Setting up an agentic research study takes under 5 minutes. But knowing what to test and how to frame it determines whether the output is useful. These templates provide the structure for each of the three research modes.
Template 1: Preference Check
Use when: You need to know which of two or more options people prefer and why.
Required Inputs
| Input | Example |
|---|---|
| Options to compare | ”Headline A: ‘Ship faster with AI’ vs. Headline B: ‘Build products people actually want‘“ |
| Target audience | ”Product managers at B2B SaaS companies with 50-500 employees” |
| Context (optional) | “These are homepage headlines for a developer tools company” |
| Sample size | 15 audio interviews |
What You Get Back
- Preference split with percentages (e.g., “67% prefer Headline B”)
- Top 3-5 themes driving the preference, ranked by prevalence
- Minority objections from those who disagreed, with real verbatim quotes
- Conditions under which preferences might change
- Data quality indicators
Example Prompt for Your AI Agent
“Run a preference check with 15 B2B SaaS product managers. Compare these two homepage headlines: ‘Ship faster with AI’ vs. ‘Build products people actually want.’ Context: these are for a developer tools company homepage. I need to know which one resonates more and why.”
When to Scale Up
If the preference split is close (55/45 or tighter), consider running a follow-up with 30 participants to increase confidence. If the split is decisive (70/30 or wider), 15 participants provide sufficient evidence for action.
Template 2: Claim Reaction
Use when: You need to know whether people believe a specific statement and what objections they have.
Required Inputs
| Input | Example |
|---|---|
| Claim to test | ”Our platform reduces customer research time by 95%“ |
| Target audience | ”VP-level consumer insights professionals at companies with $100M+ revenue” |
| Context (optional) | “This claim will appear on our website pricing page” |
| Sample size | 20 audio interviews |
What You Get Back
- Agreement rate with percentage (e.g., “45% find the claim credible”)
- Reasons for belief, ranked by prevalence
- Specific objections from skeptics, with real verbatim quotes
- Suggested modifications that would increase credibility
- Emotional reactions to the claim (trust, skepticism, interest)
Example Prompt for Your AI Agent
“Run a claim reaction study with 20 consumer insights VPs at $100M+ revenue companies. Test this claim: ‘Our platform reduces customer research time by 95%.’ This will appear on our pricing page. I need to know if it’s credible and what objections people have.”
Interpreting Results
Claims that test below 50% credibility need revision. Look at the objection themes — they often contain the specific language adjustments that would make the claim believable. A claim that “feels too good to be true” may become credible with added specificity (“reduces time from 6 weeks to 3 hours”).
Template 3: Message Test
Use when: You need to know whether a message is clear, what people think it promises, and how it makes them feel.
Required Inputs
| Input | Example |
|---|---|
| Message to test | ”Customer intelligence that compounds. Every conversation builds on the last.” |
| Target audience | ”Marketing directors at CPG brands” |
| Context (optional) | “This is tagline copy for an AI research platform” |
| Sample size | 15 audio interviews |
What You Get Back
- Clarity score (percentage who understood the intended meaning)
- What participants think the message promises (in their own words)
- Emotional associations (what the message makes them feel)
- Confusion points (specific words or phrases that cause friction)
- Suggested improvements from participant language
Example Prompt for Your AI Agent
“Run a message test with 15 marketing directors at CPG brands. Test this copy: ‘Customer intelligence that compounds. Every conversation builds on the last.’ Context: this is tagline copy for an AI research platform. I need to know if it’s clear, what people think it means, and how it makes them feel.”
Using Message Test Results
The most valuable output is often what participants think the message promises — in their own words. If their interpretation matches your intent, the message works. If there is a gap between intent and interpretation, their language often contains the fix. Participants describe what they want to hear; use their words.
Audience Targeting Guide
Getting the right audience is as important as the right question.
B2B Audiences
| Target | Targeting Criteria |
|---|---|
| Enterprise buyers | Job title + company size + industry |
| Product managers | Role + company type + team size |
| C-suite | Title level + company revenue + industry |
| Technical decision-makers | Role + technology stack + company size |
B2C Audiences
| Target | Targeting Criteria |
|---|---|
| Category purchasers | Purchase behavior + frequency + brand |
| Demographic segments | Age + location + income + household |
| Behavioral segments | Usage patterns + channel preferences |
| Lapsed customers | Previous purchase + time since last activity |
First-Party vs. Panel
Use your own customers when: you need feedback from people who know your product, testing retention messaging, understanding churn, or validating features for existing users.
Use the panel when: you need feedback from prospects, testing acquisition messaging, evaluating brand perception, or reaching audiences outside your customer base. User Intuition’s 4M+ vetted panel covers B2C and B2B audiences across 50+ languages.
Sample Size Decision Framework
| Decision Type | Recommended Sample | Cost (Audio, Professional) |
|---|---|---|
| Quick sanity check | 10 | $200 |
| Tactical validation | 15 | $300 |
| Confident directional finding | 20-30 | $400-$600 |
| Segment comparison (2 segments) | 30 (15 per segment) | $600 |
| Quantitative confidence | 50+ | $1,000+ |
| Multi-market (3 markets) | 45 (15 per market) | $900 |
Start with the smallest sample that answers your question. Agentic research makes follow-up studies cheap and fast — it is better to run two focused 15-person studies than one unfocused 30-person study.