Persona Problems

I first heard the word 'persona' back in 2016 on an automotive pitch, where we were building customer journeys for different representative audiences. Personas then were born from media targeting: stock photos layered over demographic and behavioural data to represent the audiences we'd try to find and influence.

Personas have evolved, but they're still born from the same place: downstream execution data. They're no longer stock images but LLM-powered interfaces that synthesise existing knowledge and create a compelling illusion of human characteristics. But they don't have human characteristics, so don't fall for the persona trap.

According to McKinsey, 71% of organisations now use AI in marketing, with synthetic customer research offering the lure of the ultimate efficiency gain: instant insights without the messiness of actual humans. Feed an AI the right prompts, and you'll get articulate, detailed responses about customer motivations, pain points, and preferences. It's fast, scalable, and available 24/7.

But let's be brutally honest: these are not insights. These outputs are probabilistic responses based on existing knowledge—summarised well and wrapped in language that appears human. They're sophisticated data synthesis masquerading as customer understanding.

The seduction is understandable. Teams under pressure to deliver insights quickly find AI personas irresistible. Why spend weeks scheduling interviews, transcribing conversations, and analysing contradictory responses when you can get clean, consistent answers in minutes?

Because clean and consistent isn't how humans work.

Let's be clear about what synthetic personas excel at, because they're genuinely powerful tools when deployed appropriately.

For downstream execution, AI personas shine. They excel at refining messaging once you understand your audience, testing variations of creative concepts, and scaling personalisation across channels. They're exceptional at pattern recognition and data synthesis, identifying themes across thousands of customer interactions and flagging inconsistencies in messaging.

The sweet spot is using AI personas as thinking partners, not decision makers. They can help frame research questions, structure approaches, and identify blind spots in assumptions. They're particularly valuable for teams needing to align quickly around customer understanding before investing in human research.

But here's where they become dangerous: AI personas represent the mean of a consumer segment with more aggregated knowledge than any individual in that segment. They are not unique individuals with pain, emotional drives, and motivational contradictions. They don't express authentic emotion or rationalise feelings. They don't feel shame, awkwardness, or social pressure.

Above all, they are alone. They 'talk' to users and check responses from their language models, but they do not interact with others. This matters enormously because the focus group has persisted as a core research method precisely because insights often emerge from intersections between people: the serendipity of moderated conversation, social dynamics, and group influence.

The real risk lies in using synthetic personas for strategic, upstream decisions: the foundational choices that shape product development, brand positioning, and innovation priorities.

Dr. Thomas Walter at Dentsu explains the core problem: "consumers are contradictory, sometimes irrational, often surprising. They are not the platonic ideal of a demographic segment, but real humans with tensions, habits, inconsistencies."

The Nielsen Norman Group's testing found that AI chatbots have "a tendency to want to please (sycophancy)" and "do not always model human behaviour well," providing "an unrealistic view of human behaviour" with "values, desires, and needs that are too shallow."

B2B persona expert Ardath Albee warns marketers directly: "Don't trust the AI. Assume everything you get from AI is incorrect until you've reviewed it carefully." Her research shows AI excels at "making lists of stuff, but the problem is specificity" and lacks differentiation.

The stakes are higher in upstream decisions because errors compound. A synthetic persona that misses an emotional trigger leads to positioning strategies that fundamentally misunderstand customer motivations, resulting in campaigns that feel tone-deaf or products that solve the wrong problems.

Don't believe me? Try these simple tests with whatever persona tool you're using today, tests that any competent human researcher would pass easily:

Consistency Test: Ask the same persona "When did you start using [brand]?" across different sessions. I tested five leading AI persona tools with this basic question about coffee consumption. The responses were so inconsistent—ranging from "last month" to "five years ago" for the same supposed individual—that I wondered if they were describing the same person or five different customers.

Memory Test: In a longer conversation, ask "Why did you start using [brand]?" early in the session, then ask the same question again later. Watch how the rationale shifts, contradicts, or completely changes. Real humans have consistent personal histories. AI personas have probabilistic response patterns.

Emotional Rationalisation Test: Ask "How did you feel when you first used [brand]?" and then follow up with "Why did you feel that way?" The responses reveal the shallow emotional modelling at AI's core—feelings without authentic psychological foundations, emotions without the messy irrationality that drives real human behaviour.

Social Context Test: Ask about influence from friends, family, or colleagues on their decision-making. AI personas struggle with social dynamics because they exist in isolation, unable to model the complex web of relationships that shape real purchasing decisions.

These aren't edge cases, they're the foundational basics of any qualitative exploration, just introductions to understanding someone. If AI personas struggle with these fundamentals, how can we trust them with strategic decisions?

In a world where everyone has access to the same AI tools, competitive advantage comes from maintaining authentic human connection. In the same McKinsey study, companies achieving AI excellence show 60% higher revenue growth, but they distinguish themselves through strategic balance rather than AI replacement of human functions.

The brands that thrive won't be those with the most sophisticated AI personas, they'll be those that use AI to enhance human capabilities while preserving the empathy, creativity, and contextual understanding that builds lasting relationships.

Here's the uncomfortable truth: AI personas aren't just ineffective for strategic insights, they're actively dangerous. They provide the illusion of customer understanding while systematically removing the contradictions, emotions, and social dynamics that drive real human behaviour.

The future belongs to marketing teams that use AI to enhance human capabilities rather than replace them. Use AI personas for what they excel at: pattern recognition, concept refinement, and execution scaling. But always validate synthetic insights against real customer interviews before making strategic decisions.

Maintain budget allocation for authentic human research. The most successful marketing teams invest in both AI tools and human connection, recognising that each serves different purposes in the insight generation process.

Remember that the goal is collaborative intelligence: AI handles data processing while humans provide the empathy, creativity, and strategic thinking that build lasting brand relationships.

The question isn't whether you can afford to invest in real human research. The question is whether you can afford the cost of building strategies on synthetic foundations. Every dollar spent on AI personas for strategic decisions is a dollar not spent understanding the contradictory, irrational, surprising humans who actually buy your products.

How are you balancing AI efficiency with authentic human connection in your research? Where have you seen synthetic personas excel and where have they led you astray? More importantly: are you brave enough to test your current AI personas with the simple questions above?

The answer might surprise you—or save you from a very expensive mistake.

Brandflow is written by Justin Billingsley, who has spent his career on all three sides of the industry's table: senior client, global agency leader, technology founder. First published 17 July 2025 in the Brandflow newsletter on LinkedIn.