Erika Hall, co-founder of Mule Design Studio and author of Just Enough Research and Conversational Design, recently posted on LinkedIn about synthetic user research. Her insights cut to the heart of why the distinction matters.

She says:

“I continue to see so much nonsense justified by some notional time savings. Nothing slows down design like lacking a clear, evidence-based decision-making process. … Having more tools or more data doesn’t speed anything up if you aren’t asking the right questions for the right reasons. Clarity is the real accelerant.”

With Market Research Institute International reporting that 62% of researchers surveyed recently saying “most” or “some” of their team is using AI, compared to 39% last year, we need to talk about AI’s role in the future of the discipline.

The synthetic user debate: replacing research or enhancing it?

Talking to human users to inform product design has been a guiding UX principle for decades. In 2014, the Nielsen Norman Group (NNG) told us that “UX without user research is not UX”.

As Greg Nudelman, an internationally recognised UX thought leader and author of the book UX for AI: A Framework for Designing AI-Driven Products, wrote in his article Navigating the Abyss: The Dark Side of Synthetic AI User Research Tools:

“The whole point of talking to people in the first place is to figure out what PEOPLE would find useful and joyful. What, for the love of the Matrix, is the point of talking to a robot to do your research instead of a real person?”

However, the debate about AI in research isn’t entirely one-sided. HBR research indicates that generative AI is offering businesses “unprecedented opportunities to understand customers, better assess the competitive environment, and push data-driven decision-making deep into their organizations.”

The key distinction lies in how AI tools are deployed and what role they play in the research ecosystem.

Companies like Synthetic Users* market their product with the tagline “user research without the users”, promising teams can quickly and accessibly simulate human behaviour without the burden of recruiting and interviewing real people. The appeal is obvious: instant insights, no scheduling headaches, unlimited availability.

But this convenience comes with significant risks that industry leaders are beginning to document.

* To avoid confusion, this article is a comparison of approaches, not a critique of any particular product using the synthetic user approach.

The problems with purely synthetic approaches

NNG’s extensive testing of synthetic user platforms revealed concerning patterns. For example, when researchers asked real users whether they had completed all parts of an online training program, real users honestly admitted they often started a course but didn’t finish it. Synthetic users claimed they completed all the courses, adding embellishments like “completing these courses was not just about acquiring certificates but truly understanding and integrating the learnings into my work.”

As NNG researchers noted, this highlights the fundamental limitation: synthetic users “have a tendency to want to please” and “do not always model human behaviour well”. They provide “shallow or overly favourable feedback” that fails to capture the complexity and messiness of real human behaviour.

The researchers went on to say that

“[h]uman behaviour is complex and context-dependent. Synthetic users miss this complexity. The synthetic users generated across multiple studies seem one-dimensional. They feel like a flat approximation of the experiences of tens of thousands of people, because they are.”

Perhaps most problematically, synthetic users readily generate long lists of needs without understanding priority. NNG saw this in their research:

“Real people care about some things more than others. Synthetic users seem to care about everything. This is not helpful for feature prioritization or persona creation.”

The expert criticism has been unequivocal.

In another LinkedIn post, Erika Hall emphasises the ethical dimension of synthetic user technology:

“Let’s put this whole ‘synthetic users’ thing to rest. It is unethical, indefensible, and also unnecessary, to create a product or service or policy that affects other people, without having conversations with representatives of those populations.”

Jared Spool, the influential UX design leader, reposted the comment, arguing that synthetic user technology is “absolutely the wrong direction for UX professionals to go.” He suggests focusing on tools that bring stakeholders closer to the experience of real users, not “some imaginary random poetry generator.”

In his “Navigating the Abyss” article cited earlier, Greg Nudelman said the Synthetic Users platform and tools like it should be “abandoned on the trash heap of history”.

Persona simulations: a research-grounded alternative

Persona simulations, or “persona sims” for short, represent a fundamentally different approach. They make human user research and data interactive by adding a thin AI layer on top. This distinction is crucial for understanding their value and appropriate application.

Screenshot of an excerpt from a conversation with a persona sim

A persona, regardless of how it’s created, is always a synthesis, a summary, a snapshot. It’s a compromise with the reality that humans aren’t always aware of, can’t access, or can’t manually process the vast amounts of data that represent audiences.

In traditional methods, source data is not retained in the resulting persona (or journey maps), for practical reasons. An AI-powered persona simulation, however, allows for access to much more of the underlying data, enabling traceability and explainability of the summary-level output. In other words, it maintains and makes sense of context.

Persona simulations support and enhance existing research practices. They don’t replace the fundamental human insights that drive them. They can also unearth patterns and insights that a human may not notice and address gaps in market understanding among individual professionals who work with them.

Research from companies implementing AI-enhanced research methods supports this approach.

Quoted in the HBR article cited earlier, former head of user experience research at WeightWatchers Wil Readinger found that AI tools meant researchers no longer had to choose between “the richer, contextualized data collected in an interview and the broader reach of a survey”.

This capability enables UX professionals to continuously engage with user perspectives without scheduling meetings, to generate and test hypotheses, and to refine designs before returning to real customers.

The “humans are the loop” philosophy

Synthetic-user approaches position AI as a replacement, following either “human in the loop” (where humans occasionally intervene) or “human on the loop” (where humans supervise) models.

Persona simulations reflect a “humans are the loop” philosophy, where human agency drives every decision. The AI provides acceleration and synthesis, but humans maintain control over strategy, interpretation, and application of insights.

This difference has practical implications. With synthetic users, teams risk making decisions based on artificially optimistic or generic feedback. With persona simulations, teams work with synthesised versions of real user voices and verbatim quotes, maintaining the context, nuance, and complexity that drive good design decisions.

A persona simulation should use technology to amplify, preserve, and extend human voices and behaviours and the details of their experiences.

This philosophy advocates that the goal isn’t to make things faster, it’s to make them better.

Closing the data loop

This debate exposes a limitation in how we handle research data today. Even the most thorough persona is a human synthesis of multiple research sources distilled into a static document. The final persona might reference whole research decks as sources, or include a few carefully selected quotes, but the rich connections between insights and source data are largely lost.

This creates a traceability problem. When a persona states that users are “frustrated with slow loading times”, it’s often difficult to trace that insight back to specific research moments. Which users said this? In what context? How many mentioned it unprompted versus in response to direct questions? These nuances get buried in the synthesis process.

AI-powered persona simulations offer a possibility for solving this long-standing challenge. Unlike traditional personas that create one-to-many relationships (one persona representing many data points), AI enables many-to-many mappings that would be untenable to maintain manually. A single insight can be traced back to multiple specific research artefacts, while individual data points can inform multiple personas and journey stages.

Consider the practical implications: when a persona simulation mentions a specific pain point, it can reference the exact interview moments, survey responses, or support tickets that contributed to that insight. When exploring a particular journey stage, teams can drill down to see the specific research evidence supporting each assumption or decision point.

This granular traceability becomes even more powerful when creating multiple personas or exploring different journey variations. Traditional approaches require choosing between depth and breadth: either create a few detailed personas or many shallow ones. AI synthesis allows for both multiple nuanced personas and traceable connections to their underlying research foundation.

Screenshot from a fictional example created with Urbina AI persona sims

This results in what we might call “living data”. Instead of insights becoming fossilised in static documents or disconnected from their broader context, they remain dynamic, contextualised, and query-able. Teams can explore different angles on the same research data, generate new personas for emerging use cases, or validate assumptions by examining their research foundations. And they can do this while maintaining clear audit trails back to original sources. Further, in an enterprise setting, they can theoretically trace back to a persistent data layer of research that AIs use as a shared source of ground truth.

This capability addresses one of the fundamental criticisms of traditional persona work: that insights become disconnected from their evidence base over time. With AI-powered simulations, the research foundation remains accessible and interrogable, making the insights more trustworthy and actionable while addressing bias concerns through maintained audit trails.

Finally, there’s the issue of simple loss of data and context. With traditional personas, an individual or small team (the researcher or researchers) or the persona deliverable (a document) is the interface to all the underlying data, offering limited access and coverage. The researchers make a choice of what to filter out for various reasons, the least desirable of which is simply having a summarised result that is readable in a few pages or slides. Simulations can offer access to much more of the underlying data, which could still be cleaned up and filtered by humans, but much more of the data stays accessible through an AI layer rather than a few pages of a presentation deck.

A comparison of traditional personas vs simulated personas

Persona sims as context engineering

Context engineering is an emerging field focused on providing AIs with the right background information for given tasks. It can also be thought of as a new wing of knowledge management. Some adherents prefer the term “context architecture”, as it is the AI version of content strategists’ age-old “right content, right audience, right time” adage.

Design requires what we might call “behavioural context” – understanding how people think, feel, and act throughout their journeys. Without this behavioural context, AIs can analyse data but struggle with providing grounded, useful, human-like responses.

Persona simulations are a type of context engineering, structuring psychological and situational insights to control AIs. They provide experiential context through journey stages and emotional states while maintaining behavioural consistency across interactions. This enables AIs to generate more contextually appropriate content.

Addressing AI bias in synthetic users or persona simulations

Bias is a silent killer in design decision-making.

Besides the effort to avoid human bias in decision-making, when AI models generate or simulate human behaviour, they can reflect both the explicit and hidden patterns within their training data. They can also reflect prejudice within the organisations that design the underlying model. Whether user researchers are building synthetic users or persona simulations, they must carefully manage bias.

Models often inherit broad cultural, linguistic, or behavioural biases from general AI training data. These biases are both ethical and practical concerns because they distort user insights, leading to skewed assumptions about priorities, language comprehension, or emotional responses.

For example, if an underlying model has been trained predominantly on English-speaking Western data, it may underrepresent how users from other linguistic or cultural backgrounds express needs or frustrations.

Persona simulations mitigate some of these risks by grounding responses in validated research data. Because their foundations can be traced to identifiable sources – interviews, analytics, and feedback – they allow for transparency and correction. Teams can audit where an assumption originates, see whether it reflects a small subset of participants, and decide whether to rebalance or expand the dataset. This traceability makes persona simulations inherently more accountable than synthetic users.

Still, bias can enter through omission. A persona simulation that reflects only a narrow slice of the audience will replicate those blind spots at scale. The countermeasure is diversity of input: blending qualitative and quantitative data, cross-checking AI-generated insights against real user sessions, and setting up systematic “bias reviews” as part of the research cycle.

Responsible teams treat bias management as part of their design governance. That means documenting data sources, versioning personas as new research arrives, and regularly testing the outputs against real-world feedback. Maintaining persona simulations becomes part of the organisation’s quality management framework, helping ensure that empathy and evidence remain aligned as both human and artificial perspectives evolve.

When synthetic might be “better than nothing”

Even critics acknowledge there are scenarios where synthetic approaches may have value. NNG suggests synthetic users could be useful “for desk research and generating hypotheses, but not for final decision-making” and recommends they be used to “help you prepare for research studies with real users.”

The HBR study found that 45% of market researchers were already using generative AI, with most employing it to analyse transcripts and data. However, the most successful applications focus on what the study authors call “filling existing gaps in market understanding by obtaining insights and evidence that aren’t available in conventional data” rather than replacing conventional research entirely.

The “better than nothing” argument applies when organisations face a choice between synthetic research and no user consideration at all. In these cases, even AI-generated insights might encourage a more user-centred mindset than pure assumption.

Extremely controversial will be the finding from the research team at WeightWatchers that found that participants are often more forthcoming when being interviewed by AI rather than by people.

However, this pragmatic approach requires careful handling. NNG warns:

“We’re extremely concerned about the likelihood that teams that start with synthetic research will get too comfortable and won’t move on to real-user research.”

Both synthetic users and persona simulations can serve a strategic function in organisations resistant to UX investment. Many teams struggle to secure budget for proper user research because stakeholders don’t understand its value. However, their approaches differ significantly.

Synthetic users might provide the illusion of research without the foundation, potentially undermining the prospect of ever securing investment in user understanding.

Persona simulations, when built from even modest research foundations, can demonstrate the power and utility of user research while highlighting gaps that require real user validation, thus demonstrating the ROI of further research investments. The key is transparency about limitations and maintaining the goal of expanding, not replacing, real user research.

The democratisation factor

Another significant advantage of persona simulations over traditional personas is democratisation of insights – the research itself.

Not every team member is trained in user research methodologies, but most design decisions benefit from user perspectives. Persona simulations, or AI layers more broadly, can provide a guided window into research findings that allows content designers, UX and user interface (UI) designers, and product managers to apply user insights without needing deep research expertise.

Persona simulations help scale the capacity and capability of user researchers, making research insights more accessible to the consumers of research with appropriate context and focus. They also drive consistency across teams and projects. Instead of each team member interpreting research differently, persona simulations provide a shared foundation for user understanding that improves collaboration and alignment.

Rather than certain projects having no access to research support or consistent methods, an AI-guided approach can maintain a minimum standard of research by extending the reach of user researchers’ work to more projects that might otherwise have been deprioritised.

Finally, democratisation can improve cohesion among outputs when different teams engage with shared personas, ensuring that marketing, product, and design teams are working from the same understanding of users and customers, rather than developing conflicting interpretations of who they’re serving.

Considerations for implementing synthetic users or persona simulations

For teams considering either approach, the implementation philosophy matters enormously. NNG’s guidance for synthetic users emphasises treating outputs “as hypotheses that need testing” and avoiding their use “as a replacement for real-user research.”

Persona simulations require similar discipline but offer more-robust foundations.

They work best when they’re

• built from multiple research sources rather than assumptions
• clearly linked to original research for traceability
• used to accelerate synthesis rather than replace insight gathering, and
• applied by teams that understand their limitations and maintain commitment to user validation.

The path forward

The distinction between synthetic users and persona simulations reflects deeper questions about AI’s role in human-centred design. Are we using AI to bypass or to deepen our understanding of real people? Are we trying to reduce complexity or to increase our knowledge?

HBR’s research suggests the answer lies in strategic deployment. The data shows that when it’s properly deployed, “AI offers firms unprecedented opportunities to understand and engage with customers, better assess the competitive environment, and extend data-driven decision making” deep into their organisations.

The key word is “properly”.

As Greg Nudelman notes:

“Sending a robot to the Bahamas may be cheaper and faster than going yourself – this is true! But what is the point of doing so exactly?”

The same logic applies to user research. The goal isn’t efficiency for its own sake, but deeper empathy and a more comprehensive understanding that leads to better design decisions.

The most promising approaches use AI to augment human capabilities rather than replace them.

Persona simulations can make research insights more interactive and accessible while maintaining the grounding in real human experience that makes UX research valuable. Organisations should use AI to enhance their connection with users, not to avoid it.

The choice between synthetic users and persona simulations isn’t about tools. It’s about values. Do we believe that understanding real people is central to creating meaningful experiences? The answer to that question will shape the future of research and experience design.

If you want to learn about ucLoops, Urbina Consulting’s AI-assisted methodology for designing and working with persona simulations, join our cohort training courses. Or contact us for information about our other persona and journey mapping services.