AI in market research: What’s ready. What’s not.

AI is driving rapid change in the market research industry, as it is in most sectors. Most of us market researchers (95%) are using AI tools regularly or experimenting with them, and half of us (51%) are using them regularly, according to Qualtrics’ 2026 Market Research Trends Report. Some of us are even using “synthetic data” most of the time, that study says. For a time-machine perspective, in my first full-time ad agency job (1991, Dallas, Texas. Big hair. Shoulder pads.), we had a secretary pool with shiny new NeXT computers and a row of fax machines. So, my mind is officially blown in the current moment, but I’m determined to roll with the changes, to quote REO Speedwagon (and further “date” myself).

At the same time, we’re hearing from some clients and prospects that they don’t want AI-assisted research, despite the increased speed to insight and cost-efficiency that it can provide. The hesitancy is quite understandable. It’s a big change and people have concerns about accuracy.

Let’s dispel the anxiety and look at where AI is most helpful right now in the realm of research and insights, where it is ready for the right applications, and where it’s not quite “there” yet — but getting “there” quickly.

The lowest hanging fruit: Expediting market research design and analysis.

AI is enhancing current market research practices in several ways, including:

• Identifying hypotheses, questions that would be interesting to ask, prompts and response categories that may have been missed, making survey design faster and easier
• Automating the process of coding open-ended survey responses, processing thousands of comments in a matter of minutes
• Extracting and synthesizing insights and creating first drafts of reports

Eva Gott, Partner and Head of Research at ThinqInsights, one of our strategic research partners, said:

“There is this overarching idea that AI is just going to take over everything, like Skynet from the Terminator. I get the instinct, but where we are right now, the AI we’re using is more like a smart intern who helps you work more efficiently. I like to use AI for routine tasks: checking data, proofing, exploring possible survey answer choices. AI doesn’t get tired. If you give it exact directions, it executes perfectly, without fatigue.”

When it comes to analyzing qualitative research, AI technology has been life changing. As much as I love reviewing a big pile of interview transcripts (and I really do), it’s more efficient for AI to assist. When moderating interviews, my practice is to jot down high-level notes that nudge me on what I want to make sure to have AI pull out of the transcripts later. This cut several days out of our latest qualitative project. I’m not saying that human transcript review is obsolete; I still find it helpful to review some of them, to jog my memory regarding insights. However, with AI, I don’t need to review all of them in their entirety, and pulling evidentiary quotes is a breeze.

Another of our research partners, Sunseed Research, feeds qualitative responses into its proprietary learning model, which it can then query. The AI pulls up transcript excerpts and video clips about the topic at hand and provides its overview of the themes for the analyst to consider. This has become standard practice for many researchers.

Ripe for the right applications: AI-moderated interviews.

One increasingly popular approach is using an AI chatbot to moderate interviews, sometimes called conversational research. As with everything, there are pros and cons that must be considered for each application of the method.

AI moderated interviews can be an effective way to acquire insights at scale for reasonable costs, especially for audiences that are hard to recruit and engage in research. Some researchers have also found that interview participants are more forthcoming when being interviewed by AI rather than people, according to Harvard Business Review.

The biggest downsides are the lack of nuances and body language that a human moderator can pick up. For example, in a recent interview I moderated, a client’s customer put his hand on his heart when talking about how hard the brand’s name change would be for him. Another shortcoming is AI’s inability to pivot in the moment if needed.

What the pros and cons result in is the need to carefully consider whether the method is appropriate for the purpose, as with any research study. If the objective is to get a deep understanding, empathetic view of how cancer patients feel, we likely would not recommend this method. But if the objective is to get some quick reactions to potential positioning directions or creative executions, it can be a viable approach.

Gott explained this methodology further, to dispel concerns that some marketers have:

“People tend to think the AI moderator is a living, breathing thing that could go rogue. But that’s not how it operates. It’s very well structured. You’re taking a discussion guide and creating a very specific script for the AI to use, with discrete question and probes. It’s not going to come up with its own questions. What it gives you is the ability to do many more interviews, because it’s running them simultaneously. It’s an efficiency play, again. If I need an hour-long interview, I’m still going to use the traditional qualitative route. But if you only need to ask people five to seven questions, this can be so much faster and more affordable than traditional qualitative.”

She also pointed out that as our respondent base gets younger, conversational research platforms that look more like text messages can be more engaging than other methods, because it mimics how they communicate.

Maturing rapidly: synthetic personas and digital twinning.

Getting real people to give feedback is getting increasingly difficult, according to Gott.

“It’s survey fatigue, number one. Everything you do now people are asking for your review. People are getting tired. And there’s something broken in the survey sample world. There are a lot of bots and a lot of bad actors trying to get more money for doing surveys.”

A sample quality management platform can stop those bad actors, but sampling challenges are one reason why researchers are increasingly turning to synthetic data, in addition to saving time and money.

An article in Harvard Business Review in Fall 2025 entitled “The Tools that are Transforming Market Research,” provides an in-depth discussion of the synthetic persona and the digital twin. In the synthetic persona approach, an AI model is provided with demographic, psychographic and behavioral information about a customer type or segment to create a persona that is representative of that type or segment. They can then ask the model to answer questions as if it were that type of person. If a company has solid individual-level data about a group of customers from past interactions or research, it can create digital twins, who can then take part in research in place of the people they represent.

Researchers are growing increasingly comfortable with synthetic responses. In 2025, 40% said they would be comfortable with future research having 51% or more synthetic responses, according to Qualtrics research. And among researchers who have adopted synthetic data, 45% now view it as their most reliable data source.

Gott and I discussed that “synthetic data” could use a rebrand. As she put it:

“I hate the word synthetic, because it sounds like it’s made up, when it’s based on actual survey data. To me, it’s more like modeled or simulated.”

She recommended considering a hybrid approach, in which part of the sample is from a panel-based sample and part is modeled. Another best practice in this area is using synthetic data to move faster on early-stage testing, followed by using human panels to validate high-stakes decisions.

Regarding the “digital twins,” a 2025 Columbia Business School study called the Digital Twins Initiative, tested over 2,000 digital twins concluding that they were more likely to carry “pro-human” and “pro-technology” biases and more likely to provide socially desirable answers. Ultimately, the CBS team concluded that “while digital twins show promise, they are not fully ‘ready for prime time’ yet.” That said, 2025 was a long time ago, in AI time. Some researchers feel these methods are absolutely “ready for prime time.”

A few questions to ask when choosing your research approach.

• Is this an objective that AI can accomplish, or does it require more human empathy?
• What’s most important to this project — efficient scale or emotional depth?
• If empathy and emotional depth are required, where can I still plug AI in to support efficient study planning and reporting?
• How important is the ability to see respondents’ body language or pivot mid-interview?
• Is there a point in this process during which synthetic or modeled data is enough or can augment the sample?

Remember: AI is not a replacement for human judgement.

Clearly, AI can do many things well. However, it is not a replacement for an insightful human researcher, by any stretch of the imagination. AI sometimes has strong biases and makes unsupported logic leaps that need to be checked. So, while AI “agents” may take over more research projects more completely, humans are still needed to provide the strategic judgement that makes AI useful. Think of AI as handling the scale and speed, while human researchers provide empathy, nuanced interpretation, and essential subject matter knowledge.

The bottom line: Do not fear AI-assisted research, for the humans are still driving.