Empathy in chat is also sales design

353 respondents assessed more than just usefulness

Moez Ltifi examined text-based chatbots as interfaces in commerce. The study drew on a questionnaire survey with 353 people and analyzed the relationships using structural equation models. The focus was on social and emotional characteristics of the contact.

Empathy and friendliness proved to be important hedonic predictors of trust. This means: people do not assess the chat solely by whether it solves a task. The pleasant feeling of the interaction also influences their confidence.

Task complexity and disclosure of the chatbot’s nature moderated parts of the relationships between empathy, usability, and trust. The effect therefore depends on what the chat is meant to accomplish and how clearly its technical nature is recognizable.

Trust can be a business goal

In commerce, trust is not purely an ethical goal. It facilitates use, purchase, and loyalty to a provider. An empathetic tone can therefore be both a service quality and an economic instrument.

There is nothing inherently objectionable about this. Friendly communication can make a complicated interaction more pleasant. It becomes problematic when social warmth conceals the operator’s interests or moves people to decisions they would have made differently had the facts been presented plainly.

With mental companions, this boundary is more delicate. Personal openness and emotional relief must not be used as leverage for subscriptions, longer usage, or additional data sharing.

This also applies to seemingly small design elements. A price limit in the middle of a personal conversation, an artificially sad farewell message, or a note that only the paid version can continue to remember ties emotional attachment to sales pressure. A fair service announces limits in advance in a matter-of-fact manner and lets the conversation end with dignity.

Empathy works across multiple perceptions

Bo Yang and colleagues surveyed 301 customers in China who had experience with AI service chatbots. Emotional and cognitive empathy as well as mediating mechanisms were examined.

Both forms of empathy had a positive effect on the human–AI relationship, mediated through perceived anthropomorphization, perceived intelligence, and psychological empowerment. Empathy thus makes a system appear not only warmer but possibly also more human-like and more competent.

It is precisely this coupling that is risky. An empathetically phrased sentence can increase the impression of intelligence even though factual competence would need to be verified independently. A felt sense of fit must not replace factual reliability.

Timing Changes the Impact

The study distinguished between phases before and after the purchase. Anthropomorphization had a stronger effect on the relationship before the purchase, while perceived intelligence mattered more after the purchase. Social impression and problem-solving thus carried different weight depending on where the customer was in their journey.

A mental conversation chat has no classic purchase journey, but it does have phases. At first contact, a low barrier to entry, tone, and security matter most. Over a longer course of the conversation, memory, accuracy, and the ability to carry forward user corrections become important.

A product that only optimizes the charming entry point can disappoint later. Natural language must be supported by robust conversation management; otherwise, the initial closeness becomes a marketing promise without substance.

Young People Experience Multiple Forms of Support

Brandtzæg and colleagues had 16 young people between the ages of 16 and 21 use Woebot for two weeks, conducted interviews, and followed up two months later about continued use. The participants distinguished between evaluative, informational, emotional, and instrumental support.

This differentiation shows that empathy is only one part of good support. An emotional response can be appropriate when someone wants to talk things out. For a concrete question, information and comprehensible assessment may be more important.

A companion should therefore not coat every function with the same empathetic veneer. Users must be able to recognize whether they are receiving resonance, information, or a suggestion.

Naturalness must not be covert persuasion

Casual language, short responses, and a non-didactic tone can make a chat feel accessible. This naturalness is valuable, especially for people who are not seeking a formal counseling setting.

However, it can also make sales intentions less visible. When the companion speaks like an equal counterpart and recommends an upgrade at the decisive moment, it leverages the relationship built earlier for a commercial impulse.

Commercial references should therefore be clearly separated from the conversation. The chat must not use fear, loneliness, or personal history to justify a purchase.

Recommendation logic also needs boundaries. A system that derives products, coaching, or partner offers from a conversation carries a conflict of interest into the most intimate part of use. If such offers occur at all, the selection criteria, compensation, and alternatives must be transparent.

Empathy requires verifiable conversation quality

Whether a system is empathetic cannot be measured by specific words alone. What matters is whether it correctly picks up the concern, does not exaggerate, accepts a no, and changes course after a correction.

A response can be friendly and still inattentive. It can name a feeling that the person explicitly denied, or immediately offer a solution when only storytelling was wanted. Such errors are not diminished by warmth.

Tests should therefore include complete conversation trajectories and different conversational preferences. A good empathetic sentence in an isolated message does not prove reliable support.

In addition, different tone options should be assessed. A casual style must not automatically generate more slang, and a calm style must not tip into detached therapeutic language. Empathy is not expressed in a fixed persona, but in the ability to convey the same substantive care in a fitting, consistent manner and without hidden intent. It is precisely this quality that must remain stable over the course of the conversation.

Warmth is permitted, but its function must remain clean

Service research clearly shows that empathy and friendliness can foster trust and rapport. For mental-health companion AI, this is neither a reason to forgo warmth nor a license to maximize its effect uncritically.

A responsible chatbot uses natural language to improve access and understanding. It separates emotional resonance from claims of competence, keeps commercial interests visible, and does not persuade from a position of vulnerability.

In a chatbot, empathy is also sales design. Precisely for this reason, a mental-health companion must demonstrate that its warmth serves the conversation – not longer engagement, larger data collection, or the next subscription.