When marketing is perceived as the technology of producing content faster, creating copy for ads or as an automated way of running non-strategic programs, that’s when you’d see the term ‘AI’ mentioned. But some of the more promising developments are exploring a different kind of question: Can artificial intelligence increase a marketer’s understanding of customers?
However, MIT Sloan research indicates that it may-if AI is trained to be a “specialist” in just one marketing research task rather than a “generalist” seeking to replace human marketers. A recent study considered whether large language models (LLMs) could uncover customer needs and desires from interviews, product reviews, Facebook posts, and other types of customer-generated data.
Great marketing begins with figuring out what customers really want. And that’s easy to say, but most of the time customers aren’t spelling out their needs in exactly the right language that a product team or a marketer needs to know.
A customer could say, “I run out of battery power of my phone too soon, especially while traveling.” The above tells a fact about the customer, but there might be a better need – the customer wants to be able to use their smartphone for extended periods of time without a wall plug nearby. That’s what a professional market researcher would do.
MIT Sloan professor John Hauser and his team have explored whether AI could help. The study is especially relevant because it moves away from using the copy of AI and focuses instead on whether LLMs can contribute to the initial stage of marketing – capturing the customer’s voice.
The researchers contrasted professional analysts with large language models trained to diagnose and articulate customer needs. Rather than asking a general language chatbot what customers wanted, the researchers trained them with supervised fine-tuning on hand-annotated examples from market research papers.
This difference matters because it blows a hole in the argument that the more intelligent and capable a general-purpose AI is, the better the marketing insight it will generate. In fact, the study found that training method and task are the key.
The researchers first applied the approach to wood-stain products and then duplicated it across other categories of products and services. “The precision at which the models pinpoint customer needs and map them to ideal solutions was comparable to (or even superior to) our trained professional analysts,” according to MIT Sloan.
The outcomes were remarkable. In the study reported, the fine-tuned model was able to identify all 8 primary customer needs and all 30 secondary customer needs, while professional analysts were able to identify 87.5% of the primary needs and 80% of the secondary needs.
That doesn’t mean the research demonstrates that AI has a superior grasp of customers than veteran marketers. The test was structured around a narrowly defined analytical challenge: identifying, digesting, and classifying customer needs from research material.
There’s an interesting intersection in this research and SEO and digital marketing teams use to get the ranking signals from search engines. Search marketers start out using keywords, search volumes, competitor pages and rankings opportunities, but they don’t always help you tell us how come you’re searching.
What if someone is searching for washing machine won’t drain? The keyword is what someone actually entered, but it does not say everything about the issue. The person might want to know if it’s possible to take care of the problem by themselves or if it’s risking serious damage to keep using the washing machine, how much repairs might be, and when to call for help.
Those are customer needs, not just keywords. Analyzing the data with the help of AI could help marketing professionals discover customer needs in the flood of the customer feedback, and then use these needs to write better content.
It’s not just about finding the most commonly used words, but about getting to the root of why people are using them in the first place. This may mean analyzing video testimonials, survey results, comments on social media, reviews of products, or insights from interviews with the users.
MIT Sloan points out that even professional analysts might spend hours poring over all that data before they get to the far more valuable stage of really understanding the insights, bringing them to life, and integrating them into the innovation and product development process. AI might help to speed that initial part.
It is perhaps a way of having the research robotic, doing more of the preliminary sorting and pattern-drawing, and giving the human researcher more time to identify the significance of what is observed. So it isn’t the researcher being replaced, but the repetitive work around sorting and pattern-trawling that is offset.
Perhaps the most critical point in MIT Sloan’s study is that the researchers did not just depend on a general LLM. Instead, they employed supervised fine-tuning using carefully hand-curated voice of the customer research.
This assisted the model in understanding the way these analysts are used to articulating customer needs. According to reports by MIT Sloan, the team realized that just over 1,000 previously known customer needs were enough for the supervised fine-tuning step in their research.
That distinction is especially important for companies just getting started with AI. Asking a general-purpose bot a question like “What do my customers want?” is quite different than creating a well-defined process that instructs the model to examine true customer data using an approach optimized for the specific task.
What does this mean for content marketing? Rather than structuring content around keyword variations, marketers might leverage customer comments to uncover the questions, frustrations, expectations and issues that surface time and again in direct customer conversations.
From those results, the business can generate blog topics, FAQ questions, product pages, service pages, comparison posts, videos, and anything else they want to create. The shift goes from “What keyword do we want to target?” to “What problem does the customer want to solve?”
This matters even more when search goes conversational. With AI-based search, queries are less about matching exact words and more about the user’s intent.
The MIT Sloan study also offers a good reason to be cautious. Researchers discovered that the base LLMs alone were not enough to match the skill of professional analysts – the fine-tuned models performed more strongly because they were trained to do the task.
Translation: Enterprises should be wary of trusting a standard AI output as a substitute for market research conducted by a human. Although an AI tool can search through data for trends, the quality of those trends relies on the data source, as well as what data science, training, and validation was implemented.
There also needs to be a broader context for a decision that simply might not be available in the data. A model may highlight that customers have grumbled about a product feature over and over, but a team still has to work out whether fixing it is a good idea business-wise, technically and strategically.
Role of research Perhaps, if AI can cut down the time consumed to go through customer data, the role of researcher could shift, over time. Rather than putting in all the effort for collecting and plotting observations, he or she could be validating the findings, extracting strategic insights, and recommending a way forward.
This aligns with the ways that MIT Sloan has explained the potential value of this technology. The idea is to free an analyst’s time to do other (more valuable) activities, not to remove the analyst from their job altogether.
That’s an important distinction as marketing has always been a mix of analysis and judgment. AI can do the analysis faster but it can’t do it without people with knowledge of the market, the customer and the business.
Customer research the way you used to do it This type of customer research can cost a lot of time, money and skills. That means it can be easier for larger companies to do this kind of research than for startups or small businesses.
MIT Sloan suggests an intriguing opportunity here. Entrepreneurs won’t be able to produce large formal research projects, but they will have conversations with customers, interview transcripts, reviews and other first-party data that can be analyzed using AI tools trained for the purpose.
That might enable some kinds of customer-needs analysis to be easier. It won’t free you from the obligation to do a good study, but it might enable a smaller team to identify trends that are worth exploring before doing a big study.
This concept has now been elaborated upon by MIT Sloan. In September 2026, MIT Sloan released a research commentary on using generative AI to predict potential market reactions prior to taking a decision.
The authors also note one major caveat to keep in mind: “AI simulations shouldn’t be used as a proxy for gold-standard market research. They’re more of a way to give initial directional feedback, identify possible issues, and suggest areas for which more in-depth research could be helpful.”
Hence, there’s an even bigger correlation between MIT Sloan research and SEO than AI-created content. The research points to a possible move away from content that is based on search behavior to content that is based on better customer needs.
What consumers search for What customers want to know 4. Search data can tell us what people are searching for. Customer research can tell us why people are searching for it.
Use both together and you could have the most powerful of all your content strategy. Rather than writing numerous pages targeting variations of the same keywords, content marketers could focus on the issues and craft resources that resolve those issues from the customer’s point of view.
Lesson 1: Use real customer data first Time and again, the most effective marketing uses actual customer input whenever feasible. Demands for reviews, interviews, surveys, support conversations and the like will benefit you far more than requests to an AI to predict your audience.
Number two is that it’s important to specialize. Our MIT Sloan study demonstrates a tangible benefit from an LLM that has been specially “tuned” for a particular customer needs task, versus a general purpose LLM.
Third, there is still the need for human insight. AI can be useful in revealing patterns, but it is up to marketers to decide whether they really are relevant, representative, and useful.
Finally, Customer needs can be a beneficial link between research and content strategy. When marketers identify the true need that is behind the search query, they will be able to develop the content that resolves the actual question rather than by words people have typed in to a search engine.
The initial wave of generative AI in marketing was Production-heavy Generated copy, images, ideas, and that sort of thing.
MIT Sloan research indicates another avenue. AI can be woven more and more deeply into ways to process customer information, reveal needs, emulate likely responses, and inform decision-making prior to any marketing action.
That may be an even more profound departure than automated editing.This future marketer might not be using the tool simply to create more content, but rather to get to know customers in a way that wasn’t otherwise feasible or affordable. In the end, the edge may come from knowing how to use the tool to make sense of all the data.
