Synthetic Empathy – How AI Is Scaling UX’s Errors

It is tempting to believe that the problems of customer experience can be solved with artificial intelligence. When language models are able to analyze vast volumes of feedback, produce fluent UX discourse, and simulate empathy, it creates the impression that understanding has finally been achieved. This impression, however, is misleading. The crisis of customer experience is not computational, but epistemic. The problem is not a lack of analytical capacity or technological sophistication, but a growing uncertainty about what we are actually measuring and under what conditions understanding can be formed at all.

Language models speak the language of customer experience fluently. They use the correct terminology, recognize emotional states, and construct plausible interpretations of user motivations. It is precisely this fluency that creates the illusion of understanding. When language flows effortlessly, organizations begin to believe that the phenomenon itself is under control. This continues a trajectory already visible in earlier stages of customer experience practice: first the model drifted away from reality, then measurement stopped seeing the human, and now language itself offers a false sense of certainty. The more convincingly AI speaks about experience, the easier it becomes to assume that experience has been understood.

Synthetic Empathy and the Closed Loop

In the age of AI, customer experience has entered a new phase that can be described as synthetic empathy. Language models are capable of generating user personas, simulating feedback, and producing analyses that mimic empathetic understanding. They write reports on user frustration, identify friction points, and propose improvements in ways that appear human and persuasive.

The problem is not that these tools are inaccurate. The problem is that they enclose customer experience entirely within a linguistic loop. When UX practice begins to rely on AI-generated synthetic feedback, the last remaining need to encounter a real human disappears. Experience no longer emerges from interaction with reality, but from a system in which AI analyzes fiction produced by AI and calls it understanding.

This is not a misuse of artificial intelligence. It is the logical outcome of a customer experience paradigm in which experience is primarily defined through language. If experience is understood as a narrative that can be analyzed, summarized, and optimized, then a language model is the perfect instrument. At the same time, UX ceases to be an observation of real human behavior and becomes a self-reinforcing structure.

Language Is a Slow Signal, State Is Fast

At this point, a fundamental theoretical problem in the relationship between AI and customer experience becomes visible. Language is a slow signal. It is retrospective, symbolic, and already interpreted. Experience, by contrast, unfolds as a simultaneous, embodied, and intensive state. When experience is translated into text, a substantial portion of it is lost.

Intensity, rhythm, physiological tension, and contextual pressure do not travel easily through words. They are not readily explicit and resist smooth articulation. Yet these elements are precisely what shape how people act within systems: how they hesitate, adapt, comply, or withdraw. Experience is not a story; it is a state in which biological and cognitive realities are tightly coupled.

Language models operate at the symbolic level. They process words, meanings, and probabilities. Human experience, however, takes place at the biological and cognitive level, where meaning precedes language. This creates a fundamental mismatch: the human is not a prompt. When we attempt to understand experience through language, we are already too late.

When UX Believes Experience Is What People Say

The history of customer experience measurement is largely the history of surveys. Surveys are assumed to provide a direct channel to human experience. In reality, they measure a retrospective narrative about experience. This limitation is not new, but AI amplifies it.

When UX practice relies on AI-analyzed feedback, we learn to believe that words are equivalent to experience. Linguistic expression is granted the status of reality. In doing so, we lose the ability to detect what remains unsaid. Language models do not read between the lines, nor can they perceive the cognitive tension that arises from the mismatch between a user’s goal and the system’s affordances. This is not a deficiency of AI; it is intrinsic to its nature.

The factors that matter most for experience are often precisely those that are not articulated. A frustrated user may lack the capacity or motivation to describe what went wrong. Adaptation is interpreted as satisfaction, and silence as approval. When UX equates experience with what people say, it stops seeing the human as a whole.

Dynamic Baselines and Statistical Blindness

At this point, the distinction must be stated as clearly as possible. Language models predict the average human. They are built on massive datasets in which individual differences are flattened into probabilities. Customer experience, however, does not occur at the average. It always emerges as a deviation from an individual’s momentary baseline. Experience is defined by the moment in which a person’s state does not match what is normal – not by moments in which everything proceeds as expected.

This is precisely why language models appear to perform exceptionally well when nothing is wrong. They describe typical behavior accurately. They fail at the point where customer experience becomes genuinely meaningful: situations in which an individual’s cognitive, physiological, or contextual state deviates from the norm. This is not a technical flaw, but a structural consequence of attempting to infer deviation from averages.

Language models operate on statistical probability. They predict the most likely next word. Human behavior, by contrast, unfolds relative to a situational baseline. Each person is their own reference point, and that reference point is constantly shifting.

Cognitive load, stress, time pressure, and environment reshape behavior from moment to moment. This dynamic baseline cannot be inferred from language or statistical aggregates. When UX integrates AI into the interpretation of customer experience, it creates a structural blindness. The model may be precise within its own framework, but it fails to capture the individual’s lived state.

This problem cannot be solved by adding more data or refining algorithms. It arises from attempting to understand experience at the wrong level. Statistical probability reveals nothing about the state a person is in when they encounter a system.

Scaled Control Illusions

Previously, metrics gave organizations a sense of control. AI now multiplies that illusion. AI dashboards offer elegant summaries, condensed insights, and seemingly deep understanding of the customer. Organizations feel that they understand more than ever before.

At the same time, contact with real human experience weakens further. The more AI tells us about experience, the less we encounter it directly. The uncertainty that originally defined customer experience as a research endeavor disappears. In its place emerges a confidence rooted in linguistic fluency rather than observation.

Toward Signal-Based Understanding

This does not mean that AI is useless, nor that customer experience thinking has reached a dead end. It means that the starting point is wrong. Perhaps experience should not be asked about, but observed. Perhaps it should not be modeled as journeys, but as states. Perhaps instead of averages, we should attend to intensity, rhythm, and accumulated tension.

This is not a solution, but a change in direction. The question is not how to design a better survey or a more sophisticated language model, but whether we are willing to abandon the language-based illusion of control and learn to see experience as it unfolds.

And it is precisely at this point that the rethinking of customer experience now arrives.