A consumer research API lets an AI agent or a script put a question in front of real consumers and get back findings it can cite: how many people reacted a certain way, why, and in their own words. Two jobs show what that changes for a brand team. A claim test tells you whether people understand and believe a line before it goes on pack or into a campaign. A brand-switching study tells you why buyers left, what triggered the move, and what holds the ones who stayed.
This guide is for the team that wants the evidence rather than the integration. It follows both studies as an agent runs them, then covers how to read the results and how this differs from survey APIs and synthetic panels. Developers wiring the same flow into a service will find every request and response in the developer integration guide. For packaging and product concepts, see how to run a concept test from an AI agent.
Run your first consumer study from an agent. Signup includes 3 free interviews with your own participants; panel recruiting is billed separately. Start free →
What does a consumer research API do?
It turns a question a brand team would hand to a research agency into a short conversation with an agent, or a handful of calls from a script. You describe what you want to learn, supply the claim or the behavior you want explained, pick an audience, approve a price, and receive a structured report when interviews finish.
The difference from other data APIs is the source of the answer. A consumer research API does not look anything up. It goes out and asks people.
A consumer research API is a programmatic interface that commissions interviews with consumers and returns the evidence as structured data. The calling agent supplies a research brief, a stimulus such as claim wording or a competitor’s pack, and an audience definition. The platform converts the brief into a saved interview plan, recruits matching participants, runs AI-moderated conversations that follow up on each answer, and analyzes the transcripts. The result is a report in JSON: findings grouped by learning goal, each with a prevalence count of how many participants said it, verbatim quotes linked to the source interview and, for voice and video, the timestamp in the recording, participant profiles, a judgment on whether the sample supports each conclusion, and recommended next studies. Because every claim points back to a person, the agent can quote the evidence in a brief or a deck without paraphrasing what consumers said.
For a brand team, that means a question about shoppers no longer ends at “we don’t have data on that.” It ends with a study and an answer with sources, which is the core of consumer insights research run from a conversation.
Which consumer studies can an agent run through the API?
Every study type below runs through the same API. What changes is the stimulus, who takes part and what the report answers.
| Job | Study type | Stimulus | Who takes part | What comes back |
|---|---|---|---|---|
| Claim test | Concept test with a claim card, or an in-depth interview with the wording in the brief | The exact claim wording | Category buyers, often split by what they use now | What the claim promises to people, where belief breaks, what proof would make it credible |
| Brand-switching study | In-depth interview | None; participants reconstruct the switch | Recent switchers, lapsed buyers and loyal buyers | The trigger, the moment of decision, what would win buyers back, what holds loyal buyers |
| Shopper-trip study | In-depth interview | None; participants reconstruct a recent trip | Recent shoppers of a retailer or channel | Where the trip broke down and what shaped the choice at shelf |
| Concept or pack test | Concept test | One image per concept | Category buyers | First read, appeal and hesitation for each concept |
| Prototype test | Prototype test | A live link, in voice or video | Consumers or your own customers | Where people got stuck and why, in their words |
Participants come from a vetted 4M+ consumer panel, with B2B audiences recruited to order per study, or from your own customers, invited by email or study link. Concept and pack tests have their own workflow in the concept test guide, and brand lines have theirs in the tagline testing workflow. The rest of this guide covers the two studies a brand team most often needs answered in words: claims and switching.
How does an agent run a claim test?
The scenario: a household brand is launching a bleach-free bathroom cleaner and has two candidate claims for the front of the bottle, “Cleans like bleach, without the bleach” and “Plant-powered clean for every surface.” The team needs to pick one, and it needs to know what shoppers think each one promises before the bottle goes to print.
A claim brief differs from a concept brief in four ways. It gives the wording verbatim, because a one-word change can move belief. It names the proof the brand holds, so the moderator can ask what evidence a shopper would need without revealing it. It lists the claims shoppers already see from competitors, because belief is relative. And it splits the audience by current behavior, because a bleach user and a shopper who already avoids bleach hear the same claim differently.
The brand manager briefs the agent in plain language:
Run a claim test for our new bleach-free bathroom cleaner. We need to choose between two front-of-bottle claims: “Cleans like bleach, without the bleach” and “Plant-powered clean for every surface.” For each, I want to know what shoppers think it promises, whether they believe it, what would make it believable, and which one would make them pick up the bottle. US shoppers who bought a bathroom cleaner in the past month, 30 voice interviews.
The agent sends the brief to the planning step, and the planner comes back with the choice that matters most for a claim. Here is that exchange in the shape the API returns, abridged, with IDs marked EXAMPLE:
POST /api/public/v1/studies/study_EXAMPLE_07/customize-plan
{ "message": "Claim test for a bleach-free bathroom cleaner. Claims: 'Cleans like bleach, without the bleach' and 'Plant-powered clean for every surface'. Learn what each promises, whether shoppers believe it, what proof they need, and which one they would pick up.", "decisions": "human" }{
"response_type": "question",
"questions": [
{ "id": "q_EXAMPLE_1", "text": "Who should we recruit?", "options": ["Shoppers who use bleach cleaners now", "Shoppers who already avoid bleach", "Both groups"] }
],
"study": { "id": "study_EXAMPLE_07", "provisioning_status": "draft" }
}The brand manager answers “both groups,” and the planner writes the plan. Each interview plays the claim back first: in your own words, what is this bottle promising? Then the moderator asks whether the participant believes it, what makes them doubt it, what they would need to see on the label or in use, and how it compares with the cleaner they buy now. Only then does it ask which claim would make them pick up the bottle. With both claims on one study, the moderator captures a first reaction to each before any comparison; for a monadic read, run one study per claim with the same plan and audience.
The agent shows you the saved plan and a cost estimate before anything is spent; in Claude you approve with one click, elsewhere you confirm in the conversation. Most panel studies fill within 24 hours.
What comes back is organized by learning goal. For each claim, the report says what shoppers took it to mean, where belief broke and why, which proof points they asked for, and which claim moved them toward the bottle, each finding with its count and the quotes behind it. The most useful finding in a claim test is often the gap between what the brand meant and what shoppers heard, because that gap is fixable before print.
How does an agent run a brand-switching study?
The scenario: a mid-priced shampoo brand is losing regular buyers to other brands and wants to know why. Sales data shows who left. It cannot show what happened at the moment they decided.
A brand-switching study needs no stimulus. Participants reconstruct their own switch, and the method is a timeline. The moderator anchors on the last purchase of the old brand, walks back to the first doubt, and walks forward to the first purchase of the new one. Along the way it separates four forces: the push away from the old brand, the pull of the new one, the hesitation that delayed the move, and the habit that held them as long as it did.
The brand manager briefs the agent:
Run a brand-switching study for our shampoo. Talk to US shoppers who bought our brand regularly and switched to another brand in the past six months, and to shoppers who still buy it every time. For switchers, I want the last purchase before the switch, what triggered it, what they buy now and why, and what would bring them back. For loyal buyers, what keeps them. 20 voice interviews per group.
The agent turns this into a saved plan with screener questions that sort participants by what they bought before and what they buy now. It shows the plan and the estimate for approval in the same way as the claim test, and fielding starts once you approve.
The report separates the trigger from the cause. A trigger is a moment: a price change, an empty shelf, a friend’s recommendation, a new formula. The cause is what had built up before it: a scalp that never felt right, a scent that stopped feeling special, a sense the brand was not for them anymore. A win-back offer aimed at the trigger can earn one purchase; a fix for the cause is what keeps the buyer. The loyal group shows the other side, so check the attributes loyal buyers name first against the ones switchers stopped noticing.
Ask follow-ups in the same thread. Which switchers would come back, and on what condition? What did loyal buyers say about the new formula? What did people who switched to a store brand have in common? The agent answers from the report and cites the quotes.
How should you read claim and switching results?
Read prevalence as a count, not a percentage. A finding stated as n of N keeps the sample size visible wherever the finding travels, and it stops a strong quote from standing in for a majority. To see how a finished report presents counts, quotes and recordings, explore an example study.
For claim tests, read three layers separately:
- Comprehension. Did shoppers read the claim the way the brand meant it? A playback that differs from the intent is a wording problem.
- Credibility. Do they believe it, and what would they need to see? The doubters’ quotes tell you which proof to add to the label.
- Motivation. Does it change what they would pick up? A claim can be understood and believed and still move no one.
For switching studies, read the switchers’ language beside the loyal buyers’. When both groups describe the same attribute in different words, that difference is usually the story. And read the win-back conditions as conditions: “I would come back if” tells you more than “I might come back.”
For both, ask the agent who disagreed with each finding, and check the report’s coverage judgment before you conclude. If the sample does not support a learning goal that touches the decision, have the agent scope a focused follow-up instead of stretching the result. Before the next brief, the agent can also check what earlier studies already say about the same claim or category; see how to ask questions across past research.
How is this different from survey APIs and synthetic panels?
The three tools answer different questions, and a consumer research program can use all of them.
| Survey API | Synthetic panel | Consumer research API | |
|---|---|---|---|
| Who answers | People completing a fixed questionnaire | A model generating responses | People in a moderated conversation |
| What you get | Counts and distributions per question | Simulated answers | Findings with prevalence, reasons and verbatim quotes |
| Follow-up questions | Fixed by the questionnaire | Generated | Asked live in each interview, laddering into reasons |
| Best fit | Sizing a known answer across a large sample | Fast directional checks before fieldwork | Learning why people believe, doubt or switch |
A survey API is the right tool when you already know the answer options and need to measure them at scale. A consumer research API is the right tool when the options are the thing you don’t know yet: why a claim lands as doubtful, which proof would fix it, what made a loyal buyer leave. Teams often run interviews first to find the reasons, then size them with a survey whose answer options come from what consumers said. Synthetic respondents are also available on User Intuition, in chat, for a directional first read before fieldwork with people.
For a wider comparison of vendors, see the best customer research APIs.
How does User Intuition run consumer research through an API?
User Intuition exposes the full study lifecycle as a REST API and as a hosted MCP server, so an agent can run both studies above from a conversation or a script. For the product overview, see the research API and MCP server for AI agents; for the tool-by-tool reference, see the MCP research API.
- Depth. User Intuition’s AI moderator runs 20–30 minute interviews that ladder 5–7 levels deep, in voice, chat or video, with interviews in 80+ languages.
- Audience. A vetted 4M+ consumer panel, with B2B audiences recruited to order per study, or your own participants.
- Quality. Every panelist call is reviewed by the User Intuition team, panelists pass a video identity check, and only interviews that pass quality scoring are billed; see how quality works.
- Trust. Participants rate the experience at 98% satisfaction, and User Intuition holds 4.9/5 on G2 and 5/5 on Capterra.
- Price. A 20-interview panel study is $1,200, recruiting included.
- Connect. Connect it in Claude, ChatGPT, Codex or Cursor in a few minutes (setup). Scripts call the REST API with an API key, and the developer integration guide shows each call.
The MCP server and the REST endpoints share one data model, so a claim test a brand manager starts in Claude can be tracked by a backend service and reported into a team channel against the same study ID.
Put a claim or a switching question in front of consumers this week. Create a free account →