How Public Affairs Teams Can Pressure-Test Messages Faster
Messaging in public affairs and corporate communications largely comes down to one thing: people read the same words differently. A policy frame that lands with an industry expert can confuse a voter who just wants to know what the fight is about. An internal town hall script that sounds right at headquarters can read as out of touch to the people in the field who actually do the job.
The real problem is figuring out how an audience will react before the message goes out. Polling tells you what people think, not why a specific phrase loses them. Focus groups get you real language and real logic, but they take weeks and cost more than some budgets can justify. Gut checks are only as good as the room they happen in.
And none of that research holds still once it’s done. A new development is outlined, a news cycle shifts perception, and research that was solid a few months ago is suddenly being asked questions it was never built to answer. Most clients can’t justify commissioning something new every time the landscape shifts slightly.
There has always been a gap between a draft and testing it with real people. Most public affairs work just lives inside it.
Synthetic audiences are how that research keeps earning its keep
Synthetic audiences fill that gap. A synthetic audience is an AI-built group of personas designed to show how different segments are likely to react to new messages.
In short, we use AI to take a client’s existing audience research, verbatim transcripts, and fielded findings and build a set of personas grounded in what real people actually said. Once they’re built, we can keep testing new messages against those personas without waiting weeks for a new round of research every time.
However, synthetic audiences are only as good as the research behind them. Personas built on real transcripts behave differently than personas built on assumption or summary. That’s the whole reason this only works if the underlying research was done well in the first place, and why we won’t build a synthetic audience without real data behind it.
What this looks like in practice
A recent engagement is a good example of how granular this gets. The foundational research was a series of focus groups across multiple locations, covering both tenured and newer audience members with different roles and reporting lines. From that research, we built five distinct personas. Tenure, location, and background all shifted how the same message landed.
Every trait in every persona ties back to a specific transcript quote, and we flag the points where the data is thinner, so the team and the client know where to push back. A single persona built on assumptions would have missed those splits entirely.
The validation step most projects skip
Before personas become synthetic audiences, real people who know the audience have to validate them. For that engagement, we built a short review document with four direct questions:
· Does this feel like someone you recognize?
· Are there people who don’t fit any of these personas?
· Where we flagged lower confidence in the data, do you push back?
· Do the directional messaging guides match how these audiences actually respond?
That validation step is what keeps the synthetic audiences honest at the build stage. The next test is the one that actually closes the loop: when the messaging goes into market and the real reaction either confirms the model or corrects it. We treat that as part of the work, not the end of it.
What the synthetic audiences actually do
A draft message goes in. What comes back is how that segment is likely to react: where it persuades, where it triggers skepticism, where the language quietly works against the goal.
That also means the personas don’t go stale once the research wraps. If a new political development changes the conversation, we can quickly run a message about it past the same validated personas and see how each segment is likely to take it, without commissioning a new poll or another round of focus groups to find out. The personas stay grounded in the real people they were built on. What’s new is the question being put to them, not the people answering it.
Where this fits in the research practice—And why real people still matter
Synthetic audiences extend the life of research. They don’t replace it. The personas are only as useful as the data behind them, and they need a real refresh, new fielded research, once people’s underlying attitudes have actually moved. Judgment about which messages matter and which fights are worth picking still comes from people, not from the model. Polling and focus groups still matter because they capture what real people actually believe, not just how a model expects them to respond.
The difference is how often a client can pressure-test their language between major research investments. The same research just has a longer impact.
Our approach to synthetic audiences
Our approach starts with real audience research, not assumptions. We use focus group transcripts, polling findings, and other validated research inputs to understand how audiences actually talk, where they are skeptical, and what language moves them.
From there, we build synthetic personas that reflect those real audience patterns and use them to pressure-test new messages as the issue environment changes. Every output still requires human judgment: what to test, what to trust, where the data is thin, and when it is time to go back to real people.
The goal is not to replace research, but to help clients get more strategic value out of the research they already have.
At THG, we pair two decades of public affairs experience with AI-enabled audience research tools to help clients pressure-test language, sharpen strategy and reach the audiences that matter.
