AI-powered user research platform

AI designs the study, runs live interviews and writes the insight report. It then connects to user behavior data to close the validation loop.

AI study designLive AI interviewsAI insight reportsValidation loop

AI study design

Describe your research question. AI asks about your goals and audience, drafts the interview guide and helps recruit the right participants.

The research assistant asks for context to sharpen your research goal
Set follow-up depth per question, and attach concept images, videos, and prototypes
Rehearse with simulated participants before publishing to check question order, length, and leading wording
Import a list or share a link; every participant gets a unique link
Set quotas by segment and watch progress in real time

What you getA rehearsed, version-locked interview guide and a participant list filled to your quotas.

app.heari.ai · Study design

AI research assistant

I want to know why new users stop using us in week one.

Your interview guide is ready: 4 topics, 8 questions, about 15 minutes. Ask me to adjust anything.

Tell me how to adjust the guide…

Interview guide

Published · v1Next: recruiting

Why new users drop off in week one

About 15 min4 topics8 questions

Research goal

Find when and why new users quit in week one, and which onboarding step to fix first.

Interview flow

Intro·Welcome and overview

2 min

Warm-up·Their last session

3 min

Core·When they wanted to quit

6 min

Wrap-up·Summary check

2 min

Core question

In your first week, when did you feel like stopping? What happened right then?

Live AI interviews

Run hundreds of one-on-one interviews at once. AI follows the guide and asks follow-up questions until it uncovers a specific experience.

Answer by text or voice; participants just open the link on their phone
Dozens of languages: participants answer in their own language and the AI follows up in the same one
Participants are told up front that an AI is interviewing them, give consent first, and can leave anytime
Every interview gets a quality score; low-effort and invalid answers are flagged
Tone and emotion analysis of voice answers flags hesitation, pauses and clear frustration, and researchers can replay the original audio
Researchers can listen live in the observation room and pause or take over; every human intervention is logged
Before the interview ends, participants confirm a summary of key points and can edit or reject it

What you getA complete record of every interview: each topic fully explored, with key points confirmed by the participant.

9:41

This interview is run by AI

About 15 minutes. Your answers are used only for this study, and your name is never shown when you're quoted.

There are no right answers. Just tell us what actually happened
Type or just speak
Skip any question and leave anytime

DeclineAgree

Run by a kids' coding course research team

9:41

3 / 8

Think back to the day of your child's trial class. How did you decide to hold off on enrolling?

My kid really liked it, but the moment class ended the teacher started talking price and didn't say a word about how my kid did.

Follow-upWhen class ended, what did you most want to know about your child?

0:12

TextVoice

Type your answer…

AI insight reports

AI surfaces findings backed by quotes and counterexamples. A researcher signs off before they reach the report.

Extract key points in bulk with custom analysis columns
Each finding shows how many participants it covers and which quotes it cites
Researchers review each one: accept, edit, or reject
Answers in every language in one report: quotes keep the original with a translation, and findings can be compared by market
Version-locked reports and export
Findings and quotes go into the knowledge base, so you can ask questions across studies later

What you getAn insight report signed off by a researcher, with verbatim quotes behind every finding.

app.heari.ai · Sample insight report

Low NPS on loan applications: what customers actually ran into

60Interviews520Supporting quotesv3 lockedApproved by researcher

Key findings

The proof-of-income step is the main reason applicants drop out midway

47%

044061118

I had to upload proof of income and didn't have it on me. Figured I'd do it later, then forgot.

Participant 044 · New customer · 03:05

Not seeing approval status was the top complaint among low scorers

40%

072089

Credit limits and interest rates weren't the main issue

18%

101

Theme distribution

Gathering documents

No visibility into approval status

Too many steps

Credit limits and rates

Customer service

The same study, in days instead of weeks

A 30-interview study, step by step. Your researchers spend their time on findings, not on scheduling and transcripts.

1Recruiting and scheduling

Traditional interviews

Contacting participants one by one and finding times, usually 1–2 weeks

Send a link; participants start whenever they like, no scheduling

2Running interviews

Traditional interviews

A researcher runs 3–4 a day, so 30 take about 2 weeks

Hundreds at once, in any time zone

3Transcription and notes

Traditional interviews

Manual transcription and note-taking, about 1 week

Automatic transcripts, organized by theme

4Analysis and reports

Traditional interviews

Researchers synthesize and write the report, about 1 week

Report ready in hours; researchers review and confirm

Connect behavior data and close the validation loop

With ThinkingAI Agentic Engine, every step—from choosing participants to validating findings—ties back to real behavior data.

Select participants by behavior

Target abandoned carts, lapsed members and more. No list exports.

Bring behavior context into interviews

AI knows what each participant bought and where they stopped, so its questions are more targeted.

Said vs. did

See what participants said next to what they actually did, and find where the two don't match.

Write findings back as user tags

Findings become user tags that ops and product teams can segment by.

Validate with business results

After launch, check each finding against behavior data: confirmed, unproven or overturned.

Finding validationLow-sugar yogurt repeat-purchase study

What the shopper said

"It tastes good. I just haven't been buying much yogurt lately."

What the behavior data shows

No repeat purchase within 30 days of the first; bought a different sugar-free yogurt twice in that time.

Too sweet for breakfast, which hurts repeat purchasesConfirmed

The ad suggests fitness, not breakfastUnproven

Price is the main reason for not buying againOverturned

heari also works as a standalone tool without behavior data.

Let your AI agents run user interviews too

Codex, Claude Code or your own AI agent can call heari directly through MCP or the CLI.

We're picking the lead selling point for the new headphones next week. Ask 120 current owners whether they care more about noise cancelling or comfort, and get me findings by tomorrow.

heari.create_study(goal: "Pick the lead selling point for new headphones", audience: "owners_120.csv")

heari.launch_interviews(study: "st_8f2k", count: 120)

✓ Study created, 120 invitations sent

heari.get_findings(study: "st_8f2k")

112 interviews completed. Most owners care more about comfort: 69 of 112 brought it up unprompted—e.g., "I wear them four or five hours a day, so comfort matters most to me."083

# Install and sign in
$ npm install -g @heari/cli
$ heari login

# Create a study, import participants, start interviewing
$ heari studies create --goal "Lead with comfort or noise cancelling?"
$ heari interviews launch st_8f2k --list owners_120.csv --count 120

# Get findings, each with supporting quotes
$ heari findings st_8f2k --format json
{ "insight": "Comfort matters more than noise cancelling",
  "participants": 69, "quotes": ["u083", "u097", "u019"] }

// Codex, Claude Code, or any MCP-compatible client
{
  "mcpServers": {
    "heari": {
      "command": "npx",
      "args": ["-y", "@heari/mcp"],
      "env": { "HEARI_API_KEY": "hk_live_..." }
    }
  }
}

Tools your agent can call

create_studyCreate a study and draft the guide

launch_interviewsSend invitations and start interviews

get_study_statusCheck progress and interview quality

get_findingsGet findings with supporting quotes

search_quotesSearch quotes by topic

ask_knowledge_baseAsk across studies

Structured output: findings come back as JSON with evidence IDs.
Human in the loop: researchers can approve invitations and insights.
Same permissions and quota: interviews your agent starts count toward your plan's quota.

Run the full workflow on a real study

From study design and live AI interviews to the insight report, we'll help you run a pilot study in heari.

Book a demo See use cases