Faster research without sacrificing quality

Hello everyone! We’re excited to share a big update to our research panel offerings that will genuinely change how many of you approach your research projects: we're bringing synthetic responses to the self-service online panels experience in Q4 2025 (Public Preview).

If you've ever been in a situation where you needed quick feedback on a concept, wanted to explore a hypothesis, or even make the business case to invest in research—but couldn't justify the time or cost of traditional panel recruitment—this is for you. We've built a solution that generates row-level survey responses using a fine-tuned language model that is fundamentally different from general-purpose LLMs (think Google Gemini, ChatGPT).

Our model has been specifically trained to understand the concept of a survey and how humans respond to surveys. It's built on a vast, structured, proprietary dataset of human survey responses, leveraging the research expertise and market research knowledge we've developed over 25+ years in the business. The result is a research tool that understands survey context, question types, and response patterns in ways general-purpose AI simply can’t.

Here’s a quick overview of what you can do with synthetic responses:

• Get fast responses: When you set up a survey using synthetic panels, you don't wait for recruitment or fielding. The system generates responses and sends them straight to your Data & Analysis tab, just like human respondents would. Early customer pilots like Gabb Wireless have seen fielding timelines drop from 7 days to 4 hours—a 98% reduction in time-to-insight.

• Validated against real human data: We've put our synthetic model through a four-step validation framework testing for generalization, data shape, diversity, and transferability—and we publish our methodology transparently. The model is also continuously trained on fresh panel data to reflect current opinions and attitudes. In a recent comparison study, our fine-tuned model outperformed general-purpose LLMs by 10–12x in matching human response patterns. Early customers running head-to-head comparisons are also seeing strong alignment, especially on attitudinal and preference-based questions.

• Mix and match methodologies: The best part is you’re not locked into one approach. You can run a fully synthetic study to test five concepts quickly, then follow up with human panels on the top two. Or blend both for comprehensive insights. Another early customer Loop Earplugs tested message resonance with both audiences and found alignment on two of the top three messages, giving them confidence to use synthetic for early-stage testing, then validate final decisions with human respondents.

• Optimize your panel budget: Instead of paying per-project, synthetic panels use a subscription credit model through Edge Audiences. You purchase credits upfront, then spend them on either synthetic or human panel studies as needed. Early access customers are seeing around 50% cost reductions compared to traditional human-only panels.

Learnings from early customers

Synthetic panels are not a full replacement for all human research. When you have high-stakes decisions, regulatory requirements, or need to understand nuanced human emotions and behaviors, human panels are still your best bet.

Through our pilots with customers, we’ve found synthetic panels work best for:

• Early-stage exploration and concept testing
• Situations where you need quick directional feedback
• Research questions focused on attitudes, preferences, and stated behaviors
• Projects where speed and cost are primary constraints
• Expanding sample sizes for harder-to-reach segments

They're less ideal for:

• High-stakes decisions requiring human validation
• Deep qualitative insights about experiences
• Research requiring regulatory approval
• Situations where stakeholders specifically need human respondent data

The sweet spot is often a blended methodology: move faster on early-stage testing, then validate high-stakes decisions with human panels. We believe synthetic data, generated using fine-tuned LLMs, provides another tool for researchers to utilize in their practice—not a replacement for what already works.

What's coming next

Available for U.S. General Population (for now)
Our Public Preview launch includes U.S. General Population synthetic respondents, and we'll be adding more audiences based on available data, market demand, and feedback from our customers.

Beyond the Public Preview launch, we're working on:

• Additional targeting criteria (education, relationship status, children)
• More audience segments beyond U.S. General Population
• Enhanced integration features
• Better visibility into credit usage at the brand level

We'd love your input

We're still learning about how researchers want to use synthetic panels in their workflows. A few questions for the Community:

• What types of research projects would you want to run with synthetic responses?
• What concerns or questions do you have about using AI-generated data?
• For those who've tried synthetic panels (any vendor), what's been your experience?
• What audiences or segments would you most want to access through synthetic panels?

Feel free to share in the comments. Your feedback directly shapes what we build next!

Want to try Edge Audiences? Reach out to your account team or check the customer hub for access information