Communicators working on narrative change around some of the defining issues of our time—building solidarity across difference, advancing equity, strengthening democracy—rarely have the time or budget for in-depth audience research behind every decision. They are still in need of a compass: a systematic way to tell what narrative approaches will work across and within audiences before committing to full production.

In our previous post, Audience Research: Increasing Utility, Democratizing Access, we wrote about how we’re building the Audience Lab, a new platform for AI-powered audience intelligence tools, on top of a decade of audience research housed inside the Narrative Observatory’s data infrastructure. Since then, we’ve prototyped a focus group tool alongside four partner organizations: Caring Across Generations, Narrative Initiative, Race Forward, and RALLY. Their insights have helped shape an MVP (Minimum Viable Product) for a tool that helps users explore the responses of real people by interacting with personas that are AI representations of our four Narrative Observatory audiences.

Each values-based audience persona’s responses are anchored in data from 13,163 real U.S. respondents in our Deep Story Survey System. The tool is built for communications professionals on tight timelines and budgets who want to develop and test hypotheses about reaching different audiences.

Putting Research To Work

Narrative strategy is essential for setting a direction. What this tool addresses is the next step: once you have a strategy in place, how do you iterate quickly and effectively on ideas, whether you’re seeking out audience-specific guidance, messaging direction, or content recommendations?

As part of the Robert Wood Johnson Foundation’s Mindsets Consortium, over the last three years, we’ve engaged with other narrative change organizations and advocacy groups in thoughtful discussion around what might make their narrative work more effective and sustainable, and built a rich research foundation:

• defining a shared narrative goal for systemic equality through a participatory process;
• analyzing racial equality in scripted media;
• examining how individualism shapes behaviors and systems;
• understanding how multiracial solidarity can be created and sustained;
• breaking down attitudes toward equality, agency, flourishing, and an equal future across values-based audiences, and more.

Now the opportunity is to put that research into practice.

The organizations we’ve been in conversation with tell us that only a small share of their creative decisions are data-informed, and that even an incremental improvement in using data to engage different audiences would be a “huge win.” We heard from several strategists across teams that the alternative to our tool isn’t rigorous research—it’s educated guesses while being on a deadline. That is the gap we set out to fill: making research accessible without sacrificing speed or accuracy.

How This Works

You can choose between an in-depth conversation with a single AI persona—best for understanding how one kind of audience thinks—and a conversation with multiple AI personas, where you can collect and compare different answers to a single question.

When you ask a question, we search the survey data for what’s actually relevant to that question—closed-ended survey stats, themes coded from thousands of open-ended responses, and verbatim quotes from actual survey respondents—and hand the persona only the best matches to work from.

Each persona is built from the real survey respondents who most closely match its values profile. The quotes a persona cites come verbatim from survey respondents’ actual answers.

Building In The Open

Our partner organizations were co-creators in building and testing the tool, with each sitting at a different point in the narrative and communications journey. They brought a wide range of target audiences and topic interests, from reaching faith communities and Latino parents to examining attitudes around climate migration, literacy, and Medicaid.

Between June and October, we worked through intake questionnaires, discovery sessions, a build phase, user testing, and retrospectives with each partner. Their feedback has been instrumental in understanding what it takes to build something that meets the wide-ranging needs of narrative change practitioners, while separating AI hype from AI-powered research tools that can offer a real edge.

• 15 practitioners across 4 organizations
• 8 moderated 1:1 sessions run on participants’ own projects
• 682 persona replies generated over 3 weeks of independent use
• 50+ minutes spent in the tool across an average of 6 sessions on 3 different days

Key Learnings

Transparency about sources is key to building trust

The partner organizations who tested the tool all have a relationship with Harmony Labs, from long-time research partner to agency peer, and have previously used our values-based audience framework and research to support their narrative strategy work. Familiarity with the Narrative Observatory and our audience model bred excitement for how this tool would unlock even more of our insights. While users initially came in skeptical that the AI powering the tool would reproduce bias or inaccuracies, knowing that there is credible data behind the personas drove their desire to return. Seeing each persona’s responses cite actual verbatims and survey data helped make the tool a trusted source.

“The characters feel like people I have been in rooms with.”

Conversing with a persona bridges the gap between understanding and application

Users typically started out wanting to understand an audience perspective on a topic, and those curiosities gradually turned into concrete asks, such as testing a message, gut-checking a framing, or assessing whether a program idea is a likely fit. Despite the tool being in the MVP stage, several users were eager to incorporate the learnings into their day-to-day work. The desire to take immediate action suggests the responses they were getting from the personas were useful. Observing how partners used the tool points to new potential features that are custom-built for specific decision-making around communications activations, such as a new call to action or a shift toward a different platform or video format.

“It’s sparked my curiosity. And it makes me want to mess with it more.”

AI personas are one piece of the greater audience puzzle

In-depth audience research can be expensive, and for many organizations the choice is between a tool like ours and no audience research at all. Users from our partner organizations saw the tool as viable enough to fill the gap and said they’d use it again. Where we’d be wary is using the tool as a substitute for fielded research, instead of as a directional research aid to spark new ideas and approaches. These outputs are AI-generated and reflect survey data from a defined time period, and shouldn’t be treated as statistically representative findings or relied on as the sole basis for a campaign decision. What it does do is democratize access to existing research and help organizations inhabit the audiences more deeply by giving voice to the responses of real people.

“Because the content of these personas are rooted in a large amount of survey data, I feel a bit more comfortable using it as ONE data point in a research project, not THE datapoint.”

Looking Ahead

This MVP set out to test whether practitioners would use AI personas grounded in our research in their own work, and they did—so we’re building on it.

We’re making the personas more distinct by bringing in more of our research data, while making the tool available to more organizations who wish to reach audiences in new ways, including the wider media-making ecosystem.

The defining ask from users was the ability to customize personas to the demographics of their own target audiences. Besides rebalancing the mix of personas, we are also adding new functionalities, such as the ability to make smaller panels, choose which persona responds, and compare how different values-based audiences respond to the same question—something that isn’t easily achieved with in-person focus groups.

A “focus group tool” was our first experiment in making our data more directly accessible to partners, but it won’t be the only one. We’re testing other ways to deliver access to the same data, so future versions of Audience Lab may look quite different.

…

If you’d like to start a conversation and learn more about the Audience Lab, please get in touch. As always, we’re interested in your thoughts and feedback on our work.