Dec 25,2025 
You need feedback on your product, your copy, or your onboarding flow. You could recruit testers, schedule calls, and wait for responses. Or you could build a synthetic user inside Claude and get a synthetic perspective in minutes.
A synthetic user is a persona you create through prompting. You describe someone with specific traits, goals, frustrations, and context. Then you ask Claude to respond as that person would. The idea is simple. The execution takes a bit of care.
This guide walks you through setting up synthetic users in Claude, writing prompts that produce useful responses, and knowing when the approach works well and when it falls short.
What Claude Offers for Persona Building
Claude Projects gives you a dedicated workspace for this kind of work. You can upload documents, set custom instructions, and maintain context across multiple conversations. This matters because a synthetic user needs consistency. If Claude forgets who your persona is mid-conversation, the responses lose coherence.
The 200,000-token context window helps here. You can feed Claude background materials like user research summaries, brand guidelines, or product documentation. You can also write detailed instructions about how you want Claude to respond, including the persona it should embody.
A regular Claude chat treats each conversation as a fresh start. Projects let you build something that holds together over time. Your synthetic user can have a backstory, preferences, and a perspective that persists across sessions.
Writing a Prompt That Builds a Real-Feeling Persona
The quality of your synthetic user depends entirely on how well you describe them. Vague prompts produce generic responses. Specific prompts produce something you can actually use.
Start with the basics. Give your persona a name, age, occupation, and living situation. Then go further. What does their typical day look like? What are they trying to accomplish? What frustrates them? What do they value?
Here is an example prompt you might use:
You are Maya, a 34-year-old project manager at a mid-sized software company in Austin. You manage a team of 6 developers and spend most of your day in meetings, Slack, and Jira. You’re skeptical of new tools because you’ve seen too many fail during implementation. You care about efficiency but you also care about your team’s wellbeing. You’re not technical yourself but you understand technical constraints. You have 2 kids under 5 and very little patience for things that waste your time.
Notice the specific details. The team size. The tools she uses. Her skepticism and where it comes from. Her competing priorities. Her life outside work. All of this gives Claude material to draw from when responding.
Asking Questions That Surface Useful Feedback
Once you have your persona established, the next step is asking the right questions. Open-ended prompts tend to work better than yes or no questions.
Instead of asking “Would you use this feature?” try asking “Walk me through how you would decide if this feature is worth trying.” The second question forces Claude to reason through the decision the way a real person might.
You can also put your synthetic user in situations. Show them a screenshot of your interface and ask what they notice first. Describe a scenario where they encounter your product and ask how they would react. Give them a problem and ask how they would try to solve it.
The responses you get will feel more grounded if you ask Claude to think out loud. Ask for doubts, hesitations, and the questions Maya would want answered before committing.
Where Prompt-Based Synthetic Users Fall Short
This approach has limits worth understanding. Claude generates responses by combining attributes and producing outputs based on patterns in its training data. When you ask it to play Maya, it draws from what it knows about project managers, working parents, and skeptical software users. The result is a plausible composite, not a real person.
The persona has no genuine history. It has no memories of past disappointments or wins. It arrives at each conversation fresh, even within a project, without the accumulated context that shapes how a real person would respond. Claude can roleplay consistency, but the underlying model does not actually remember being Maya last week.
This means prompt-based synthetic users work well for exploring possibilities and generating hypotheses. They work less well for predicting exactly how a specific audience segment will behave.
Where Evelance Takes This Further
Evelance approaches AI user research differently. Instead of generating personas through attribute combinations and statistical patterns, Evelance has built predictive personas that have absorbed real behavioral data.
The difference shows in the results. In testing, Evelance predicted how real people would respond with 89.78% accuracy. That number matters if you’re making decisions based on the feedback you receive.
Evelance maintains over 2 million personas. Each one carries identity, memory, and situational context. The personas know who they are before your test begins. They hold opinions and arrive with a full day already behind them.
You can filter by demographics, professions, and behaviors. You can measure responses across 12 psychological metrics. The system gives you something closer to real research without the time and cost of recruiting participants.
If you need quick exploration and hypothesis generation, Claude synthetic users serve that purpose well. If you need predictive accuracy for business decisions, Evelance offers a path to get there.
Making the Most of Either Approach
Whatever tool you use, the value comes from asking good questions and interpreting responses carefully. A synthetic user can surface blind spots in your thinking, suggest concerns you had not considered, and help you practice conversations before you have them with real people.
Treat the feedback as input, not gospel. Use it to refine your approach, then validate with actual users when the stakes are high.
Building synthetic users takes practice. Your prompts will get better as you learn what produces useful responses. Keep experimenting, keep refining, and pay attention to when the feedback helps and when it feels hollow. That awareness will serve you well.