The appeal of AI for market research is obvious: faster synthesis, broader coverage, and lower cost than traditional research approaches. The risk is equally real. AI-generated research looks authoritative whether or not it is accurate, and errors made at the research stage compound at every subsequent decision point in the strategy process.

The marketers who use AI most effectively in research are not the ones who use it most extensively. They are the ones who understand precisely where AI is reliable and where it is not, and who design their research processes around that distinction.

What AI Is Good at in Market Research (and the Evidence for It)

AI performs consistently well at three types of research task. The first is synthesis across large document sets. Given a body of existing research, reports, and publications, an AI assistant can identify themes, extract key findings, and surface patterns that would take a human analyst significantly longer to find. McKinsey's 2024 analysis of AI in professional services estimated that synthesis tasks of this kind are completed 40-60% faster with AI assistance, without meaningful loss of accuracy when the source material is reliable.

The second is competitive monitoring at scale. AI can track signals from public sources, including press releases, industry publications, job postings, and public financial data, across a defined competitive set, generating regular summaries of what is changing and flagging anomalies. This isn't deep competitive intelligence; it's efficient surface monitoring that would otherwise require disproportionate human effort for its value.

The third is trend identification. AI can process a larger volume of public signals than a human analyst can, identifying emerging topics and themes before they surface in formal research publications. This is most useful for early-stage content strategy and opportunity identification, where directional signals matter more than precision.

Across all three, the common factor is that AI is synthesising and organising information that already exists in accessible form. It is faster at this than humans. It is not generating new insight that doesn't exist in its source material.

The Research Tasks AI Gets Wrong More Often Than It Gets Right

Four categories of research task are genuinely unreliable with current AI tools, and the reliability problems are structural rather than temporary.

Precise statistics. AI tools generate numbers with confidence regardless of whether those numbers exist in their training data. Forrester's 2024 guidance on AI research tools for B2B marketers warned specifically against using AI-generated statistics without primary source verification. If you ask an AI for a specific market size figure and it provides one with a citation, find the original source. If you can't find it, the number is unverified and should not appear in a strategy document.

Primary qualitative insight. Understanding what your specific buyers actually think, feel, and believe requires conversations with those buyers. AI can analyse what customers have said in public forums, reviews, or published interviews; it cannot conduct the equivalent of a qualitative research programme. The difference matters because public sentiment is often unrepresentative of your specific buyer segment's actual views and motivations.

Recency. AI models have training cutoffs, and even retrieval-augmented systems have gaps and lags. If a significant market development happened recently, the AI may not know about it, or may know only early coverage that was subsequently revised. For any research task where current information is material to the conclusion, AI should be treated as one input among several rather than the primary source.

Your specific market context. AI has general knowledge about most markets; it doesn't have knowledge of the specific dynamics, relationships, and competitive realities of your particular market segment. Insights that depend on understanding your buyers, your competitive position, or your category's specific buying process need to come from people who have that context, not from a model trained on general data.

A Practical Research Stack: When to Use AI, When to Use Primary Research

A tiered approach based on question type and decision stakes prevents both under- and over-reliance on AI. The tier determines the appropriate method; the decision stakes determine how much verification is required before the research informs strategy.

Tier 1: Use AI freely. Landscape orientation, competitive surface monitoring, trend scanning, secondary research synthesis, early-stage hypothesis generation. These are lower-stakes, exploratory uses where speed is valuable and errors are caught before they influence significant decisions. AI accelerates this work substantially, and the cost of an occasional error is low.

Tier 2: Use AI with validation. Customer sentiment analysis from public sources, category-level trend identification from existing reports, industry analysis based on published data. AI accelerates this work; humans verify significant findings against primary sources before the findings are used to inform strategy. The validation step is not optional at this tier.

Tier 3: Primary research only. Anything that will directly inform a significant strategic decision: customer buying behaviour, competitive pricing and positioning, barriers to adoption in your specific market, qualitative voice of customer, market sizing for a new category. AI may help synthesise what primary research surfaces; it should not substitute for the research itself.

The Ehrenberg-Bass Institute's research on B2B marketing effectiveness consistently emphasises the importance of empirical data about your own buyers, gathered through direct research rather than inference. That emphasis applies equally to how you source the insights behind your strategy: the closer to a consequential decision, the more important it is that the data is primary and verified.

How to Validate AI-Generated Insights Before They Inform Strategy

Four steps reduce the risk of acting on flawed AI research. These steps add time to the research process, but less time than recovering from a strategy decision made on incorrect information.

Source tracing. Every significant claim should have a traceable primary source. If the AI cites a study, find the original study and confirm the finding matches what the AI reported. AI tools paraphrase, misattribute, and occasionally fabricate citations. If you cannot find the source, the claim is unverified.

Cross-referencing. If an AI-generated insight is important enough to inform a strategy document, look for confirmation in a second independent source. Agreement between two sources doesn't guarantee accuracy, but it substantially reduces the probability of acting on a fabrication or an outlier finding from one study.

Domain expert review. Before AI-generated research is used to inform a strategy recommendation, someone with genuine market expertise should review the findings. They don't need to redo the research; they need to assess whether the findings are plausible given what they know about the market. Expert review catches the category of error where AI findings are technically sourced but contextually wrong.

Flagging uncertainty. When AI-generated research is used in a strategy document, note it as such and document the verification steps taken. This creates accountability, prevents AI-assisted research from being treated with the same confidence as primary research, and gives decision-makers the information they need to weight the findings appropriately.

The goal is a research process that is faster where speed is appropriate and rigorous where rigour matters. AI handles the former well. Humans remain responsible for the latter.

The Marketing & AI track at FP Collectiv covers how to build AI research workflows that are both fast and defensible, with practical frameworks for task allocation and validation.

Sources

• McKinsey Global Institute, "AI in Professional Services: Performance Benchmarks," McKinsey & Company, 2024
• Forrester Research, "Guidance on AI Research Tools for B2B Marketers," Forrester, 2024
• Ehrenberg-Bass Institute, "How B2B Brands Grow," University of South Australia, 2023