Generative AI Is Reshaping Market Research — But Most Companies Still Haven't Caught On

A while ago, I grabbed dinner with a friend who runs a consumer brand. He ordered an Americano, looked troubled, and told me:

"We just spent 800,000 RMB hiring a research firm to produce a 120-page consumer profile report. The day they delivered it, I fell asleep after the first three pages."

I asked him about the other 117 pages.

He said, don't even bring it up. The data was six months old, and the conclusion was "Gen Z values personalized experiences." He looked around and discovered that a competitor had already redesigned their packaging in that direction last month.

This had me thinking for days.

Market research, at its core, answers two questions: What do consumers actually want? And what are competitors actually doing? The old way was sending out surveys, conducting interviews, buying industry reports — three months minimum, hundreds of thousands of RMB at a baseline, and by the time conclusions landed, the world had already moved on.

Then generative AI arrived.

An Underestimated Battleground

In 2025, Harvard Business Review published an article whose title roughly translated to "How Generative AI Is Changing Market Research." The core argument caught my eye:

Among all management functions, the one most thoroughly disrupted by generative AI is very likely marketing itself.

But here's the interesting part: over the past two years, the discussion has focused almost entirely on customer service and content creation. Chatbots can answer messages 24/7. AI can generate ten copy variations in a second. These two areas are genuinely buzzing — so much so that many people assumed the marketing AI story was already told.

It isn't. Not even close.

The next real wave of opportunity is hidden inside market research — the very segment that sounds the most "traditional," the most "heavy," the least likely to be disrupted by technology.

Why Market Research?

Think about it: when you run a research project, what eats the most time and money?

It's not the analysis. It's the collection.

In the past, if you wanted to understand Chinese young women's true attitudes toward a certain skincare product, you had to find a sampling company, design a questionnaire, send out thousands of copies, collect responses, clean the data, code it, run models. Three months gone, just like that.

Now?

You feed a year's worth of relevant social media discussions into a large language model, and within hours it produces a sentiment trend report broken down by city, age group, and spending power. You ask it, "How have attitudes toward ingredient-conscious consumers shifted among 25-year-old women in Shanghai?" and it pulls up a timeline for you right there.

Or take competitive analysis. It used to rely on analysts manually flipping through earnings reports, watching product launches, scraping product pages. Now you let AI continuously monitor a competitor's website, pricing, ad placements, and user reviews — it alerts you the moment something changes.

The core action of research has shifted from "sampling" to "census."

That shift is profound.

The biggest weak spot of traditional research is sampling bias — you survey 2,000 people and dare to extrapolate the preferences of 1.4 billion. Now you can directly process millions of real discussions. The bias still exists, but in terms of scale, it's not even the same game anymore.

Three Directions Worth Pursuing

Based on practices from academia and frontline brands in recent years, three scenarios have proven viable.

First, real-time consumer insight.

No more waiting for quarterly reports — instead, checking a sentiment dashboard every day. Which feature is getting roasted, which packaging is earning praise, which celebrity spokesperson triggered a backlash — AI keeps watch for you. One global fast-moving consumer goods brand told me they compressed their consumer feedback processing cycle from 6 weeks to 3 days.

Second, automated competitive intelligence.

A competitor changed their pricing strategy, launched a new SKU, suddenly ramped up ad spending in a particular city — these signals used to require human monitoring. Now AI systems automatically capture and flag them. The key is that it can simultaneously watch dozens of competitors — something the human brain simply cannot do.

Third, scaled hypothesis testing.

Running an A/B test used to mean scheduling, sourcing samples, securing budget. Now you can run hundreds of "what if" simulations simultaneously, letting AI filter out the obviously unworkable directions and reserve human energy for the few that genuinely need judgment.

But Don't Celebrate Too Early

Every tool comes with a cost.

The biggest risk generative AI brings to market research is the hallucination that sounds completely plausible. It can generate a report with rigorous logic, detailed data, and beautiful charts — but a key number in there might be something it simply made up.

So the smartest approach right now isn't to use AI to replace the research team. It's to let the research team use AI to handle the grunt work — collection, cleaning, initial screening — and then have humans do the one thing machines can't: judge which signals are real trends and which are just noise.

Back to my friend.

I later told him: of that 800,000 RMB, at least 600,000 could be saved next time. But don't pocket all the savings — take a portion and hire an insight analyst who understands both AI and your business.

People like that are scarcer than gold right now.

The barrier to conducting research is being flattened. But the barrier to reading research results is actually being raised. Because when data becomes abundant, fast, and cheap, what's truly valuable is whether you can spot, in that ocean of information, the card no one else can see.

That's probably the greatest paradox generative AI has left for market research.

The more powerful the tools, the more important the human.