Category: Ux | Written by Martin Möller

Ambivalent Personas with AI: How to Simulate Realistic User Behavior

Imagine designing for a user who always acts rationally, never contradicts themselves, and perfectly fits your product. Sounds fantastic—but unfortunately, this user doesn't exist. Let's call him "Perfect Paul": an AI-generated ideal persona, consistently efficient and rational. Paul has just one flaw: he's not real.

Real users behave unpredictably and contradict themselves. They emphasize privacy, yet spend hours scrolling Instagram and TikTok. They talk about sustainability but still purchase fast fashion. These inconsistencies make users human and design challenging.

So why do overly consistent AI personas like Paul fail, and how can you create ambivalent personas instead, which realistically reflect user behavior?

Why Overly Smooth AI Personas Fail

AI-generated personas often appear credible but deceptively flawless: "Sophie, 28, loves productivity apps and continuously optimizes her workflow." Yet in reality, people rarely behave this consistently. According to Pew Research, 80% of social media users are concerned about privacy but still use these services daily—a classic privacy paradox.

These contradictions aren't flaws; they're essential insights into deeper needs and motivations. Designs ignoring these ambivalences might seem ideal but often miss genuine user needs.

How to Use AI to Create Realistic Ambivalence

To generate personas capturing realistic contradictions, use a combination of actual segment data and targeted AI prompts.

Step 1: Gather Segment Data ("Data Shadows")

Start with real data sources like analytics, CRM, or user surveys. For example:

25% mobile-only users
35% aged over 45
80% rural residents
70% value privacy, yet 65% use social media daily

This data anchors your personas firmly in reality.

Step 2: Craft the Right Prompt

Create an AI prompt explicitly asking for contradictions, such as:

"Create a persona for a user over 45 who lives rurally, values privacy, but uses social media daily. Include at least two contradictions and a quote highlighting their ambivalence."

Step 3: Generate Your AI Persona

Using an AI tool like ChatGPT, you might get realistic personas like:

Persona: Lisa, 50, rural teacher
Goals: Stay connected with family and receive local news
Problems: Concerned about privacy but checks Facebook daily
Behavior: Uses privacy settings but frequently posts personal updates
Quote: "I know Facebook has privacy issues, but I just can't stop using it."

You immediately feel the tension between aspiration and reality.

Step 4: Add Detailed Nuances

Refine your persona with additional nuances:

Motivation: Values community connections over privacy fears
Contradiction: Uses a VPN for online banking but casually posts family photos

These details make personas tangible and encourage targeted discussions.

Limitations of this Method

Ambivalent personas have clear advantages: they're realistic, foster empathy, and are excellent for workshops and prototyping. Yet they're no replacement for real user research. Their weakness lies in being synthetic, based on data and AI—not direct observations. The risk is teams might mistakenly accept them as facts without validation.

Conclusion: AI Simulates Complexity but Doesn't Replace Reality

Ambivalent personas help you design more realistically by accounting for genuine user contradictions. Use them consciously—as an enhancement, not a replacement—for real user interviews, testing, and observations.

Forget Perfect Paul. Embrace ambivalence, use targeted prompts, and always validate with genuine user research. Your designs will become more relevant, effective, and most importantly: more human.

Try it Yourself: Prompt Template

Want to create your own ambivalent personas? Use this template:

Create exactly three realistic personas that collectively represent the segment below:

• 25 % of this audience are mobile-only users  
• 35 % are aged 45 +  
• 80 % live in rural areas  
• 70 % say they value privacy, yet 65 % use social media daily  

For **each** persona:
1. Ensure they match **all** four segment attributes (mobile-only, 45 +, rural, privacy-minded but daily social-media user).  
2. Include **at least two contradictions**:  
   – one between stated attitudes and actual behaviour  
   – one between goals and obstacles  
3. Provide:  
   • Name & brief demographic details  
   • Goals and motivations  
   • Pain points / obstacles  
   • Typical daily behaviours  
   • A 1-sentence quote illustrating their ambivalence  
4. Limit to 150 words per persona.  
Return the result as a Markdown table (one row per persona).

Afterward, validate your results with real users to confirm your assumptions.