Human Signal: The Data Type Your AI Agent Doesn't Have

Part 3 of the series: The Customer Truth Layer for AI Agents

Your AI agent works with a well-defined set of data types every day. JSON objects from API calls. Vector embeddings from your knowledge base. Structured outputs from function calls. Token streams from other language models. Each data type has a clear schema, predictable format, and established patterns for consumption.

Now consider the most valuable signal for customer-facing decisions: what real people actually think about your product, your messaging, your positioning. Where is the data type for that?

In most agent architectures, it does not exist. The agent has structured access to everything except the one input that matters most for customer-facing decisions — grounded, quantified feedback from the humans it is trying to serve. It can query a database, call an API, search a vector store, and generate text. It cannot ask a real person what they think and receive a structured answer.

Human Signal fills that evidence gap through real participant research. The current interface returns interviews and cited study reports that agents can retrieve and analyze.

What Human Signal Is (And Is Not)?

Human Signal means evidence from real people reacting to a specific question, concept, or experience. It is useful when an agent needs information that is absent from public sources and training data.

In User Intuition, this evidence comes from the ordinary study workflow: an approved plan, real participant interviews, and a report with references. The phrase describes the value of that evidence. It is not a separate API mode or a guarantee that every finding arrives as a typed numerical field.

An interview can explain why a person preferred an option, what made a claim hard to believe, or where the product promise was misunderstood. The agent should preserve those reasons and their source, including reactions that conflict with the main interpretation.

What Are the Three Study Modes?

Preference checks, claim reactions, and message tests are three research objectives. They use the same planning and recruitment tools rather than three special API modes.

Preference checks

Compare the options with the intended audience. Ask what each option means before asking for a preference, and allow participants to reject all the options. A tally requires an explicit, consistent question and a stated denominator; the reasons behind a choice are often more useful than a small numerical lead.

Claim reactions

Explore what people believe the claim promises, why they trust or doubt it, and what evidence would change their view. Distinguish a participant’s reaction from an independent verification that the claim is true.

Message tests

Ask participants to explain the message in their own words. Look for missing context, confusing language, and unintended promises. Connect proposed edits to interview evidence rather than treating a generated clarity score as a platform measurement.

Anatomy of a Human Signal Result

get_study_report returns report text, references, interview count, stale status, and timestamps in JSON. Use generate_report when analysis is needed, and get_interview to verify supporting messages and recording links. Preference shares, credibility scores, and ranked themes are not guaranteed typed fields in this response.

A useful agent readout adds an interpretation, the sources supporting it, exceptions, and the limits of the sample. If an application extracts themes or computes preferences, label that as application analysis and verify it against the interviews. The underlying report should remain available so a reviewer can inspect the evidence.

Do not use a small qualitative sample as a precise estimate of an entire market. Include who participated, when the interviews took place, and which perspectives may be missing.

From Episodic to Ambient

A recurring research practice keeps evidence available for future decisions. Save study IDs, briefs, audience information, reports, and source references so a later agent can understand what was actually learned.

Agents can list studies, retrieve selected study reports, and inspect supporting interviews. They can compare that retrieved evidence in their own context. Direct cross-study Intelligence Hub search is available in the dashboard; the current public API and MCP catalog do not expose Hub search, Hub sessions, or presentation generation.

Reuse works best when the earlier question and audience match the new decision. If the product or market has changed, old evidence may help design a better study without settling the question on its own.

Human Signal in Agentic Market Research

Create a metadata draft with create_study, then send the research brief through customize_study. Relay any planning questions to the user. Retrieve the persisted plan with get_study and obtain approval before recruitment. The user must choose panel or BYOP explicitly.

Once evidence is available, the agent can use it to refine a message, challenge an assumption, or propose a follow-up study. That judgment should remain traceable to the retrieved sources.

Give your agent access to participant research →

Series: The Customer Truth Layer for AI Agents

1. Your AI Agent Is Confidently Wrong About Your Customers
2. The Agent Stack Is Missing a Layer: Customer Truth
3. Human Signal: The Data Type Your AI Agent Doesn’t Have (you are here)
4. Why Synthetic Panels Can’t Replace Real Customers (And What Can)
5. Compound Intelligence: Why Your Agent Gets Smarter With Every Conversation
6. Building the Customer Truth Layer: A Technical Guide