Will Market Research Survive AI?

Listen to the audio version here

Will surveys soon become obsolete? What skills will market researchers need in the future? Are synthetic data and digital twins just hype, or are they the next frontier? In this article, we’ll break down these concepts, clarify what they mean, and build a surprising hypothesis about the future of market research.

AI is Already Reshaping Market Research

Most of us are already using AI in some capacity. We ask it research-related questions, triangulate incidence rates, and even let it draft survey questions, scales, or entire questionnaires—for inspiration, of course.

We use AI to code open-ended feedback, and some even rely on AI-powered tools to moderate online discussion groups. It helps summarize findings and makes everything sound just a bit sharper.
Now, with reasoning models and AI-driven deep research capabilities, everything is accelerating. What once took weeks of desk research can now be done in minutes.
But AI isn’t just about large language models (LLMs). Let’s paint the full picture and explore what this means for our industry—and for each of us.

What is Synthetic Data?

If you have a dataset—say, from a market research survey—you can train a synthetic data tool to generate new datasets with the same statistical properties as the original. The result? Virtual respondents who don’t exist but could very well represent your target audience.
Here’s how it works: The algorithm estimates a density function. If you plot two variables in a graph, the density function will be highest where most data points cluster and lower where data is sparse.
To create a synthetic dataset, the model generates random data points, filtering them based on density—eliminating more from sparse areas and retaining more where real data is abundant.
Nonparametric density estimators have existed for years, but deep learning has made them vastly more effective, especially when dealing with many variables.
Applications:

• Boosting datasets: Filling gaps in underrepresented demographics.

• Anonymization: Banks, for instance, can create synthetic versions of customer data to share with external vendors—without privacy concerns.

Predictive Research Tools: Faster, Not Perfect

Predictive AI tools are revolutionizing market research. Platforms like Brainsuite, Neurons, and Jumbai analyze visual content—ads, images, and videos—to predict what consumers will notice. Will they read the tagline? Will they recognize the brand?
Similarly, tools like Kantar Link encode visual data to forecast copy-test results, enabling rapid test-and-learn cycles.

But here’s the catch: These tools don’t necessarily produce better research. They just make it faster and cheaper. They are creating value when company would not have tested otherwise.

The real question for businesses is: When is "good enough" truly enough, and when do we need deeper insights?

From Correlation to Causation: The AI Breakthrough

Synthetic data, AI-assisted research, and predictive tools all have one thing in common: They speed up research while reducing costs. But they don’t necessarily improve it.

Causal AI takes a different approach. Instead of just identifying correlations, it seeks to understand the cause-and-effect relationships between actions, perceptions, attitudes, and behaviors. It answers the ultimate business question: What should we do to drive success?

While causal analysis isn’t new—driver analysis has been around for a century, and Structural Equation Modeling for over 50 years—traditional methods are too rigid for real-world complexity. Effects aren’t always linear, and context matters. A great ad might not boost immediate sales, but it can build brand equity that drives future revenue.

Causal AI finally makes these complexities more measurable—and actionable.

Digital Twins: Your Customer, But AI

Imagine an AI-powered language model, customized to behave like your target customers. That’s the concept of digital twins—virtual personas that organizations can directly engage with to ask research questions.

While it’s relatively easy to create a basic digital twin, the challenge lies in ensuring it provides realistic responses. Simply feeding it descriptive survey data leads to parroting known insights rather than uncovering new ones.

To overcome this, researchers must dig deeper—exploring underlying motives, conducting depth psychology studies, and applying Causal AI.

A compelling example: The AI Citizens that Concept M AI developed for the German election. These digital personas demonstrated strikingly natural behavior, with impressive face validity.

Replacing Quantitative Research?

Since 2023, researchers have experimented with using LLMs to fill out quantitative surveys. The results? Surprisingly good—but with significant caveats. Sometimes, the data is unreliable, and it’s difficult to predict when it will be accurate.

As digital twins become more psychologically and causally sophisticated, they’ll only improve. It’s just a matter of time before most surveys could be conducted entirely by AI.

However, this doesn’t eliminate research—it shifts the focus toward building and validating these AI models.

Replacing Qualitative Research?

AI is already moderating online discussion groups. With digital twins, it can even conduct discussions between AI-generated personas. Think of it as a multi-brain reasoning model.

Will this replace qualitative research? Likely, for simpler formats like focus groups and structured interviews.

However, AI won’t replace methods that explore the human subconscious—such as deep psychological interviews or implicit association tests. Why? Because LLMs are trained on language, which primarily reflects rational, System 2 thinking. While traces of subconscious thought exist in language, they are overshadowed by cognitive biases and rational mental models.

The future of qualitative research will lie in exploring the subconscious—an area where AI still struggles.

The Natural Limits of AI in Market Research

Imagine it’s the eve of an election, and news breaks that the leading candidate is embroiled in scandal. An AI model, unless updated in real-time, won’t account fully for this sudden shift in public opinion.

In consumer research, trends and cultural shifts are even harder to measure. AI models are trained on past data, operating under the assumption that fundamental consumer behaviors remain stable. While often true, it’s never entirely reliable.
AI’s limitations mean that some areas of research will continue to rely on direct human input, including:

• Trend tracking (e.g., campaign and experience trackers)

• Innovation research

• Unconscious consumer insights

The Future: A Symphonic AI-Human Collaboration

In the future, questionnaires will generate themselves based on decades of documented experience. Research tools will sync seamlessly. Reports will be auto-generated. Predictive models will evaluate stimuli, while standardized Causal AI will highlight what truly matters. Digital twins will automate most traditional quant and qual research.

What remains? The human element.

Think of an orchestra. AI plays the instruments, but a conductor is still needed to bring everything together. Many traditional white-collar research jobs may disappear, but expertise—deep knowledge, strategic thinking, and creative interpretation—will only grow in importance.

And perhaps most importantly, in an age where we interact more with machines, the value of human connection and validation will rise.

The future standard of research will be AI. The premium service? AI with a human.

THIS… is how you 10x market research with AI.