Personas: a modelling device and the evidence problem behind it

A persona is the studio's most portable artefact, carrying a name, a job, a frustration and a photograph from research notes into the deck and the client meeting. That form makes it persuasive and dangerous. This unit makes two claims. The defensible job of a persona is to hold a team's attention on one observed user, keeping an assumption visible long enough to argue with, and it cannot carry the evidentiary weight its form invites: it is a mnemonic built from observation, not a finding.

Treated as evidence, a persona closes a discussion. Treated as a device, it opens one, because a designer can check it against the observations it came from. This is the seventh of ten units; it takes the question of whom the design is for and hands unit eight the service behind it.

The method's own case for the persona is attention, not truth

The persona entered interaction design as part of an argument about whom software serves. Cooper (1999) argued that the industry shapes interactive products around the machine's convenience rather than the user's. He proposed goal-directed design, in which an archetypal user anchors decisions about behaviour. That user is the design's point of view, a character whose goals a designer can consult when a decision is contested.

Cooper, Reimann, Cronin and Noessel (2014) set out how personas are built from research and used to drive scenarios and interaction decisions, the account most teams inherit. Pruitt and Adlin (2006) added a lifecycle of planning, creating, communicating and using personas, complementing scenarios, walkthroughs and user testing, and Pruitt and Grudin (2003) supplied the account of the method's grip: from three years of use, they argue, personas engage a team more than scenarios alone and focus attention on aspects other methods miss.

A persona cannot be falsified, so it carries no demonstrable validity

Chapman and Milham (2006) make three arguments: it is difficult to determine how many users, if any, a persona represents, so its relevance is unknown; a persona cannot be adequately verified or falsified, and so has no demonstrable validity; and it generates political conflict while undermining a researcher's ability to settle questions with data. Until those issues are resolved, they conclude, personas should not be treated as a means of communicating data.

A persona can be a fair summary of observations. It cannot be evidence that the summary is typical, because nothing in its construction states what would count as the persona being wrong.

Teams use personas to talk, and the benefits case is consensus

Friess (2012) analysed decision-making sessions at a firm that invests heavily in personas and found them making relatively few appearances in the designers' language, with advocates of the method talking in designer-centric terms. Matthews, Judge and Whittaker (2012) studied experienced practitioners and found they used personas almost exclusively for communication rather than design, judging them abstract, impersonal, misleading and distracting. Personas cannot replace immersion in user data.

Miaskiewicz and Kozar (2011) used a Delphi process to aggregate expert opinion on the benefits claimed for personas. Their panel reached consensus that the suggested advantages have substance, including communication, assumption checking and decision support.

What personas reliably deliver is communication and engagement; what would justify their status, the anchoring of design decisions, is what observational evidence struggles to find.

The one experiment measures identification, not decision quality

The strongest contrary evidence is controlled. Salminen, Jung, Chowdhury, Sengun and Jansen (2020) ran a 34-participant experiment. They compared a persona system and an analytics system on identical user data, for identifying which user a behaviour came from. Personas gave faster completion and significantly higher identification accuracy, with further advantages in learnability and consistency.

The study shows that a persona-shaped presentation of data beats a dashboard at retrieving what is already there. That is a claim about the interface to evidence, not about the evidence.

The organisation failed that project, not the method

Ronkko, Hellman, Kilander and Dittrich (2004) report a project in which personas were never used: mass-market software for mobile devices, where the method was chosen to settle an internal disagreement about who the user was. The designers, management and marketing all agreed on the idea, yet it was never implemented, because several clients shared the same development, secrecy limited what could be said about users, and naming a group of end users carried consequences for future sales, so the team closed the experiment down. The case is often read as evidence against the method, but it separates two failures: a method never given the conditions to work, and an organisation that would have defeated almost any user-centred instrument.

Every detail in a persona informs and misleads at once

If a persona is a device, its details are design decisions with consequences. Salminen, Jung, An, Kwak, Nielsen and Jansen (2019) found that contextual photographs gave users significantly more information about a persona, while photographs of multiple people produced confusion and lowered informativeness. Turner and Turner (2011) argue that the form itself carries assumptions about who the user is. Salminen, Jung, Kamel, Santos and Jansen (2022) rated 1,000 generated faces through 6,812 crowd judgements, at 90 per cent medium quality or better, then tested them with 496 participants: the artificial pictures did not lower authenticity, clarity, empathy or willingness to use. Concreteness informs and misleads in equal measure.

The method is moving to generated personas and checkable provenance

The literature is shifting from hand-built characters to data-driven construction. Salminen, Guan, Jung, Chowdhury and Jansen (2020) reviewed 49 studies of quantitative persona creation from 2005 to 2019 and argued that the field must share datasets, code and algorithms to mature. Nielsen (2019) makes the condition explicit: personas work when grounded in data and used as a shared frame rather than as a portrait of a user. As personas become generated, what observations they rest on becomes answerable. Gu, Chandrasegaran and Lloyd (2025) compared an interactive synthetic-user chatbot with a written persona summary and found no significant gain in empathy or in the number or variety of ideas.

Contested question: is a persona evidence or rhetoric?

One position holds that a persona is not evidence at all. Chapman and Milham (2006) argue its central claim is unfalsifiable, and its use distorts a team's ability to settle questions with data. On this reading the persona's fluency is a defect. A summary in the form of a person is harder to check than a table.

The other position holds that the persona is a legitimate instrument with a defensible output. Pruitt and Grudin (2003) and Miaskiewicz and Kozar (2011) locate its value in engagement, communication and the surfacing of assumptions. Salminen, Jung, Chowdhury, Sengun and Jansen (2020) show a measured advantage for a persona-based presentation of data. On this reading the fluency is the point, because it makes a team reason about a user at all.

Settling the question needs a study the literature does not contain: build personas from documented observations, have a design team take specific decisions with them, and compare the resulting designs against a control team given the same data without personas, stating what observation would count as the persona being wrong, which is exactly what Chapman and Milham (2006) find missing. Until then the defensible position is the one taken here: a persona is a device whose value is holding attention, and every detail should trace to something observed.

The evidence is strongest on the side that limits the persona. Chapman and Milham make a logical case rather than an empirical one, fixing a standard rather than measuring a failure, while the defenders' case is consensus and practice, since Miaskiewicz and Kozar aggregate expert opinion and Pruitt and Grudin report three years of use. Only Salminen and colleagues measure an outcome, and their task is identifying which user a behaviour came from, not deciding what to build. The strongest objection is that the unit's own modest defence fails the same test: if every detail must trace to an observation, the persona adds nothing a labelled table would not, and its persuasive form becomes a cost. While the dispute is open, a designer builds a persona only from documented observations.

Where the argument is heading

The first is generation, where Salminen, Guan, Jung, Chowdhury and Jansen (2020) document the move from manual writing to algorithm. The second is synthetic detail, where the generated faces Salminen, Jung, Kamel, Santos and Jansen (2022) tested did not damage credibility. The third is provenance, where Friess (2012) and Matthews, Judge and Whittaker (2012) show what happens when a persona is untraceable and a team uses it for talk rather than decisions. A persona should be as checkable as the study it came from. That question passes to unit eight, where the object of design becomes a service the user never sees.

The sources this unit rests on

• Chapman, C. N. and Milham, R. P. (2006). The personas new clothes. Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 50(5), 634-636. doi:10.1177/154193120605000503 They argue a persona cannot be verified or falsified and that its use undermines settling questions with data. It is an argument, not a study, and sets the falsifiability standard this unit's thesis rests on.
• Cooper, A. (1999). The Inmates Are Running the Asylum. Sams Publishing. archive.org/details/inmatesarerunnin00coop Cooper argues interactive products are designed around the machine rather than the person, and proposes goal-directed design in which an archetypal user anchors decisions. It is the practitioner origin of the persona as the design's point of view.
• Cooper, A., Reimann, R., Cronin, D. and Noessel, C. (2014). About Face: The Essentials of Interaction Design. John Wiley and Sons. archive.org/details/aboutfaceessenti0000coop The book gives the operational account of goal-directed design, including how personas are built from research and used to drive scenarios. It is the version of the method teams inherit.
• Friess, E. (2012). Personas and decision making in the design process. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI). doi:10.1145/2207676.2208572 Friess analysed decision-making sessions at a design firm and found personas made relatively few appearances in designers' language, advocates of the method talking in designer-centric terms. It shows personas failing to enter the reasoning they should guide.
• Gu, H., Chandrasegaran, S. and Lloyd, P. (2025). Synthetic users: insights from designers' interactions with persona-based chatbots. Artificial Intelligence for Engineering Design, Analysis and Manufacturing, 39, e2. doi:10.1017/S0890060424000283 A between-subjects study compared an LLM persona chatbot with a written persona summary and found neither empathy nor the quantity or variety of ideas improved, calling it a complement rather than a replacement.
• Matthews, T., Judge, T. and Whittaker, S. (2012). How do designers and user experience professionals actually perceive and use personas? Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI). doi:10.1145/2207676.2208573 They found experienced practitioners used personas almost exclusively for communication rather than design, judging them abstract, impersonal, misleading and distracting. It places the method's real use in communication, not design.
• Miaskiewicz, T. and Kozar, K. A. (2011). Personas and user-centred design. Design Studies, 32(5), 417-430. doi:10.1016/j.destud.2011.03.003 They used a Delphi process to aggregate expert opinion on the benefits claimed for personas, reporting consensus that the suggested benefits have substance. It is consensus about perceived benefits, not a measured outcome.
• Nielsen, L. (2019). Personas - User Focused Design. Springer. doi:10.1007/978-1-4471-7427-1 Nielsen develops persona theory and method and argues personas work when grounded in data and used as a shared frame for a team. It gives the condition that defines a defensible persona.
• Pruitt, J. and Adlin, T. (2006). The Persona Lifecycle. Morgan Kaufmann. archive.org/details/personalifecycle0000prui They set out a lifecycle of planning, creating, communicating and using personas, positioning them not as a panacea but as a method complementing scenarios and user testing. Their framing supports reading the persona as one instrument among several.
• Pruitt, J. and Grudin, J. (2003). Personas: practice and theory. Proceedings of the 2003 Conference on Designing for User Experiences (DUX). doi:10.1145/997078.997089 They describe three years of persona use, outline the theory behind the method's engagement, and argue personas convey qualitative and quantitative data while focusing attention on what other methods miss. It is the strongest case for personas as a device for attention.
• Ronkko, K., Hellman, M., Kilander, B. and Dittrich, Y. (2004). Personas is not applicable. Proceedings of the 8th Conference on Participatory Design (PDC). doi:10.1145/1011870.1011884 They report a project in which personas failed because of unrecognised patterns of dominance in the telecom branch, commercial obstacles overriding the design work. The case separates a method's weaknesses from the organisation that defeated it.
• Salminen, J., Guan, K., Jung, S. G., Chowdhury, S. A. and Jansen, B. J. (2020). A literature review of quantitative persona creation. Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (CHI). doi:10.1145/3313831.3376502 They reviewed 49 studies of quantitative persona creation from 2005 to 2019, describing three stages of maturation and arguing that sharing datasets, code and algorithms is needed to mature. The review marks the shift to data-driven construction.
• Salminen, J., Jung, S. G., An, J., Kwak, H., Nielsen, L. and Jansen, B. J. (2019). Confusion and information triggered by photos in persona profiles. International Journal of Human-Computer Studies, 129, 1-14. doi:10.1016/j.ijhcs.2019.03.005 They found contextual photos gave users more information about a persona, while photos of multiple people produced confusion and lowered informativeness. A persona's concrete details carry costs as well as benefits.
• Salminen, J., Jung, S. G., Kamel, A. M. S., Santos, J. M. and Jansen, B. J. (2022). Using artificially generated pictures in customer-facing systems. Behaviour and Information Technology, 41(5), 905-921. doi:10.1080/0144929X.2020.1838610 They rated 1,000 generated faces from 6,812 crowd judgements, 90 per cent medium quality or better, then tested them with 496 participants in data-driven personas. The pictures did not lower authenticity, clarity, empathy or willingness to use.
• Salminen, J., Jung, S. G., Chowdhury, S., Sengun, S. and Jansen, B. J. (2020). Personas and analytics. Proceedings of the 2020 CHI Conference on Human Factors in Computing Systems (CHI). doi:10.1145/3313831.3376770 They ran a 34-participant experiment comparing a persona system and an analytics system on identical data, finding personas gave faster completion and higher identification accuracy. It is the strongest measured persona benefit, for identification rather than design quality.
• Turner, P. and Turner, S. (2011). Is stereotyping inevitable when designing with personas? Design Studies, 32(1), 30-44. doi:10.1016/j.destud.2010.06.002 They argue that stereotyping is a tendency of the method.