{"id":4953,"date":"2025-07-08T02:23:37","date_gmt":"2025-07-07T17:23:37","guid":{"rendered":"https:\/\/uism.co.jp\/?p=4953"},"modified":"2026-04-13T14:07:13","modified_gmt":"2026-04-13T05:07:13","slug":"synthetic-users-how-far-can-ai-go-in-ux-research","status":"publish","type":"post","link":"https:\/\/uism.co.jp\/en\/blog\/synthetic-users-how-far-can-ai-go-in-ux-research\/","title":{"rendered":"Synthetic Users: How Far can AI go in UX Research?\u00a0"},"content":{"rendered":"\n

Introduction <\/span><\/h2>\n\n\n\n

As UX researchers, we depend on real people to understand their needs, frustrations, and emotional journeys. <\/p>\n\n\n\n

But what if we could simulate those people instead? <\/p>\n\n\n\n

With generative AI evolving so quickly, there’s been growing interest in using “synthetic users\u201d<\/span><\/strong>, meaning AI-generated personas that respond to research prompts as if they were real. These virtual users are now being tested as a way to spark hypotheses and imagine early user journeys. <\/p>\n\n\n\n

Still, can they truly capture the emotional depth and complexity of human experience? <\/p>\n\n\n\n

To explore this, I ran a small comparative study using both real and synthetic users. The topic I chose was emotionally and socially nuanced: volunteering among people in their 20s in Japan<\/em>. This blog shares what I found, where synthetic users performed well, where they fell short, and how they might fit into the research process going forward. Let\u2019s dive in. <\/p>\n\n\n\n

Research Design <\/span><\/h2>\n\n\n\n

This study was conducted in two phases, with matched user profiles across both groups. <\/p>\n\n\n\n

Phase 1: Real Users (P01\u2013P04)<\/strong> <\/span><\/h3>\n\n\n\n
\n
    \n
  • Four individuals aged 20\u201329 <\/li>\n\n\n\n
  • Volunteering at least monthly for over three months <\/li>\n\n\n\n
  • 60-minute remote interviews focused on motivations, reasons for continuation, and emotional barriers <\/li>\n<\/ul>\n<\/div><\/div>\n\n\n\n

    Phase 2: Synthetic Users (P05\u2013P08)<\/strong> <\/span><\/h3>\n\n\n\n

    Synthetic users were generated using GPT-4, based on the same criteria used to recruit real users (age, volunteer type, frequency, motivation). The same interview guide was used to simulate 60-minute conversations. <\/p>\n\n\n\n

    \n\n\n\n
    \n

    Workflow: <\/strong><\/p>\n\n\n\n

    \n
    \n
    <\/div>Profile generation<\/strong> <\/span><\/div>
    \n

    Used real user screener criteria (e.g., age, activity type) to create four matched personas.<\/p>\n<\/div><\/div><\/div>\n\n\n\n

    <\/div>Transcript generation<\/strong><\/span><\/div>
    \n

    For each profile, a 60-minute equivalent interview transcript was generated using the same interview guide as with real users.  <\/p>\n<\/div><\/div><\/div>\n\n\n\n

    <\/div>Analysis<\/strong> <\/span><\/div>
    \n

    Used the same note-taking template and analytic framework as with real users. <\/p>\n<\/div><\/div><\/div>\n<\/div><\/div>\n<\/div><\/div>\n<\/div>\n\n\n\n

    Findings <\/span><\/h2>\n\n\n\n

    What AI got right:  <\/strong><\/p>\n\n\n\n

      \n
    • Clear and structured storytelling<\/strong>:
      AI created organized, logical narratives that reflected general motivations and barriers. <\/li>\n\n\n\n
    • Hypothesis generation<\/strong>:
      Surfaced potential UX pain points, making it useful for anticipating potential usability issues. <\/li>\n\n\n\n
    • Consistent persona: <\/strong>
      When a profile is predefined, AI performs well in responding consistently within the given character. <\/li>\n\n\n\n
    • Fast and efficient:<\/strong>
      Instantly produced responses, making it a quick tool for testing ideas and refining research materials. <\/li>\n<\/ul>\n\n\n\n

      Where AI missed the mark:  <\/strong><\/p>\n\n\n\n

        \n
      • Lack of depth and emotional change:<\/strong>
        AI responses were smooth but lacked the hesitation, internal conflict, and emotional transitions that characterize real experiences. <\/strong>Emotional changes were either missing or happened too quickly to feel believable. <\/li>\n\n\n\n
      • Abstract and idealized episodes:<\/strong>
        Stories felt polished and simplified, often glossing over struggles or negative emotions. This made them sound more like summaries than lived experiences.  <\/li>\n<\/ul>\n\n\n\n

        Comparing User Journeys <\/span><\/h2>\n\n\n\n

        When mapping out users’ journeys before volunteering participation, key differences emerged. <\/p>\n\n\n\n

        Real users<\/strong> often discovered volunteering through friends or community ties. Their early emotions included hesitation, uncertainty, and concerns about fitting in. Over time, these softened as they gained real-world experience. <\/p>\n\n\n\n

        Synthetic users<\/strong>, by contrast, expressed interest from the start, actively collected information in advance, and reported that the application process itself was a key pain point.   <\/p>\n\n\n\n

        Phase<\/strong> <\/td>Real Users<\/strong> <\/span><\/td>Synthetic Users<\/strong> <\/span><\/td><\/tr>
        Awareness<\/strong> <\/td><\/path><\/svg><\/span> Heard from friends or neighbors 
        <\/path><\/svg><\/span> Unsure if the activity was trustworthy 
        <\/path><\/svg><\/span> Low online visibility <\/td><\/path><\/svg><\/span> Found via online search
        <\/path><\/svg><\/span> Initially motivated and curious 
        <\/path><\/svg><\/span> Some info was vague or hard to compare <\/td><\/tr>
        Consideration<\/strong> <\/td>        		<\/path><\/svg><\/span> Searched YouTube\/blogs\/social media 
        <\/path><\/svg><\/span> Anxious due to fear of judgment or unclear expectations
        <\/path><\/svg><\/span> Info felt vague or outdated <\/td><\/path><\/svg><\/span> Read blog posts and FAQs 
        <\/path><\/svg><\/span> Mild concerns quickly resolved 
        <\/path><\/svg><\/span> Noted formal or unclear application process <\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

        These differences matter. Real users navigated emotional uncertainty, influenced by social cues. Synthetic users followed a cleaner path, acting more independently. This gap can lead to very different design implications, and highlights how AI can help surface user types that may be underrepresented or harder to reach in traditional research. <\/p>\n\n\n\n

        Comparison in Response <\/span><\/h2>\n\n\n\n

        Let\u2019s compare a few answers from each group. These side-by-side comparisons show how synthetic users can echo real perspectives, but tend to lack the emotional complexity and raw nuance that come through in real interviews. <\/p>\n\n\n\n

         Q. How did you feel on the first day of volunteering? <\/h3>\n\n\n\n
        \"Profile<\/figure>
        \n
        \n

        Real User (P04)<\/strong> <\/span>
        “I was tense the whole time at first, but it turned out to be really fun. Everyone was welcoming and easy to talk to. I felt like they saw me as a grandchild. That made me feel safe. It reminded me how important human connection is, and I was able to enjoy myself while feeling at ease.”<\/em> <\/p>\n<\/blockquote>\n<\/div><\/div>\n\n\n\n

        \n

        <\/path><\/svg><\/span> Emotionally layered, relationally anchored, shows change over time <\/strong><\/span><\/p>\n<\/div><\/div>\n\n\n\n

        \"A<\/figure>
        \n
        \n

        Synthetic User (P07)<\/strong> <\/span>
        “I was very nervous, but a staff member said, \u2018Just sit next to us and you\u2019ll be fine,\u2019 and that helped me relax.”<\/em> <\/p>\n<\/blockquote>\n<\/div><\/div>\n\n\n\n

        \n

        <\/path><\/svg><\/span> Emotionally flat, quickly resolved, lacks internal development <\/strong><\/span><\/p>\n<\/div><\/div>\n\n\n\n

        Q. What keeps you continuing volunteer work?  <\/h3>\n\n\n\n
        \"Profile<\/figure>
        \n
        \n

        Real User (P01)<\/strong> <\/span>
        “If I quit, those kids might not have anyone to make sure they get a meal. Sometimes children who are emotionally overwhelmed come to eat, and I see them gradually start smiling.”<\/em> <\/p>\n<\/blockquote>\n<\/div><\/div>\n\n\n\n

        \n

        <\/path><\/svg><\/span> Morally driven, emotionally specific, socially embedded <\/strong><\/span><\/p>\n<\/div><\/div>\n\n\n\n

        \"A<\/figure>
        \n
        \n

        Synthetic User (P08)<\/strong> <\/span>
        “What keeps me going is the feeling that my presence is needed. It\u2019s the only time I\u2019m not expected to give anything back.”<\/em> <\/p>\n<\/blockquote>\n<\/div><\/div>\n\n\n\n

        \n

        <\/path><\/svg><\/span> Idealized, internally focused, summarizes value rather than narrating experience <\/strong><\/span><\/p>\n<\/div><\/div>\n\n\n\n

        Future use of AI in UX Research <\/span><\/h2>\n\n\n\n

        AI can be a powerful support tool in the early stages of UX research, especially planning and idea development. Here is where synthetic users can add value:  <\/p>\n\n\n\n

        \n
        \n
        \"\"<\/figure>
        \n

        Generating hypotheses<\/strong>: <\/span>
        AI can simulate realistic user types and suggest motivations, helping researchers form and prioritize early ideas. This makes it useful for exploring a wide range of behavioral patterns and expand the scope for screener and segment design. <\/p>\n<\/div><\/div>\n\n\n\n

        \"\"<\/figure>
        \n

        Refining research tools<\/strong>: <\/span>
        AI responses are useful for fine-tuning screeners and interview guides before talking to actual participants. <\/p>\n\n\n\n

        <\/p>\n<\/div><\/div>\n\n\n\n

        \"\"<\/figure>
        \n

        Scenario simulation<\/strong>: <\/span>
        AI can roleplay user flows and decision-making processes quickly and consistently, making it easier to visualize potential user journeys.  <\/p>\n<\/div><\/div>\n<\/div><\/div>\n<\/div><\/div>\n\n\n\n

        That said, AI still struggles to capture emotional nuances and the complexity of social dynamics. It should not replace real users, especially when gathering insights or analyzing behavior, but rather be used as a complementary method to enhance research. <\/p>\n\n\n\n

        Conclusion <\/span><\/h2>\n\n\n\n

        This case study shows that synthetic users can support UX researchers in the early stages of research by helping structure ideas and simulate user scenarios. However, when it comes to understanding how people truly feel, act, and make decisions, real users are still irreplaceable.  <\/p>\n\n\n\n

        Only real users can reveal the deeper \u201cwhy\u201d behind behavior. As researchers, our job is to dig beneath the surface, asking deeper questions, uncovering the complex emotions behind every action, and understanding the social context that AI is still far from reaching.  We welcome new tools like ChatGPT for what they offer. But when we want to understand how people truly feel, nothing compares to hearing it from a real person, in their own words. <\/p>\n\n\n\n

        At Uism, we\u2019re always exploring how emerging tools can support more meaningful research without losing sight of what matters most: real human experience<\/em>.  <\/p>\n\n\n\n

        Curious how synthetic users might fit into your process? Let\u2019s talk. <\/p>\n","protected":false},"excerpt":{"rendered":"

        Can AI-generated synthetic users replace real ones? A UX case study with Japanese users in their 20s explores the potential and limits of AI in user research.<\/p>\n","protected":false},"author":29,"featured_media":4940,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_locale":"en_US","_original_post":"https:\/\/uism.co.jp\/?p=4936","footnotes":"","wp-seo-meta-description":"Can AI-generated synthetic users replace real ones? A UX case study with Japanese users in their 20s explores the potential and limits of AI in user research.","wp-seo-meta-robots":[]},"categories":[374],"tags":[165,284,312,319,321],"class_list":{"0":"post-4953","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-ux-thinking","8":"tag-ux-research","9":"tag-humancentereddesign","10":"tag-generativeai","11":"tag-syntheticusers","12":"tag-aipersona","13":"en-US","14":"c-entry"},"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/uism.co.jp\/wp-json\/wp\/v2\/posts\/4953","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/uism.co.jp\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/uism.co.jp\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/uism.co.jp\/wp-json\/wp\/v2\/users\/29"}],"replies":[{"embeddable":true,"href":"https:\/\/uism.co.jp\/wp-json\/wp\/v2\/comments?post=4953"}],"version-history":[{"count":16,"href":"https:\/\/uism.co.jp\/wp-json\/wp\/v2\/posts\/4953\/revisions"}],"predecessor-version":[{"id":9382,"href":"https:\/\/uism.co.jp\/wp-json\/wp\/v2\/posts\/4953\/revisions\/9382"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/uism.co.jp\/wp-json\/wp\/v2\/media\/4940"}],"wp:attachment":[{"href":"https:\/\/uism.co.jp\/wp-json\/wp\/v2\/media?parent=4953"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/uism.co.jp\/wp-json\/wp\/v2\/categories?post=4953"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/uism.co.jp\/wp-json\/wp\/v2\/tags?post=4953"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}