Synthetic Focus Groups: Do They Work in Market Research?

Synthetic focus groups simulate real human interactions using AI-generated respondents. You basically collect data about your customers, feed it into a synthetic market research tool, and it creates synthetic users that respond based on that information. Besides textual chats, you can also use them to run surveys, interviews, and user testing studies.

But what’s the point of synthetic respondents? One might argue that you’re just getting answers you could’ve guessed anyway. So what’s so novel about this approach if you’re not really discovering any new information?

Sure, it's cheaper than traditional research methods, which can cost a lot and take time, what with recruiting participants and bringing a credible market researcher on board. Plus, you get simulations that never get tired, are available 24/7, and cost a fraction of the money. You can keep asking questions and tweak your study as needed without worrying about fatigue or conflicting schedules.

Now, there are plenty of other pros and cons to synthetic focus groups, and in this post, we’ll cover most of them, along with how they work and some real-world use cases.

All About Synthetic Users & Synthetic Market Research

Synthetic market research leverages AI models and machine learning systems to generate realistic “synthetic users” based on actual audience demographics, psychographics, and behavioral inputs. These synthetic personas behave like virtual customers and respond in human-like ways, letting marketers run research studies without relying entirely on real participants.

Now, are they accurate? That’s debatable.

For example, a study from Google DeepMind and Stanford found that GPT-based “synthetic people” matched real participants’ survey answers about 85% of the time, showing AI can closely mimic real users when trained on good-quality human data.

In the study, researchers first conducted two-hour qualitative interviews to gather comprehensive life histories from a diverse group of 1,000 people. These interviews were then fed into GPT-4o to create synthetic personas, or “agents.” Next, both the real participants and their corresponding AI agents were asked to complete a set of established social science tests, such as the General Social Survey (GSS) and the Big Five Personality Inventory.

The results showed that the agents replicated participants’ GSS responses with 85% accuracy.

A similar pattern showed up in a real-world newsroom use case. In a Digiday Podcast episode, Tracy Yaverbaun, GM of The Times and Sunday Times, explained how her team has been experimenting with the synthetic research technology. To test it, they built an AI-generated audience panel with their database of 642,000 subscribers.

The Times then used those synthetic respondents to help finalize a name for a new business podcast. And just like in the DeepMind study, they compared the synthetic answers against real users. The synthetic panel’s responses were about 92% accurate, only one percentage point below the traditional research benchmark of 93%.

But here’s the catch: not all of the responses made sense.

When asked what new experiences they’d like from the Times, the synthetic users were all in on summaries, AI explainers, and more automation. The actual loyal customer base, though, was far less excited about introducing more AI into the product.

So it’s a bit touch-and-go, depending on how you look at it. Synthetic user technology is still in its infancy. For now, synthetic respondents offer a space for experimentation, a sandbox to play around in, rather than a replacement for real-world qualitative research.

What are Synthetic Focus Groups (SFGs)? Pros, Cons, & More

As mentioned before, synthetic focus groups (SFGs) represent AI personas or simulated users that represent different types of users. These personas talk to each other, react to what others say, and discuss ideas together, much like participants in a real focus group. Instead of analyzing only existing data (as in traditional AI analytics or predictive modeling), SFGs actively generate new qualitative insights through conversation.

The reason they’re gaining traction these days is simple – traditional focus groups aren’t really doable for small businesses and startups.

It’s the same case for big brands. They’re expensive, time-consuming, and logistically complex. You have to recruit participants, schedule sessions, design a study plan, and either moderate the discussion yourself or hire a professional moderator, which makes it even more expensive.

Conversely, synthetic focus groups allow you to test multiple scenarios, messages, and product concepts quickly and affordably. Because the participants are AI personas, there’s no bandwagon effect, no single “loud voice” dominating the conversation, and far less social pressure that can distort feedback. They also eliminate moderator bias, since the AI moderator follows a predefined discussion structure.

How SFGs work in practice

Most synthetic research tools work in a pretty similar way when it comes to focus groups. You start by picking your panel and sharing your research goal along with a discussion guide. From there, an AI moderator runs the session for you, introducing topics, asking follow-up questions, and keeping the conversation on track.

One of the big advantages is that you can run dozens of focus groups at the same time, each one tailored to a different customer segment or persona. The results come back instantly, complete with full transcripts and easy-to-digest insights, like themes, sentiments, points of agreement or disagreement, and quotes.

The best part? You can experiment freely. No idea is too risky or too bold to try. For instance:

• Marketing teams can run risky campaign ideas through a synthetic focus group first to see how different users react, which headlines resonate, and what messaging falls flat.
• At the same time, a UI/UX team might prototype a new checkout flow and use synthetic users to walk through it, finding areas where people get confused or drop off.
• Meanwhile, a product team could test three different pricing models — subscription, pay-as-you-go, and freemium — by running parallel focus groups and seeing which option feels most appealing to each segment.

Since everything happens fast and without real-world risk, teams can try bigger, bolder ideas and improve them before committing real time, money, or customers.

SFGs can also be used to fill knowledge gaps and forecast demand. In retail, for example, they can help refine new products and promotions to capture new market segments. They can also improve ROI for every dollar invested and drive profitable growth through higher revenue, stronger margins, and faster inventory turnover.

Pros and cons of synthetic focus groups

We’re clearly in an era where it feels like you don’t even need people to conduct market research anymore. But is this the next big thing in MarTech or just another AI gimmick?

The main criticism here is that AI models depend on historical data, so they’re basically remixing what’s already known instead of producing something truly new. But that’s not necessarily a dealbreaker if your goal is to understand how your existing customer segments might react.

There’s also the fact that AI doesn’t feel. It’s not human. It doesn’t get annoyed, tired, or biased in the same way people do, which can actually be a plus because those emotions often influence real interview responses. On the flip side, it’s hard to fully trust the feedback synthetic users generate. After a certain point, the data can get diluted and may stop being a true representation of real human experience.

Remember the whole “AI content ouroboros” thing?

So why should you use synthetic focus groups? For starters, a lot of businesses simply don’t have customers who are excited about answering surveys or joining focus groups. In those cases, the synthetic approach can be a practical alternative.

You don’t have to spend money recruiting or managing human participants. AI participants can operate 24/7, which makes your research insanely scalable. You also get the flexibility to create synthetic users from virtually any demographic, psychographic, or behavioral segment. And it’s fast. Most importantly, because synthetic customers are based on generalized customer data, they don’t come with the same compliance, privacy, or consent risks as working with real customer data.

So, How Do You Create Synthetic Respondents?

Human behavior is complex — but sometimes, not really. With the right customer data, you can predict a lot of things. And synthetic customers are a way to do this. Since they’re virtual avatars built from your existing customer data, they can be used to simulate real customer behaviors and preferences.

The simple way to generate one would be through LLMs like ChatGPT. Just describe your target audience, add your product information, and prompt it to adopt the “persona” of that audience.

You can, for all intents and purposes, use these “AI personas” to test product concepts, marketing campaigns, and business strategies. However, be prepared for generic outputs and, at times, stereotyped representations of certain demographic groups.

A better way is to use professional synthetic market research tools available online, like Delve AI’s Synthetic Research Software, built specifically for this purpose.

What sets it apart from generic generative AI tools is how those synthetic users are built: they’re derived from personas grounded in real customer data, including CRM records, web analytics, social audience intelligence, and 40+ public data sources. That includes Voice of the Customer signals pulled from reviews, ratings, community forums, blogs, news channels, and more, so your synthetic respondents are anchored in how real people actually think, talk, and behave, not just how an AI model guesses they might.

Here’s how it works. You generate personas > create synthetic research panels > run surveys and interviews.

Step 1. Generate user personas

Delve AI’s online persona generator currently offers six types of personas, including Customer Persona and Research Persona. In this case, we’ll use the Research Persona tool.

After signing up or logging in, go to Research Persona and upload your market research documents. This can include interview transcripts, past user profiles, surveys, or industry news, basically anything relevant to your audience. No research material? No problem. Just add a solid description of your target audience, click Create Personas, and Delve AI will develop personas based on the data you’ve provided.

The platform uses AI and machine learning to create unique audience profiles. Depending on your database and use case, you’ll get one or multiple personas. Each segment includes persona details, audience distribution, influencers, and journey maps.

Click Persona Details to see user demographics, buying triggers, psychological drivers (like goals, motivations, needs), lifestyle, career status, and core challenges. You’ll also find insights into their communication preferences, social networks, brands, shopping sites, YouTube channels, podcasts, subreddits, and more.

Each of your personas is refreshed monthly with new data, and you can continuously improve their accuracy by adding new research materials.

Step 2: Synthetic user panels

Now that the personas are in place, the next step is to generate synthetic users. To begin, go to the Synthetic Research dashboard and purchase the number of users you need. Next, create a panel with this group and give it a name, like Marketing Insights Group.

You’ll then be taken to a screen where you can choose which persona or persona product to use for generating synthetic users. For this example, select a segment from Research Persona named Edward Collins.

Once selected, the software generates synthetic users based on that persona segment. Each user comes with a Start Chat option, allowing you to interact with them directly (see Digital Twin for more details).

Step 3: Surveys and interviews

Once the synthetic users are ready, go to Surveys & Interviews and select Create Study on the dashboard. You’ll be prompted to:

• Choose your synthetic user panel
• Enter your survey or interview name (e.g., Product Market Fit Survey)
• Select the number of users (e.g., 100)
• Upload a CSV file containing your questions (eg, multiple-choice, rating scale, open-ended, and ranking questions)

The platform runs your survey across the panel and automatically turns the results into charts, themes, and topics.

Click the Data button to see how many people responded to each question and who they were. You can also check out the complete survey transcripts and sort them by question or by respondent.

Want to go beyond surface-level answers? Use the chat feature to follow up and ask users why they made certain choices. You can also explore other marketing or product questions. For example, in the sample below, we asked a user what influenced their decision to complete a purchase online.

Because everything is structured more rigorously and tied back to real datasets, our platform can run more controlled simulations than a generic chatbot prompt ever could. You’ll get fewer stereotypes, more consistency, and results that are much closer to something you can actually act on.

Getting Started With Synthetic Market Research

Now, you may be thinking: Will synthetic market research do away with traditional, human-based research studies?

Right now, that’s highly unlikely.

Synthetic users are great for early-stage testing and predictive modeling, but emotions and subconscious decision-making aren’t something they’ve mastered yet. Because everything is text-based, you can’t really pick up on tone, body language, or the subtle expressions that usually accompany how people talk or react.

AI respondents can also miss cultural and social context, like social trends or slang, since they’re trained on past data. So, while synthetic users are incredibly useful, they’re best seen as a complement to human research, not a replacement.

There are also several things to keep in mind before and during your use of synthetic users:

• Don’t cut corners on data. The quality of synthetic respondents depends entirely on the data used to build them. They’re only as good as the inputs you provide.
• Watch out for bias. AI can introduce bias into your study. For example, it often has a tendency to be overly positive and focus on the best possible outcome.
• Test multiple tools. Try different synthetic market research platforms before committing to one. Ask yourself: Are the personas realistic? Does the tool support the use cases I want to test?
• Start small. Begin with low-risk, simple tasks that don’t have a major impact on your business. This helps you evaluate credibility and accuracy without putting too much at stake.
• Validate with human research. Test the results against a traditional test group made up of real human participants to check for accuracy and reliability.

AI is reshaping a fast-moving, competitive market where companies are under pressure to launch better products, faster. While product development is time-consuming, it can’t be skipped if you want to build something truly valuable and user-centered. And synthetic users, used alongside actual customer feedback, can help teams bring personalized products to market, without replacing the need for human insight.

Final Thoughts

For all its follies, artificial intelligence is only going to get better and more sophisticated in the future; after all, just look at how far we’ve come since 2022. In perhaps a year or two, synthetic users will be free of most biases and will improve decision-making.

They can already speed up the pace of insight. With traditional market research, by the time results arrive, the market may have already moved on. Synthetic users dramatically reduce the time to insight, allowing teams to test new concepts, ideas, and incentives 24/7. They help move hypotheses into action without long wait times.

Synthetic respondents also go beyond analyzing past data and can tell you what customers are likely to do in the future. They can show how different customer groups think and behave at scale.

That said, synthetic focus groups cannot replace human participants. They do not operate with empathy or think with emotions, at least not yet. So, while they can prove to be useful, they should be used to complement, not replace, human inputs.