Everything about Synthetic Users, AI led UX Research. I learnt so you don't have to.

AI led User Interviews

I came across Synthetic Users while looking for faster ways to run research in telecom. Recruiting the right participants is genuinely hard. Prepaid subscribers are difficult to pin down.

My core userbase in B2B, Enterprise IT managers, they won't sit through a 45-minute session for a gift card. Research budgets are always fighting sprint timelines.

So I ran a live demo, dug into the platform, and tested it against real research problems. Here's everything I found.

4 mins

My Role

UX Researcher

Platform

SyntheticUsers.com

AI led UX Research by

Various LLMs + RAG

What is Synthetic Users?

Synthetic Users is an AI research platform that lets you build, interview, and test with AI-generated participants. No recruitment. No scheduling. No no-shows.

You configure the participants (they have personalities, yes). You run the study. You get research-grade output.

How the platform actually works

You start by creating participants. Not generic personas. Detailed ones.

Example participant: Maria Fernando Quiroz
Age / Location
29, Doha Qatar
Profession
Night Shift Clinical Lab Scientist
Break patterns
Scheduled breaks + micro-stretches tied to instrument cycles
Content engagement
Captioned micro-learning, quiet playlists, short news
Income
$18,000 USD/year

Each participant also gets a Big Five personality radar: openness, conscientiousness, extraversion, agreeableness, neuroticism. Every attribute is editable directly in the interface.

That level of specificity is what separates Synthetic Users from just prompting ChatGPT with a persona description.

Personality Graph example, because what distinguishes human from AI? Personalities with different sensitivities.

Then you pick a study type

Seven options, each built for a different research moment:

Study Type
Best for
Best for
Research Goal
Define what you want to learn, the platform structures the interview
Custom Script
Your questions, your flow
Concept Testing
Validate a design or idea against a specific audience
Problem Exploration
Understand pain points before you've built anything
Ethnographic Research
Explore how people live and work in context
Multi Concepts
Test multiple versions against the same audience
Prisma
Synthesis mode for comparing findings across studies

Each study runs 5 to 10 interviews automatically and stores every transcript.

The output is not what you'd expect

Transcripts read like real qualitative data. Participants respond in first person, express hesitation, and emotions. One transcript in the demo included:

"shifts slightly in seat"

You can ask follow-up questions to individual participants directly from the interface. The Report tab generates a structured executive summary automatically. Key themes, impact notes, variations across participants. The kind of writeup that normally takes a researcher days to produce after five interviews.

The Knowledge Graph maps everything as a visual node diagram:

Four Features worth knowing about syntheticusers.com

Iris, research assistant

An AI agent you can ask questions across all your studies. Ask it what themes participants mentioned about connectivity and it synthesises findings across every transcript on demand.

Prisma, multi-study synthesis

Runs comparative analysis across multiple concept tests. The demo used it to compare audio-only, video, and text-only content formats side by side and generate a single synthesis report.

RAG support

Upload your own data: support logs, past interview transcripts, market research. Synthetic Users pulls from your knowledge base before responding. Outputs become specific to your domain, not generic LLM guesses.

Model Shuffle v2

Routes each interview across multiple models simultaneously: GPT, Gemini, Claude, Mistral, Llama, Hermes. Reduces single-model bias and makes responses more realistic across participants.

Iris, research assistant

An AI agent you can ask questions across all your studies. Ask it what themes participants mentioned about connectivity and it synthesises findings across every transcript on demand.

Prisma, multi-study synthesis

Runs comparative analysis across multiple concept tests. The demo used it to compare audio-only, video, and text-only content formats side by side and generate a single synthesis report.

RAG support

Upload your own data: support logs, past interview transcripts, market research. Synthetic Users pulls from your knowledge base before responding. Outputs become specific to your domain, not generic LLM guesses.

Model Shuffle v2

Routes each interview across multiple models simultaneously: GPT, Gemini, Claude, Mistral, Llama, Hermes. Reduces single-model bias and makes responses more realistic across participants.

Now the Big Question. How Accurate is Synthetic User Research?

Synthetic Users measures something they call Synthetic/Organic Parity. They benchmark AI interviews against real ones across four weighted dimensions:

Real users still surface things Synthetic Users won't.

Treat 85 to 92% as a ceiling, not a promise.

How I'd use this in telecom

Running B2C and B2B in parallel means you're always stretched across segments. Synthetic Users lets you test assumptions about three very different audience types in the same week:

Segment
What to test
Prepaid students
Data limit frustration, self-service flows
Postpaid professionals
Family plan logic, upgrade triggers
SME procurement managers
Centralised billing, SLA priorities

Where this is going

This is the part I find most interesting. Right now Synthetic Users is a research acceleration tool. But the trajectory points somewhere more significant.

Real-time design feedback Imagine running a Synthetic Users study the moment a Figma prototype is ready. Not after two weeks of recruitment. The same day. Designers get directional feedback instantly.

Localisation and cultural testing at scale For teams building across MENA, Southeast Asia, or any region with hard-to-reach user populations, Synthetic Users could become the primary tool for cultural calibration. Test Arabic RTL flows with a panel built on regional behavioral data before you fly anyone in for in-person sessions.

Accessibility and edge case simulation Users with specific accessibility needs, users in low-connectivity regions, users navigating under cognitive load. Synthetic Users can model these scenarios on demand. No recruitment delays, no participant safety considerations, no ethical review board timelines.

My Two Cents Summary: AI Led User Research

I learnt so you don't have to.

Real Users are not replaceable and are experiencing nuanced situations given the rapidly growing AI powered world, they must be evolving with the frequent information sharing and their needs/wants are also changing.

Synthetic Users is not a replacement for talking to people.

It's a way to do more early-stage exploration and pivot fast

It's a sure way to do more early-stage exploration and pivot fast

My Two Cents Summary: AI Led User Research

I learnt so you don't have to.

Real Users are not replaceable and are experiencing nuanced situations given the rapidly growing AI powered world, they must be evolving with the frequent information sharing and their needs/wants are also changing.

Read the Next Blog

© 2026 Design Portfolio All rights reserved. Designed by Bisma Munawar