Skip to content

Knowledge base

Crisis communication simulation: test the statement before you issue it

By Artificial Societies
Published
Updated

Crisis communication simulation tests a crisis statement, and the way you plan to deliver it, on simulated versions of the audiences that will judge it, while the wording can still change. Artificial Societies, which models audiences as networks of AI personas, tested 24 messages and creatives for a major consumer goods company on 1,174 personas in five societies, one per audience, gathering 328,576 responses in under two weeks, according to its crisis case study. The client chose its crisis strategy from the findings, which also identified a likely backlash against certain activations.

What did Artificial Societies’ crisis simulation find?

Artificial Societies’ crisis simulation found that the client’s best path lay in evidence-backed messaging that directly addressed its steps to mitigate and resolve the issue, according to the crisis case study. The client, a major consumer goods company, faced a headline-grabbing crisis that threatened its customer credibility and its licence to operate. Its board wanted to repair the company’s reputation and retain market share. Views differed widely on the source of the crisis, who was culpable and what should follow. In our reading, the finding gave the board a statement it could make without first settling the argument about blame.

On delivery, Artificial Societies’ simulation showed how and when to reach each key audience. It found a likely backlash against certain activations in the broader context of the issue, and the delivery mechanisms that produced the greatest reputational uplift among the simulated audiences. A statement that tests well as text can still fail in the activation that carries it, so weigh the delivery result as heavily as the wording.

The client used the findings and recommendations to decide its crisis strategy and communications. It was then able to run follow-on simulations of how to deliver the chosen messages and where activations might backfire by audience segment. Every reaction in the study is simulated, and the case study does not report whether or how the statements were issued, or how the public responded.

Why not test a crisis statement on a real panel?

A real panel was not an option for the consumer goods company in Artificial Societies’ crisis case study, for three reasons the study gives. The crisis needed fast-turnaround insight. Resolving it meant hearing from hard-to-reach stakeholders, including policymakers, regulators and commercial decision-makers, the recruitment problem our guide to hard-to-reach audiences covers. Many of the options on the table were also highly sensitive, and a human panel would have seen what the board was still weighing.

Crisis audiences also read the same words differently. Our crisis engagement kept them as separate societies: Policymakers and Regulators, Media, Influencers, Consumers, and Commercial Customers. A sentence that reassures a customer can read to a regulator as an admission, and agreement on the wording inside the company says nothing about how any of the five will take it.

How does crisis communication simulation test a statement?

Crisis communication simulation, as Artificial Societies ran it in the crisis case study, has three stages:

  1. 1.

    Build each audience. We built 1,174 AI personas grounded in real, anonymised profiles and conversations from social listening and deep research. We also mapped each audience’s social influence networks, the mechanism our page on networked audience simulation explains.

  2. 2.

    Take a baseline. Before any exposure, we surveyed each society’s view of the company and the issue, covering sentiment and likely purchasing behaviour.

  3. 3.

    Expose, compare and reset. With the client, we tested a battery of messages, statements and creative activations, 24 in all, each targeted at the relevant audiences, and compared reactions after exposure with the baseline, the design described in our guide to message testing. Throughout, we built up and then erased the societies’ memories of each message and its delivery, before and after every exposure.

The memory reset lets the same simulated respondent judge every draft as if it were the first. A person who has read one draft statement cannot forget it before reading the next, so a human study avoids order effects by giving each draft to a separate sample. In this crisis, that would have meant a fresh sample for every draft from audiences the company could not reach in the first place. Erasing memories removed cross-contamination and order effects, which let Artificial Societies compare 24 messages and creatives on equal terms across 328,576 responses. For a communications lead, the useful output is the change in each audience’s response against its baseline, not the response count.

How is crisis communication simulation different from a tabletop exercise?

A tabletop exercise tests the crisis team and its plan, while crisis communication simulation tests what the team plans to say to each audience. FEMA’s Homeland Security Exercise and Evaluation Program (HSEEP) doctrine, dated January 2020, defines a tabletop exercise as a discussion-based exercise in response to a scenario. Players apply their knowledge and skills to problems a facilitator presents. The outcomes the doctrine lists are recommended revisions to plans, policies and procedures, and an after-action report.

Most organisations in Ethisphere and Morrison Foerster’s 2023 Global Crisis Management Benchmarking Report already plan and drill: 88% of organisations surveyed had a written crisis management plan, and 79% of respondents said they ran crisis response drills on key risk areas. The report describes tabletop exercises as hands-on practice in walking through the incident response plan. A drill shows whether the team can decide and sign off under pressure, but it puts no draft in front of a regulator, a journalist or a customer. Run the tabletop to test who decides and how fast, and the simulation to test what they decide to say.

How long does a crisis communication simulation take?

Artificial Societies’ crisis communication simulation took under two weeks from start to finish, according to the crisis case study. That included building the five societies, taking the baseline and running the message tests with memory resets. The table sets the crisis beside the published durations of five other Artificial Societies engagements, one of which counts only fielding time on a society that already existed.

EngagementPublished durationWhat the duration covers
Consumer goods crisisUnder 2 weeksWhole project, including construction of five societies
Hyperscaler earnings call48 hoursStart to finish
Consumer goods product innovation48 hours of fielding across two rounds; under 24 hours per roundFielding on a society already built for the study
Teneo strategy messagingLess than 3 weeksWhole engagement, including society construction
Transport policy positioning3 weeksTime to deliver results
Post-merger integration6 weeksStart to finish

Source: Artificial Societies’ case studies linked in each row, read 9 November 2026. Teneo’s engagement took place in late 2025. The commercial due diligence study publishes no duration.

A society built before an incident takes construction time out of a crisis response. In the hyperscaler engagement, the four societies were standing rather than one-off, so they were ready for the following quarter, and the client had full access to Radiant, our platform, to re-run scenarios as the numbers firmed up. None of Artificial Societies’ seven published case studies, as of 9 November 2026, gives a turnaround for a crisis test on a standing society, so agree that deadline when you scope the work.

How far can you trust simulated reactions to a crisis statement?

Trust simulated reactions to compare your options, and check behaviour separately. Artificial Societies’ Survey Evaluation Report (January 2026) records 86% distribution accuracy, the overlap between simulated and human opinion distributions, across 1,000 surveys. The comparator is a 91% human self-replication ceiling, set by people answering the same question twice. The evaluation page sets out the method, and our article on AI persona accuracy explains it. Those figures measure survey opinion, not reputational recovery.

Two results from Artificial Societies’ benchmark of 1 October 2026 bear on crisis work, where the question is how each group reacts to a particular wording. Within demographic groups on the 2024 General Social Survey, the benchmark put Artificial Societies’ total variation distance from the real answers at 0.063, against 0.222 for the best of nine off-the-shelf large language models (LLMs). Total variation distance is the share of answers that would have to change to match. In the benchmark’s Albertson survey experiment, which asks one question in two wordings, Artificial Societies reproduced the shift in answers to within 1.4 percentage points, against 6.2 for the best LLM.

Corrections lose force over time, and the same holds in simulation. Fact-checks of false COVID-19 claims cut belief by 45% against a control group in Carey and colleagues’ 2022 human study, and by 46% in the synthetic British public of Artificial Societies’ replication of four UK misinformation experiments, published in September 2026. Two months later, that post reports, both effects had shrunk to 8% of their original size, the human figure by Artificial Societies’ calculation. The synthetic match relied on a recall rate taken from published research, since simulated respondents do not live through the weeks between surveys. Our article on prebunking and debunking covers what that means for timing a correction.

Simulation stops at behaviour. The purchasing measure in a crisis test records likely behaviour as the personas state it, so it ranks messages against each other and does not forecast sales. Establishing what customers actually buy after the statement takes sales data or human research.

How do you start a crisis communication simulation?

A first crisis engagement follows the shape of Artificial Societies’ crisis case study. Bring the response options you are choosing between, the audiences whose reactions decide the outcome and your publication deadline. We build a society for each audience from anonymised profiles and conversations gathered through social listening and deep research, then take each society’s baseline.

We then work with your team to test the messages, statements and activations, resetting memory around every exposure. Results come back as findings and recommendations. In the hyperscaler and transport-policy engagements, the client also had full access to Radiant to re-run analyses, and the crisis client was able to run follow-on simulations of delivery.

We do not publish pricing. Engagements are usage based and scoped with your team, with no rate card; to scope a first test, book a call.

Frequently asked questions

Which platforms run reputation scenario simulations that include stakeholder sentiment?

Artificial Societies runs reputation scenario simulations that measure sentiment audience by audience. In its crisis case study, it built one society per stakeholder group and surveyed each for baseline sentiment before any message was shown. The Teneo study reported approval scores, emotional sentiment and verbatim reactions for six narratives, and the transport-policy study added damaging news coverage and hypothetical scenarios to show where each narrative held up or collapsed.

What is an AI PR crisis simulation?

An AI PR crisis simulation tests draft crisis statements and activations on AI personas that stand in for the audiences a company must persuade, before anything is published. In Artificial Societies’ version, personas are grounded in real, anonymised profiles and conversations, and connected through mapped influence networks so that opinion forms among connected people. The output is each audience’s reaction to each draft, measured against a baseline taken before exposure.

Which crisis communications firms offer AI-powered message testing?

Firms that advise on communications have run AI-powered message testing with Artificial Societies. Its crisis case study was carried out with a leading communications firm for a major consumer goods company. Teneo used Artificial Societies to test six competing narratives for a major US company in late 2025. When you choose a firm, ask whether results come back by audience, with individual reactions you can inspect.

Can crisis simulation test reputation scenarios for a technology launch or under political scrutiny?

Yes. Artificial Societies’ transport-policy study built 1,500 personas of Washington, D.C. opinion leaders across more than 360 organisations and 800 job titles. It tested two narratives against damaging coverage, and 250,000 responses to more than 170 questions exposed a standing credibility gap. The Teneo study tested how a major US company should position a new technology strategy before launch. Our page on technology reputation testing covers scenarios for technology companies.

Can crisis communication simulation model journalists’ reactions?

Journalists can be modelled as their own society, and Artificial Societies’ crisis study included a Media society and an Influencers society. Its earnings-call case study included a Financial Media society of more than 1,600 personas representing business and finance journalists, and the simulated press coverage matched the real next-morning framing. Growth led the headline, the margin reset appeared as a cost story and financing emerged as the risk. That is one reported match, not a guarantee of future coverage.

Can we test confidential or market-sensitive statements?

Yes, because no draft reaches a real stakeholder. Artificial Societies’ crisis client tested highly sensitive strategic options without putting them to live participants. In the earnings-call engagement, selective-disclosure rules kept unreleased numbers and draft scripts away from real investors, so the team tested them on simulated audiences. In that engagement, no draft, number or script left Artificial Societies’ private, secured environment. Whether a particular disclosure is permitted remains a question for your counsel.

What does a communications team receive at the end of a crisis simulation?

A communications team receives findings and recommendations first: Artificial Societies’ crisis client decided its strategy from them, then was able to run follow-on simulations of delivery. The transport-policy client received a flagship advisory report with full access to Radiant, from segment cuts to individual persona reactions. Teneo received a written report and an interactive platform showing approval scores, emotional sentiment and verbatim reactions.

How many crisis messages should we test?

Test the options your board could realistically issue, and include the activations that would carry them, because Artificial Societies’ crisis study found a likely backlash against certain activations. Artificial Societies’ published engagements range from two narratives with their proof points in the transport-policy study, through six competing narratives for Teneo, to 24 messages and creatives in the crisis study.

Do we need to supply our own data to build the audiences?

Not necessarily. Artificial Societies built its crisis societies from real, anonymised profiles and conversations gathered through social listening and deep research. For its consumer goods product study, it gathered more than 100,000 anonymised profiles with social listening data partners. Its commercial due diligence study added client-supplied expert interview transcripts and a primary survey of 130 decision-makers, with one digital twin per respondent.

How much does crisis communication simulation cost?

Artificial Societies does not publish pricing, and its engagements are usage based. For comparison with human research, Artificial Societies’ consumer goods product study reported almost 80% cost savings with over twice the sample size of alternative human research; that was product research, not a crisis test. To get a figure for your own crisis test, book a call with the team.

Sources