns senior executives, use their roles and the observations available about them to define the proposed work.

For Teneo, we helped a major US company decide how to position its technology strategy before public launch. The Teneo case study describes a late-2025 engagement testing six narratives across three societies. The Teneo study describes a General Population society of 2,381 personas, specified to match census distributions, alongside Washington D.C. and Tech Leaders. The team could examine the public response alongside specialist reactions.

We gathered 189,756 unique responses from 30 deep-dive questions across the six narratives, according to the Teneo study. We identified the strongest overall narrative and adjusted messaging for each audience segment. Teneo received a written report and an interactive platform to inspect approval scores, emotional sentiment and verbatim reactions. The study records delivery in less than 3 weeks, including society construction. The practical gain was a tested choice of narrative with the audience differences still visible.

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

The transport-policy study tested two competing narratives and their proof points after establishing baseline attitudes, then exposed the simulated audience to damaging coverage and hypothetical scenarios. The transport-policy study records 250,000 responses from more than 170 questions across three phases. We diagnosed a standing credibility gap and identified where the arguments survived or failed under pressure. The client received an advisory report with recommendations on positioning, proof points and risk, plus access to inspect segment cuts and individual persona reactions. No message reached a live audience during that engagement. For a public affairs lead, the useful output was a view of which arguments to commit to before entering the debate.

Our hyperscaler earnings-call study records a disagreement between the company’s expectation and the simulated analysts’ priorities. The investor relations team expected a reduced margin guide to dominate the call. The personas concentrated on demand durability and financing commitments; the live questions followed those themes. We treat this as a case-study result about the concerns anticipated for that call.

The earnings-call study covered 2,400+ personas across Sell-Side Analysts, Buy-Side Investors, Corporate Ecosystem and Financial Media, producing nearly 125,000 responses. The work tested investor reactions to alternative prepared remarks and the likely spread of the story. According to the case study, simulated coverage matched the following morning’s framing: growth in the headline, margins as a cost story and financing as an emerging risk. The team received a pre-call briefing and access to retest its script as the numbers firmed up. The rehearsal let the investor relations team prepare for concerns that differed from its own reading of the announcement.

Our audience simulations require diverse observations, either public or client-held. That is the scope we would establish with you before building a society. If the intended audience leaves no such trace, we would begin with research that obtains it. The published method gives you concrete inputs to discuss, from commentary and public records to your existing research.

We also distinguish anticipated responses from actual behaviour. Our discussion of simulation validity explains the gap inherited from surveys: what people say can be a weak predictor of what they do. We would use simulation to narrow the options, then confirm through field research where the decision requires it. In confidential policy or investor work, the immediate value is testing material before exposure; any later claim about behaviour needs evidence of that behaviour.