Program

Audience Community of Practice: Generative AI

January 21, 2025

The Audience Community of Practice met with audience strategist Mayra Báez on Jan. 30 to discuss using generative AI to craft more engaging journalism. Below is a recap of the session:

Before diving in, it’s crucial to understand the frameworks and vocabulary, as they make all the difference when working with LLMs. Approach matters—what you say and how you say it has an impact. You get to define how AI works for you, not the other way around.

Why market research matters

We spend a significant amount of time and effort distributing our reporting, and that alone can feel like a beast to tackle. As a result, investing time and resources into market research often feels outside the realm of possibility—especially when many of our teams are understaffed and overworked. However, our audience is far more likely to engage with and benefit from the information we provide if it resonates with them—if it addresses their real needs, rather than what we think they need. That’s why we need audience-informed approaches that help us identify and address unmet needs and pain points. By understanding what our audiences truly need, we can deliver greater value—not just improve our reach or boost vanity metrics.

Understanding market research

Knowingly or unknowingly, we already use market research in our audience work. We conduct primary research when we gather insights directly from our audience—via focus groups, surveys, interviews, and other methods. Then there’s secondary research, which involves leveraging existing research that has already been done and analyzed. Secondary research is valuable for providing context—it offers basic demographics, trends, and industry details. Plus, it’s often free. For example, the Reuters Institute’s annual report on journalism, media, and technology is a great resource.

Primary vs. secondary research: Why primary methods matter

According to Mayra, primary methods are more valuable for audience work because they are designed to answer specific questions about our audience, our organizations, and our work. The downside? These methods are often expensive and time-consuming. But this is where AI can help. It gives us access to tools we didn’t have before, allowing us to iterate faster and refine our understanding of our target audience. In primary market research, we use both quantitative and qualitative methods.

Quantitative research is all about numbers. It answers the what—what our audience does as a whole. We can use surveys, data analytics, and other methods that drive much of our audience work. There are also more advanced techniques, such as eye tracking and biometrics, that provide deeper insights.

Qualitative research, on the other hand, is about stories. It answers the why. Why are people invested in a particular format, story, or angle? It helps us spot trends, identify gaps between what we planned and the actual experience of our audience, and dive deeper into their pain points. Examples of qualitative research include focus groups, interviews, usability testing for products, and social media listening (a form of digital ethnography).

Using specific tools vs. large language models (LLMs)

Tools like Swaggable, Atlas, and Dovetail provide highly technical, in-depth analysis, but they come with barriers—namely, cost and the need for technical expertise. With AI, however, you can simply communicate in natural language, and it will walk you through the process. This makes AI much more accessible. In Mayra’s experience, it offers more than enough valuable insights to make informed decisions about audience work.

How to build an audience research process with LLMs

1. Identify your eesearch goals and methodology

First, ask yourself: What does my audience need to know, understand, feel and do? This is a brainstorming exercise to help you think about the audience you want to attract—not just the one you have right now. Next, decide whether you need quantitative research (the what) or qualitative research (the why).


Pro Tip: If you’re not familiar with different research types, ask the AI about them to explore which might suit your needs best.

2. Gather your data

Once you’ve chosen your method, it’s time to gather data. This is where the hard work begins, but LLMs will later play a crucial role in processing and analyzing the data you collect.

3. Use AI to analyze your data

Once your data is gathered, you can use AI to analyze it. You can ask the LLM to mimic the functionality of a specific software tool (e.g., Atlas, Dovetail) and have it analyze the data based on your prompt. This can range from simple prompts to more complex ones, depending on your familiarity with data analysis tools. You could also ask the data to analyze the data as though it is an expert with a certain number of years of experience and include parameters around the tone of the response. For instance, you may want the AI to respond as though it is addressing a group of experts or academics. Alternatively, you may want the response in layman terms.

Pro Tip: If you’re working with sensitive data in the cloud, ensure that it’s anonymized. LLMs can potentially expose private information. Alternatively, you can use tools like Anything LLM to run the model locally on your computer, ensuring full control over your data.

4. Validate

Once the AI processes the data, it may provide preliminary results that might seem unpolished. It’s important to validate the information. Review it yourself, or involve your team members to ensure the insights align with your actual findings.

5. Organize your data

Use your preferred content management tool (Word, Excel, etc.) to organize your data. Then, you can ask AI to generate audience personas based on the insights gathered. For example: 
“Use this information to create a buyer persona for a Gen X individual between 35-45 years old who lives in an urban area.”

*An audience persona is a semi-fictional representation of your ideal user, built on data and insights. If you’re working on a product, instead of creating buyer personas, you can ask the AI to help you build journey maps for your products.

6. Continuously improve

This is an iterative process. You can continue gathering feedback and analyzing it with AI to refine strategies and implement changes on an ongoing basis.

Best practices to keep in mind

• Respect privacy and ethics: Always be transparent about how you’re using the data and ensure you follow your organization’s privacy guidelines.
• Check for bias: Remember that how you frame your prompts will shape the AI’s response. Be aware of your own biases when interpreting results.
• Validate your findings: Don’t rely solely on the AI’s output—always validate with real-world data and insights from your team.
• Clean data is key: Ensure that your data is well-organized and understandable to the AI. The cleaner the data, the better the results.

Using large language models in market research can be a powerful tool to accelerate your audience insights and streamline the process. By leveraging AI, you can go beyond traditional methods, refine your research approach, and make more informed decisions—without the steep learning curves and resource demands of traditional market research tools.

Relevant links and individuals:

• Posts from Khalil A. Cassimally on The Audiencers
• https://www.swayable.com/
• https://dovetail.com/
• https://atlasti.com/
• https://jan.ai (used by one member to run a local AI on their computer)

• https://anythingllm.com/ (free too that helps run the LLM of your preference on your computer instead of cloud-based serves, processed locally in your computer, good for anonymizing information gathering)