Using Synthetic Panels To Find Quality Customer Insights

Extreme competition and the need for rapid customer insights have driven market researchers toward AI-assisted tools like synthetic panels. These “look-alike” audiences or virtual users allow one to get audience feedback without recruiting real customers.

But are synthetic respondents truly a match for human responses or just mediocre, make-do replacements?

Marketing and product development cannot rely on guesswork alone. You need to talk to your users to know what they need, want, and how they make buying decisions. Yet, not all companies have the time or the capital to employ traditional panels.

In this post, you’ll find out how synthetic panels address this gap in market research and provide a safe environment to test risky ideas.

The Advent of Synthetic Data

According to the European Data Protection Supervisor, “synthetic data is artificial data… generated from original data and a model that is trained to reproduce the characteristics and structure of the original data… synthetic data and original data should deliver very similar results when undergoing the same statistical analysis.”

So, it’s data that’s generated using artificial intelligence and machine learning models trained on real human data. Synthetic data is designed to closely mimic real-world data, making it suitable for a variety of use cases while preserving the patterns present in the original dataset.

In market research, this training data can encompass all sorts of first-, second-, and third-party information, like customer survey responses, social audience insights, industry reports, and more.

But if the result is just a repetition of all that we already know, then why do we need it in the first place? Well, NVIDIA lists out the top reasons quite nicely: data scarcity, data privacy, data quality, and testing.

Basically, synthetic data:

• fills gaps that exist in current market research.
• overcomes general data protection laws (since it only “mimics” and doesn’t use individual data).
• does away with most of the biases associated with real datasets and LLMs.
• offers a sandbox environment to test products, features, and messaging before launch.

Another good point is that synthetic data is cheap; hence why it’s all the new rage in marketing AI.

Practical applications of synthetic data in customer research

Synthetic data has uses in industries unrelated to marketing, like medical research and automotive manufacturing. For instance, many self-driving car companies use simulated driving scenarios to help AI models learn how to respond to different road situations (say, pedestrians, roadblocks).

However, in marketing, the applications are currently limited to personas (user, customer, or buyer) and synthetic panels.

To begin with, you can use synthetic customer data to create AI personas, which can then be used to understand consumer behavior and verify hypotheses. This is especially useful if you want to get insights into niche or hard-to-reach markets.

The more popular alternatives are synthetic panels. With synthetic datasets, you can create not one but multiple “synthetic respondents,” virtual participants that act as a stand-in for your original human respondents.

What Are Synthetic Panels And Why Do They Matter?

Synthetic panels are built using generative AI models – often trained on consumer data – and consist of synthetic respondents that can participate in surveys, interviews, and other types of research studies. This data can be public or private, including user demographics, buying behavior, purchase preferences, and more.

To give you an idea of what they look like, here’s an example of a synthetic panel created for a grocery planner app using Delve AI’s Synthetic Market Research Software.

The brands on board are currently using synthetic panels to run studies to test and reiterate a wide range of concepts involving product innovation projects and marketing campaigns. The result is usually qualitative and quantitative in nature, containing graphs, themes, response transcripts, and an in-depth research report, as shown below.

[Delve AI synthetic panel survey results – graphs]

[Delve AI synthetic panel survey results – themes]

[Delve AI synthetic panel survey results – synthetic user response transcripts]

[Delve AI synthetic panel survey results – detailed study report including key insights, themes, sentiments, takeaways, follow-up items, and more.]

The average cost of using these panels is around $2 per synthetic user (with Delve AI, it’s $0.99 per user). Compare that with traditional research methods, where human participant incentives alone can go from:

• $50 to $150 per person for online surveys
• $400 to $600 per person for in-depth interviews
• $15,000 to $30,000 per session for focus groups

And suddenly, synthetic panels start making a lot of sense from a cost perspective. These are just the numbers for more generic participants; the price point is much higher for specialized audiences. Beyond finances, the time required to screen participants, frame questions, and cross-check answers is also a barrier.

Despite all this, people might not really be saying what they’re feeling most of the time.

Synthetic panels help you find the difference between what consumers say and what they actually do. They can be a quick, cost-effective solution for early-stage research and exploratory studies.

How do synthetic panels work?

The process of creating panels and generating synthetic responses is simple. We’ll discuss it in more detail in this section.

It all starts with data collection. You gather the audience data you need from existing customer profiles, survey responses, customer support transcripts, sales records, web analytics, and more, then feed it into your selected synthetic panel provider, such as Delve AI.

Of course, the type of software you pick depends on your research goals and data. As per sources online, there are three types of synthetic research models available in the market today:

• LLM wrappers built around larger LLMs, like ChatGPT.
• Machine learning-powered models that can learn from past research to generate unique responses.
• Foundational large language models that leverage a combination of human and publicly available data.

Because their responses are derived from publicly available information, LLM wrappers lack the distinctiveness of typical human responses. ML models also have, albeit in a different way, a limitation: their output depends entirely on the quality of the input data used for training.

Foundational large language models appear to address these shortcomings by leveraging both internal and external data sources. Where one source may fall short, the other helps fill the gap, making the output often more nuanced and contextually informed.

They are also way more effective in dealing with highly specific market segments.

A platform like Delve AI, for instance, takes the latter approach. It combines your existing customer data with 40+ public data sources, including Voice of Customer data from ratings, reviews, community forums, blogs, and more, to first create data-driven personas using AI and machine learning technologies.

[Delve AI research persona segments]

These AI-generated personas help you identify audience segments within your customer base and provide details on consumer preferences, goals, motivations, buying barriers, pain points, items of interest, and more. You can then use one, two, or a combination of these persona segments to generate hundreds of synthetic respondents.

[Delve AI synthetic panels dashboard]

Filters also let you recreate highly targeted research panels by narrowing audiences by attributes such as gender, industry, or location, and by setting the desired distribution.

[Delve AI synthetic panels dashboard – Filters]

Most synthetic research platforms don’t work this way. You only get to describe your target audience – sometimes with supporting documents – and the system generates synthetic customers based on those data points, representing different types of users.

Efficient when answering questions that don’t require much research, but not very useful in cases where one needs dependable qualitative insights. Furthermore, this type of synthetic panel data is typically generated as a one-time output, whereas marketing is an ongoing process. On platforms like Delve AI, you get synthetic panels that are continuously updated using real-world feedback.

So if you want your synthetic respondents to be on point, ask your synthetic panel provider a few important questions before getting started:

• What model does it run on? And how often, if at all, are the synthetic users updated?
• Are they capable of producing unique responses rather than generic outputs?

Understanding the model’s limitations and potential biases is very important. It’s also a good practice to validate responses from synthetic audiences against those from real human customers whenever possible. Doing so will help correct inaccuracies and ensure your insights are grounded in real-world behavior.

Algorithm limitations, bias amplification, and reliability

Here's the thing: we need to be realistic about what synthetic panels can and cannot do. You need to know the capabilities and limitations of the technology that you’re eager to use. Like any AI system, their outputs are only as reliable as the data and models used to build them.

Treating synthetic responses as the unquestionable truth can and will lead to poor decisions and bad outcomes.

Algorithm limitations.

Synthetic panels rely on algorithms that learn patterns from existing data. This means they don’t truly understand human behavior; they simply reproduce patterns they’ve seen before.

Because of this, they often struggle with subtle human nuances, emotional complexity, and highly contextual situations. If a perspective isn’t well represented in the training data, or it’s outdated or skewed, the model may fail to capture it.

Bias amplification.

AI models are trained on human-generated data, and it may already contains implicit biases (cultural or behavioral). When AI learns from this data, those biases can become amplified in the model’s responses. It is also likely to have confirmational bias (favoring information that supports existing beliefs while ignoring other information).

Plus, AI lacks the judgment required to determine whether something is ethically, socially, or logically right or wrong. It just reproduces patterns from the data it was trained on.

Reliability and quality.

Now, how do synthetic respondents compare to real people? We know that AI-generated responses can simulate opinions, but they cannot experience emotions or personal contexts the way humans do. Some online articles suggest that synthetic panels lack the variation found in human responses, moving towards middle-ground answers and lacking extreme opinions.

In addition, they tend to reproduce what is already known rather than something original.

Yet, the same can be said for most low-quality market research. Weak surveys, unenthusiastic participants, biased samples, or poorly framed questions rarely produce meaningful insights for research teams. But all of this makes it easy to fall into the trap of thinking that, because traditional research methods have more or less the same limitations, synthetic panels can replace them entirely.

In reality, both approaches have trade-offs.

The quality of the results, in the end, depends less on the tool itself and more on how it is used. The researcher or marketer plays the central role, deciding which audience segments to study, what questions to ask, and how to interpret the results.

Using Synthetic Panels in Market Research: Concept Testing, Surveys & More

Synthetic panels are simply another tool in the research toolkit, and like any tool, their value depends on how they’re used. When applied correctly, they allow marketers and product teams to explore ideas, test assumptions, and gather feedback long before committing real resources to development or campaigns.

They can support many types of research activities, including:

• Concept testing – Evaluate new product ideas before development.
• A/B testing – Compare visual assets, messaging, or layouts.
• Market messaging validation – Identify which positioning resonates with target audiences.
• Benchmarking studies – Compare perceptions, usage, or performance across segments.
• Customer satisfaction surveys – Run CSAT or NPS-style surveys.
• Focus group simulations – Explore opinions, themes, and discussion-style feedback.
• In-depth interviews – Gather detailed qualitative responses.
• Pricing research – Test acceptable price ranges and perceived value.

Delve AI’s synthetic panels allow you to compare concepts, analyze consumer reactions, and get directional insights early in the decision-making process. Since they’re built using your customer data and industry-specific information, they demonstrate strong reasoning capabilities and generate responses highly relevant to those areas.

The platform supports a wide range of research questions, helping you structure both quantitative and qualitative studies:

• Multiple choice questions
• Rating scales and Likert scale questions
• Matrix questions for evaluating multiple attributes
• Open-ended questions for detailed insights
• Ranking questions to understand priorities
• Demographic questions for segmentation
• Image choice and click-map questions for visual feedback
• Dichotomous (yes/no) questions
• Benchmarkable questions for performance comparisons
• Pricing questions to assess price sensitivity

So users can test different messaging strategies and creative concepts without risking brand reputation or causing fatigue among real customers. As our personas and panels are continuously updated with real-world data, they also provide a more recent view of your audience behavior and preferences.

Are Synthetic Customers the Future of Market Research?

Perhaps sometime is the not-so-distant future, yes. But for now, synthetic panels and customers should not be a substitute for real-world testing.

Contrary to the sensationalist claims made by some synthetic market research platforms, it’s not a good idea to replace your entire user research workflow with synthetic respondents. They should, instead, complement and support real market research.

Ultimately, there’s still a lot we don’t understand about the efficacy and drawbacks of synthetic panels and synthetic data as a whole. So start small, see how (whether) it works for you, and take it from there.

If you're looking for ways to combine synthetic panels with traditional market research, try Delve AI’s all-in-one consumer insights platform!