AI Prompts for Market Research Surveys

Design a market research survey to help us answer this specific business question: [state the decision you need to make].

Target respondent: [specific profile—who should answer this survey and why]
Sample size goal: [how many responses you're targeting]
Survey context: [when and how it will be sent—post-purchase, in-product, panel, etc.]

Survey design requirements:
1. Start with 2-3 screener questions to confirm the respondent matches our target profile
2. Include at least 3 open-text questions (not just rating scales)—place them after the rating questions so respondents are primed to elaborate
3. For each key hypothesis we're testing, write both a direct question AND an indirect behavioral question that tests the same hypothesis without telegraphing the answer we're looking for
4. Include one question designed to surface unexpected reasons—something like "What almost stopped you from [action]?"
5. End with one question that invites new information: "Is there anything about your experience with [topic] that we haven't asked about that you think we should know?"

Business hypotheses we're testing: [list 3-5 specific hypotheses about customer behavior or preferences]

Most market research survey advice is wrong about the most important thing. Guides tell you to "keep surveys short" and "avoid leading questions"—both true—but completely ignore the fundamental problem: most survey questions are designed to confirm what the company already believes rather than discover what it doesn't know.

ai prompts market research survey workflows are most powerful when they challenge that instinct. The best use of AI in survey design isn't generating questions—it's generating the questions you wouldn't have thought to ask.

Before: The Weak Prompt

The most common survey prompt: "Write a customer satisfaction survey for [product/service]."

What comes back: a 10-question CSAT survey with a Net Promoter Score question, some rating scales, and an open-text field at the end. This is technically a customer satisfaction survey. It's not research.

The problem isn't the format—it's the goal. A CSAT survey tells you how happy customers are. It doesn't tell you why they're happy, what would make them happier, what would make them leave, or what they wish you'd build. Those questions require different question design and a different prompting approach.

Why It Fails

A generic survey prompt fails because it doesn't specify what the survey is trying to discover. "Customer satisfaction" sounds like a research objective but it's actually just a metric. A real research objective sounds like: "We need to understand why customers who churn in months 3–6 are leaving—specifically whether it's a product gap, a competitor pull, or a value expectation mismatch."

That objective produces completely different survey questions than "how satisfied are you?"

⚠️ Common mistake: Designing surveys around metrics you want to report rather than decisions you need to make. If you're building a survey to justify an existing decision or to show leadership that customers are happy, you're not doing market research. You're doing confirmation theater. Real research should have the potential to surprise you—and a well-designed AI survey prompt will generate questions you genuinely don't know the answer to.

After: The Improved Prompt

Here's a market research survey prompt that produces genuinely useful questions:

Design a market research survey to help us answer this specific business question: [state the decision you need to make]. Target respondent: [specific profile—who should answer this survey and why] Sample size goal: [how many responses you're targeting] Survey context: [when and how it will be sent—post-purchase, in-product, panel, etc.] Survey design requirements: 1. Start with 2-3 screener questions to confirm the respondent matches our target profile 2. Include at least 3 open-text questions (not just rating scales)—place them after the rating questions so respondents are primed to elaborate 3. For each key hypothesis we're testing, write both a direct question AND an indirect behavioral question that tests the same hypothesis without telegraphing the answer we're looking for 4. Include one question designed to surface unexpected reasons—something like "What almost stopped you from [action]?" 5. End with one question that invites new information: "Is there anything about your experience with [topic] that we haven't asked about that you think we should know?" Business hypotheses we're testing: [list 3-5 specific hypotheses about customer behavior or preferences]

What this does: The "direct AND indirect behavioral question" instruction is the most important design element in this prompt. Direct questions like "Do you value X?" produce aspirational answers. Behavioral questions like "When did you last [action related to X] and what prompted it?" produce real data about actual behavior, which is what decisions should be based on.

⚡ Pro tip: Run this hypothesis-generation prompt before designing any survey: "We're studying [topic] with our [customer segment]. Generate 8 potential research hypotheses—specific, testable claims about how our customers think, behave, or decide—that we might want to validate or invalidate. At least 3 of these should be hypotheses we might be uncomfortable finding confirmed, because those are the most valuable to test." The hypotheses you're afraid to test are usually the most important ones.

Breaking Down Each Survey Element

Screener questions are the most underused element in market research. If your survey goes to the wrong respondents, every answer is noise. A screener for a churn research survey might ask: "How long did you use [product]?" and "Which of the following best describes why you stopped using it?" before asking anything about their experience.

Open-text questions produce the richest data but get the least useful responses when placed first. Place them after at least one or two easier questions—once respondents are engaged, their open-text answers are typically 40-60% longer and more specific.

Indirect behavioral questions look like: "Think about the last time you [relevant action]. What were you trying to accomplish when you did that?" instead of "How important is [feature] to you?" The first question gets a story. The second gets a rating that may not reflect actual usage patterns.

⚡ Pro tip: For competitive research surveys, use this variant: "Write 5 survey questions that reveal why customers chose a competitor over us—without using the word 'competitor' or asking them to compare. Instead, ask about their evaluation process, what they were looking for, what they found, and what made them confident in their choice. These questions should surface competitive intelligence without putting respondents in a frame that makes them defensive."

Variations for Different Research Goals

Product development research:

Write a 10-question product research survey for existing customers of [product]. Goal: identify which of these 5 potential features would drive the most value for our core users: [list features]. Include: a behavioral question for each feature that tests whether they already do the thing manually (indicating real demand), a question that forces a tradeoff between features (to reveal true priorities), and an open-text question about problems we haven't solved yet.

Pricing research:

Write a Van Westendorp pricing sensitivity survey for [product]. Include the four standard Van Westendorp questions adapted for our specific product, plus 3 additional questions that help us understand what's included in the price perception (what do they assume they get for each price point?). Note: Do NOT ask "how much would you pay?"—this is a known leading question that produces unusable data in pricing research.

What this does: The Van Westendorp method is the most underused quantitative pricing research technique in B2B marketing. It produces an "acceptable price range" rather than a single number—which is what pricing teams actually need. I'm not 100% sure why more teams don't use it, but the AI-assisted version significantly lowers the barrier to running it.

⚡ Pro tip: After collecting survey responses, use AI to analyze open-text data: "Here are 50 open-text responses to the question '[question]': [paste]. Identify: (1) the 5 most common themes, (2) any responses that don't fit the common themes (outliers that might reveal something important), (3) specific language that customers use to describe [problem/solution] that we should use in our own copy. Don't summarize—identify patterns and pull specific quotes that best illustrate each theme."

Save and Reuse This

A well-designed survey framework—screeners, behavioral questions, open-text placement, hypothesis tests—takes significant iteration to get right for your specific research goals and customer type. Save your survey design prompts in PromptABCD with notes on what each survey was trying to discover and what it actually found. The gap between those two things is where your best future research questions come from.

Analyzing Survey Results with AI

Designing the survey is only half the work. Turning responses into actionable findings is where most teams either lose momentum or revert to confirmation bias—focusing on the data that confirms what they already thought.

Here are the results from a market research survey with [X] respondents: [paste quantitative results or summary] [paste sample of open-text responses] Analyze these results with the following constraints: 1. Start with what surprised you—findings that don't match conventional assumptions about [topic] 2. Identify any data points that seem to contradict each other and propose an explanation 3. Segment the results by [demographic / behavioral variable if available] and note if different segments show meaningfully different patterns 4. List the top 3 actionable insights with a specific recommendation for each 5. Identify the most important unanswered question that this survey surfaced but couldn't resolve—that's the next survey topic Do not simply summarize what respondents said. Interpret what it means for [specific business decision].

What this does: The "start with what surprised you" instruction is the most important bias-correcting element in survey analysis. By leading with surprising findings, you force the analysis to surface information that challenges assumptions—which is the entire point of doing research in the first place.

⚡ Pro tip: For any survey with more than 100 open-text responses, use an AI-assisted thematic coding process: "Here are 100 responses to the open-text question '[question]': [paste]. Code each response into one of these themes: [list 4-5 themes you've identified]. For any response that doesn't fit the existing themes, suggest a new theme label. After coding, give me the percentage breakdown and one representative quote for each theme." This turns qualitative data into quantitative insight without losing the nuance of the actual responses.