Participant Recruitment API for Programmatic Interviews

A participant recruitment API lets software recruit people for research without a person clicking through a dashboard. The caller describes an audience, gets an estimate, and launches fieldwork by code. The distinction that matters for AI agents is whether the API only delivers participants or also runs the interviews and returns the evidence. This guide sets out what a recruitment API should cover, how marketplace APIs differ from research APIs, and how programmatic recruitment works through the User Intuition participant recruitment platform and its research API and MCP server.

What is a participant recruitment API?

A participant recruitment API is a set of operations that let software do what a research coordinator does by hand: define who qualifies, check whether that audience is reachable and at what cost, and start recruiting. A complete interface exposes the audience catalog, a screener, an estimate, a launch and a progress signal. Most interfaces on the market expose the recruitment half only. The participant is delivered to a survey link or a scheduling page, and the team runs the interview, the quality checks and the analysis through other tools and other integrations.

How do recruitment marketplace APIs differ from research APIs?

A recruitment marketplace API delivers participants into a tool you run elsewhere; a research API recruits, interviews and returns the findings in one flow. Three kinds of API answer the question “can software recruit participants?”, and they differ in where the work stops.

A recruitment marketplace API, such as Prolific’s, is the right tool when the team already owns the interview and analysis stack and only needs participants delivered into it. A research API is the right tool when the caller is an agent or a product that needs the whole loop to run without a person stitching tools together.

What should an agent be able to do through a recruitment API?

An agent acting for a researcher needs six things from the interface, and the absence of any one of them pushes a human back into the loop:

1. Describe the audience in structured terms. Country, demographic attributes and screener questions as fields, not as a brief a coordinator interprets.
2. Get an estimate before committing. Cost, expected timeline and feasibility for that specific audience, returned as data the agent can show to the person it works for.
3. Stay within limits a person sets. The launch must need a credential scoped for spending and either a person’s confirmation or a budget a person set, so an agent never spends beyond what someone allowed.
4. Run the interview in the same flow. If the API hands participants to another tool, the agent now needs a second integration and loses the link between the participant and the evidence.
5. Follow progress without polling blindly. Webhooks or a progress endpoint that report completed interviews and finished reports.
6. Read results as structured data. Findings, verbatim quotes and the source interview for each one, so the agent can cite rather than summarize.

How does programmatic recruitment work on User Intuition?

User Intuition exposes recruitment and interviewing as one sequence through the REST API and the MCP server, which share the same operations under different names. The agent guide in the docs walks through the sequence with scopes and retry rules.

1. Create the study. POST /api/public/v1/studies/ records the study and its recruiting_method: panel for the 4M+ vetted consumer panel, byop to invite your own participants, or synthetic respondents for a directional first read in chat.
2. Design the plan. The customize-plan operation turns a research brief into a saved interview plan and returns any open questions, such as who qualifies or how to show a concept. The agent relays them to the researcher. This call never starts recruitment.
3. Target and estimate. For a panel study, targeting_attributes takes qualification IDs from the targeting-attribute catalog, screener_questions adds study-specific screening, and country_code sets the market. The estimate operation returns cost, balance and expected timeline for that audience.
4. The agent launches within what a person allowed. The saved plan and the current estimate are shown to the researcher, who approves the launch, unless a workspace owner has set a monthly panel budget, in which case one-time launches within it go ahead without a new approval. A REST launch requires a key with the panel:launch scope, and a changed plan or price needs a fresh estimate.
5. Interviews run. Participants move into 20–30 minute AI-moderated interviews in voice, video or chat, in 80+ languages, and the moderator ladders 5–7 levels from each answer to the reason beneath it. Most panel studies complete within 24 hours; B2B audiences such as IT buyers or clinicians are recruited to order per study.
6. Results come back as JSON. Webhooks announce completed interviews, finished studies and ready reports. The report returns findings with prevalence counts, verbatim quotes and the source interview behind each quote, and the agent can ask follow-up questions across the whole research library with cited answers.

MCP clients see the same sequence as tools: create_study, customize_study, list_targeting_attributes, estimate_panel, launch_panel, get_fielding_progress and get_study_report. The MCP research API guide covers the tool surface in detail, and the interview API integration guide shows the REST calls with request and response bodies.

Why do spending limits belong in the API?

An agent that can launch paid fieldwork without limits is a budget risk, and most teams will not connect one to a credential that can spend. User Intuition separates the estimate from the launch, so the agent does all of the preparation while a person decides how much of the spending it does. The estimate runs with ordinary read and write scopes, and a launch then goes one of two ways. By default, a person approves each launch: in Claude, which supports MCP Apps, a review card shows the plan and the quote side by side for a one-click approval, and in ChatGPT, Codex and Cursor the person confirms in the conversation. Alternatively, a workspace owner or admin sets a monthly panel budget, and the agent launches one-time panels within it on its own. A launch that would exceed the remaining budget is refused with the amount left and the reset date. REST clients launch this way, with a key that carries the panel:launch scope.

How are participants targeted and screened?

Targeting through the API uses the same catalog the dashboard uses. The targeting-attribute endpoint lists qualification IDs for age band, income decile, region, category purchase and other standard attributes, 180+ in all, and the study adds its own screener questions on top. Country is set on the estimate and again on the launch so the quote and the fieldwork always refer to the same market. Panelists reach the study only after the same vetting every User Intuition study uses: a video identity interview reviewed by the team before acceptance, signup verification, screening of every voice interview for AI-generated or coached answers, and longitudinal tracking across studies. Only interviews that pass length, depth and coverage scoring are billed, and failed sessions are re-fielded at no cost.

How do you test a recruitment integration before spending money?

Build against the sandbox first. The sandbox quickstart gives a staging API base, a sandbox MCP server and a dashboard where a staging API key is created. Study design and reporting run against staging data with AI respondents, so the create, plan, estimate and report calls can be exercised end to end, and a three-response synthetic batch produces a real report at no cost. The public TypeScript examples add a fixture mode that uses fictional data and makes no network calls at all. The docs recommend evaluating an agent against a fixed set of briefs and recording the study ID, request IDs, final plan, estimate and source references for each run before any live launch.

Which questions are worth asking a recruitment API provider?

Ask whether the API estimates, gates, interviews, scores, cites and sandboxes. Six questions cover it:

• Does the API return a study-level cost and timeline estimate before launch, or only per-participant pricing?
• Can a person set the spending rules, per launch or as a monthly budget, and does the platform enforce them, or is that left to the caller’s own code?
• Does the interview run through the same API, or do participants get handed to another tool?
• Are quality checks applied per interview, and does billing exclude interviews that fail them?
• Do results come back as structured findings with the source interview behind each quote?
• Can the integration be tested against a sandbox with no spend?

A provider that answers yes to all six supports an agent that recruits, interviews and reports without a person in the middle, except where a person should be: deciding what the agent may spend.

Sources

• Build a research agent, User Intuition docs
• Sandbox quickstart, User Intuition docs
• API introduction, User Intuition docs