Young people do not need a chatbot for everything

Two weeks of use instead of a brief demonstration

Brandtzæg and colleagues invited 16 young people aged 16 to 21 to use Woebot for two weeks. They subsequently conducted in-depth individual interviews. During the interviews, participants also viewed a prototype designed to provide information to young people.

Two months later, participants reported on their continued use of Woebot. This design combines concrete product experience, reflection, and a later retrospective. It goes deeper than a one-time first impression, but with 16 individuals it remains a small qualitative study.

The results therefore do not provide percentage figures for all young people. They show how young people differentiate support from chatbots and which questions become relevant for subsequent experiments and design.

Emotional support is not merely friendliness

Emotional support can mean that a system picks up on stress, formulates comfort, or leaves a thought without an immediate solution. The value lies not solely in positive words. What matters is whether the response fits the concern.

A reflexive “That sounds hard” can seem attentive the first time but become formulaic over the course of the conversation. Even more problematic is when every emotional expression is immediately interpreted as a problem. Young people do not need a chat that dramatizes normal fluctuations.

A system is therefore emotionally supportive even when it accepts corrections, does not exaggerate, and picks up on the person’s tone. Warmth without precision is not enough.

This also includes tolerating silence and brevity. Not every brief message is a mandate for a long intervention. Sometimes support consists of not immediately dissecting a thought but leaving room for the next sentence with a small, open response.

Information requires different quality rules

Informational support follows a different logic. Here, the chat should explain something, organize options, or make a reliable source accessible. The response does not need to be emotionally deep but rather understandable, current, and verifiable.

A companion that phrases facts and personal assessments in the same tone blurs this distinction. Especially young users should be able to recognize when the system knows something, when it suspects, and when it is merely asking a conversational question.

Sources, cautious phrasing, and the possibility to verify details are therefore not foreign elements in a natural chat. They protect the relationship from false authority.

Evaluative support must not become a verdict

Evaluative or interpretive support helps people assess their situation. A chatbot can reflect that a reaction seems understandable or offer an alternative perspective. This can be relieving when someone is stuck in their own thoughts.

The boundary lies between offering and determining. A sentence like “Perhaps what bothers you most about this is…” leaves room. The assertion “You react this way because you have fear of commitment” turns little context into an alleged truth.

This restraint is especially important with young people. A system should not narrow development through psychological labels. Good assessment opens up ways of thinking and remains correctable.

Instrumental help ends at the world

Instrumental support means practical help with a task. A chatbot can formulate a plan, draft a message, or sort steps. However, it cannot itself accompany someone to an appointment, resolve a conflict, or bear the consequences of a decision.

Ta and colleagues found in their analysis of 1,854 Replika reviews and 66 open-ended responses companionship as well as emotional, informational, and evaluative support, but no tangible support. The technical limitation becomes evident precisely where language would need to transition into action.

The product must not obscure this transition. A suggestion is not a completed task, and a text window is not a social network. Instrumental help from AI remains preparation and structuring.

Friendliness builds trust – also in sales

A study by Moez Ltifi with 353 respondents examined text-based chatbots in retail. Empathy and friendliness proved to be important hedonic predictors of trust. Task complexity and the disclosure of the chatbot's nature influenced parts of the relationships examined.

The context is not mental support, but the finding is relevant: social and emotional design influences trust even when the actual purpose is commercial. A chat phrased in a friendly manner is perceived not only as more pleasant but possibly as more credible.

With young target groups, this effect must not be used uncritically. Warmth should facilitate access, not create uncertainty about the provider's competence and interests. A sympathetic tone does not make false information more correct.

Choices must be reflected in behavior

A product can offer different conversation directions: talking, understanding, seeking a new perspective, or finding a next step. Such options are only useful if they actually change the system's behavior.

Those who have chosen to talk should not receive an exercise after every message. Those seeking information do not need several emotional follow-up questions first. And those wanting a different perspective should receive a perspective as an offer, not as a diagnosis.

Choices make the forms of support visible and give users control. At the same time, they prevent the chat from imposing its preferred helper role on every conversation.

The selection must not become a rigid box. A conversation can change, and young users must be able to say, without restarting, that they now only want to share something, hear an opinion, or are actually looking for something specific. Control does not come from four nice buttons, but from the system’s willingness to actually respond to such changes of course.

Low-threshold does not mean boundless responsibility

The study with young people shows potential and concerns not as opposites. Chatbots can offer access to various forms of social support. Their easy accessibility in particular makes them interesting for people who would otherwise hesitate.

However, this strength does not imply responsibility for everything. Emotion, information, assessment, and practical help require different skills and boundaries. A system should recognize or ask about the desired function rather than automatically running through its entire helping program.

Young people do not need an all-knowing digital friend. They need a clear, respectful conversational space whose tone is natural, whose functions are selectable, and whose boundaries do not only become visible when something goes wrong.