User Research in the AI Era: Synthetic Users & Agents

TL;DR

• Synthetic Users: Simulate personas to test copy, flow, and basic usability before you bug real people.
• Research Agents: Autonomous agents that can browse 50 competitor sites and summarize their pricing/onboarding.
• Analysis: AI transcribes and codes interviews instantly. You focus on the insight.

The New Toolkit

1. Synthetic Users (Simulation)

Tools like Synthetic Users or UserEval let you spin up 50 AI participants.

• Prompt: "You are a busy marketing manager. Try to sign up for this product. What is confusing?"
• Result: Instant feedback on unclear copy or weird navigation.
• Caveat: It's a simulation. It generates "plausible" reactions, not real human emotion. Use it for obvious friction points.

2. Research Agents (Competitive Intel)

Don't screenshot 20 websites manually. Use an Agent.

• Task: "Go to these 10 competitor sites. Sign up for the free trial. Screenshot the onboarding flow. Summarize the 'Aha Moment' for each."
• Tool: Browserbase, MultiOn, or custom Puppeteer scripts with LLMs.

3. Automated Synthesis

• Dovetail / Maze AI: Automatically tags transcripts with themes (e.g., "Pricing Concern", "Feature Request").
• Your Job: Connect the dots. AI sees the pattern; you see the opportunity.

What AI Can't Do

• Empathy: It can't feel the user's frustration.
• Context: It doesn't know that the user is stressed because their boss is yelling at them.
• Novelty: Synthetic users are based on training data (the past). They might not "get" a truly new paradigm.

Do This Next

1. Run a Synthetic Test. Take a screenshot of your landing page. Ask Claude/ChatGPT (acting as a user) to "Roast this page. What is confusing?"
2. Try an automated transcription tool for your next call. Don't take notes; focus on the conversation.