Building Synthetic Users with "Guts First": Why Logic Isn't Enough
To build synthetic users that truly act like people, we must stop chasing perfect logic and start engineering the "gut" instincts and emotional biases that drive real human decisions.
At Synthetic Users, our job is simple to say and hard to do: replicate human behaviour well enough that teams can run user research faster, cheaper, and earlier.
We do that today with Large Language Models (LLMs), retrieval systems, and careful prompting. This gets you surprisingly far. But there’s a deeper scientific reason why we believe the field can go further, and why the next leap isn’t “more reasoning” but better guts.
This post ties together three threads:
Damásio’s neuroscience: Why emotion is the engine of choice, not a distraction from it.
The "System 1" Trap: Why the AI industry is building "smart" models when researchers need "human" models.
The Solution: How synthetic users can be engineered to have "gut feelings" so they don't just act logical but act real.
1. Damásio: Emotions are programs, feelings are the music
To understand why users behave the way they do, we look to neuroscientist Antonio Damásio (a personal hero of ours and it's not because he is Portuguese). He makes a critical distinction that most people miss:
Emotion is a largely automatic action program. It is the body preparing to act (fight, flee, freeze, approach). These changes happen to your nervous system before you are even aware of them.
Feeling is the conscious experience of those changes. It is the mental "music" playing in your head after the orchestra (your body) has already started.
The "Somatic Marker" (The Gut Check)
Damásio’s Somatic Marker Hypothesis argues that our bodies "tag" situations with value. When you face a choice, these markers fire as fast "gut signals" that bias your decision before you start logically analyzing the pros and cons.
If you damage the part of the brain that handles these markers, you don’t lose IQ. You lose the ability to make decisions. You can list the pros and cons of a washing machine for three hours, but you can't pick one.
Why this matters for UX Research and Synthetic Users:
Most products don't fail because users can't logically explain features. They fail because users:
Say they want X, but impulsively choose Y.
Bounce instantly because the UI "felt off."
Over-index on one bad memory (bias).
Buy a product because it matches their identity, not its utility.
These are Somatic Marker behaviours. They aren't rational calculus; they are emotional biases. If we want Synthetic Users to behave like your real customers, we can't ignore this layer.
2. The Trap: Why "Smarter" AI isn't always "Better" AI
Andrej Karpathy (and originally Daniel Kahneman) popularized a framework that is essential here:
System 1: Fast, automatic, intuitive, emotional.
System 2: Slow, deliberate, calculating, logical.
The Industry Sprint to System 2
Right now, the AI industry is obsessed with making models smarter (System 2). They are building methods like "Chain-of-Thought" (forcing the AI to think step-by-step) to solve math problems and write code.
The Synthetic User Problem
This is great for a coding assistant, but it can be fatal for a synthetic user.
If you simply scale up System 2 reasoning, you get a hyper-rational agent.
Imagine a synthetic user evaluating a sketchy Fintech app:
A "System 2" AI might say:
"The interest rate is 2% higher than the market average, and the onboarding flow is efficient. I will sign up."
A Real Human (System 1) might say:
"The logo looks cheap and they asked for my SSN too fast. I don't trust it. I'm out."
The real human reaction happens in under two seconds. It’s a gut check. If synthetic users are too logical, they will accept products that real humans would reject. We don't just need smart models; we need models that can simulate human irrationality.
3. How to make AI "Human-System-1"
The goal isn't to build a biological brain, but to build a system that replicates human preference patterns. To do that, the AI architecture needs "guts."
Here are the practical levers that could be used to achieve this:
3.1 Fine-tuning for "Snap Judgments"
Humans don't always deliberate. We often choose by instinct. To mimic this, models can be trained on data that reflects rapid categorization and "first answer that comes to mind" scenarios. This nudges the model away from writing a dissertation and toward giving a reaction.
3.2 Bounded Rationality (Simulating Laziness)
The human brain is an energy-miser. We don't solve a logic puzzle every time we buy toothpaste.
This can be mimicked in AI by limiting the "compute" budget for certain decisions. By forcing the model to answer quickly without "thinking/revising" too deeply, you get outputs that rely on heuristics (mental shortcuts)—just like a tired user shopping on their phone at 8 PM.
3.3 Affect and Drives (The "Mood" Layer)
Humans don't just think in words; we think in states (anxious, bored, excited).
A "guts-first" model would ideally possess a dynamic internal state based on established psychological dimensions (like the Circumplex Model of Affect):
Valence: Is this positive or negative?
Arousal: Am I calm or energized?
Example Scenario:
State A (High Trust/Excitement): The user is in a "gain-seeking" mode. They might overlook typos in a UI and click "Buy."
State B (Anxiety/High Arousal): The user is in "risk-avoidance" mode. The exact same UI typos now trigger a "scam" flag, and they leave.
By maintaining this "mood state" across a session, a synthetic user would behave consistently. They wouldn't just analyze a product; they would react to it based on their simulated emotional state.
3.4 Persona + Memory
Gut instincts are shaped by history. To be believable, a synthetic user needs a persistent memory of "past experiences" to form biases. If a synthetic user "had a bad experience" with a subscription model in its backstory, it should instinctively hesitate when it sees a "Subscribe" button today.
4. The "Guts First" Stack
If we take Damásio seriously, we can't just slap a persona on a chatbot. A truly effective synthetic user stack needs:
The Habit Engine (System 1): A fast, biased model that reacts to stimuli based on persona and "mood."
The Rationalizer (System 2): A slower layer that steps in only when the task is novel or difficult, or used to justify the gut decision after the fact.
Here’s a practical end-game:
Use a big model + tools to produce correct answers. Shuffle between models if you can.
Distill that into a smaller fast model (habit engine).
Run the fast model by default with low token budget.
Escalate to slow System-2 scaffolding only when confidence is low.
This is a first pass at mirroring human cognition: Gut first, reasoning second.
A quick GPT Summary for those who skip to the end
Damásio was insightful: Emotion isn't the enemy of decision-making; it's the foundation of it.
The Industry Gap: While the world rushes toward hyper-logical AI, user research requires human-like, biased, emotional AI.
The Path Forward: By engineering "somatic markers," affective states, and bounded rationality, the industry can move toward synthetic users that don't just think like your customers: they choose like them.
References & Further Reading
On Neuroscience & Emotion
Damásio, A. (1994). Descartes' Error: Emotion, Reason, and the Human Brain. (Foundational text on the Somatic Marker Hypothesis).
Damásio, A. (1999). The Feeling of What Happens: Body and Emotion in the Making of Consciousness.
Russell, J. A. (1980). "A Circumplex Model of Affect." Journal of Personality and Social Psychology. (The source for Valence/Arousal modeling).
On AI & System 1/2
Kahneman, D. (2011). Thinking, Fast and Slow. (The origin of the System 1 vs 2 framework).
Karpathy, A. Public talks on LLM psychology and the "OS" analogy. Karpathy.ai
Wei, J., et al. (2022). "Chain-of-Thought Prompting Elicits Reasoning in Large Language Models." NeurIPS. (The academic basis for "forcing" System 2 logic).
Simon, H. A. (1955). "A Behavioral Model of Rational Choice." (The origin of "Bounded Rationality"—essential for understanding user limitations).
Releated Articles
More articles for you
Teaching Synthetic Users What Real People Actually Think
Synthetic Users without calibration are individually believable, but collectively wrong. The missing piece is calibration, not better models.
The Lie We Tell Ourselves About Customer Research
Most research asks what people say. The problem is people don't do what they say. This piece breaks down the gap between stated and revealed preference — and why behavioral modeling, not better interviews, is how you close it.
Two ways to run research with Synthetic Users and why the difference matters
Iris, what is the difference of using agents to accelerate research.
Synthetic Users vs digital twins
You don’t need a twin for “a parent in rural Ohio who shops weekly at Walmart, prefers fragrance-free, and has a toddler with eczema.” You sample a parent profile with relevant traits and constraints, add retail and dermatology context, and generate behaviors consistent with both.
Two major papers. One shared direction.
LLM-powered Synthetic Users have crossed from concept to validated method. This proves they can predict human behavior accurately, letting teams run fast, low-cost behavioral experiments without replacing real participants.
Gartner says we lead. That's kind of them.
Gartner’s latest report on AI-powered synthetic user research cites Synthetic Users as a leader.
Introducing Shuffle v2
Shuffle v2 is a feature that intelligently shuffles between multiple large language models via a routing agent to produce more realistic, diverse Synthetic Users with better organic parity.
Chain-of-feeling
Synthetic Users use a “chain-of-feeling” approach—combining emotional states with OCEAN personality traits—to produce more human-like, realistic user responses and yield richer UX insights.
Generative Agent Simulations of 1,000 People
A paper that thoroughly executes a parity study between Synthetic and Organic users.
21 Peer reviewed papers that support the Synthetic Users thesis
Here is a compilation of all the papers that help make a case for Synthetic Users.
Why we shuffle between models — to ensure both parity and diversity!
Synthetic Users balances aligned and unaligned models to maintain diversity and authenticity in simulated users while ensuring ethical standards and user expectations are met.
Latest press articles for Synthetic Users
Synthetic Users and AI are transforming research methodologies, offering innovative, cost-effective alternatives to traditional human subject studies.
Comparison studies. The opportunity lies in the deviation.
When we compare different studies, especially looking at what synthetic (artificial interviews) and organic (real-world interviews) data tell us, we often find they mostly talk about the same things but there's also a bit where they don't match up. This gap is super interesting because it's like finding hidden treasure in what we thought we knew versus what we might have missed.
How we deal with bias
Harnessing the power of AI in our Synthetic Users, we strive for a balance between reflecting reality and ethical responsibility, ensuring diversity and fairness while maintaining realism.
The transition to Continuous Insight
The transition towards Continuous Insight™ aligns research activities more closely with the dynamic needs of the business and ensures that product development is continuously informed by up-to-date user insights.
The Art of the Vibes Engine
Large language models (LLMs) like GPT-4 serve as powerful "vibes engines," empathizing with diverse groups and generating contextually relevant content. Their applications span market research, customer support, user experience design, and mental health support, offering invaluable insights and personalized experiences. While not infallible sources of truth, LLMs enable creativity, personalization, and connection within the realm of human language.
There is a faster and more accurate way to do research. Use Synthetic Users.
How Synthetic Users is changing the research process.
The wisdom of the silicon crowd
In the light of an ancient parable, we explore a new paper that dives into how ensembles of large language models match the prediction accuracy of human crowds. It reveals that combining machine predictions with human insights leads to the most robust forecasting results.
Three research papers that helped us build ❤️ Synthetic Users
For the sceptics amongst us who need more tangible research in order to engage with this brave new world. Full disclosure: we are part of the sceptics.
What is RAG and why it’s important for Synthetic Research
Ahead of our RAG launch we explain Retrieval-Augmented Generation (RAG) and how it enhances Synthetic Users by providing increased realism, contextual depth, and adaptive learning, with profound implications for market research, user experience testing, training, education, and innovative product development.
Synthetic Users system architecture (the simplified version).
Foundation models underpin Synthetic Users with advanced capabilities, enhanced by synthetic data and RAG layers for realism and business alignment, all within a collaborative multi-agent framework for richer interactions.
Saturation score. How do we know how many interviews to run?
Determine your interview target for achieving topic saturation using our efficient approach, leveraging the historical wisdom of research pioneers. This method ensures deep insights with theoretical sampling at its core.
How Synthetic Users are gaining depth
Synthetic Users are evolving to address criticism about their generalist nature by incorporating representative data sets and personal narratives.
How we compare interviews to ensure we improve our Synthetic Organic Parity — 85 to 92%
How do we know we are right? How do we know our Synthetic Users are as real as organic users? We compare.
Synthetic Users: Merging Qualitative and Quantitative Research, in seconds.
At Synthetic users we are blurring the lines between qualitative and quantitative research. Here's how we are going about this transformative approach.