Generative artificial intelligence is increasingly able to compose and edit personal messages; many teenagers now turn to language models for help crafting replies to conflicts, apologies or emotionally charged conversations. The technology shortens the learning process by which people used to acquire skills for expressing themselves and interpreting others’ words.
AI‑mediated communication and early framing
Jeffrey Hancock, Mor Naaman and Karen Levy coined the term "AI‑mediated communication" in 2020 to describe situations where an intelligent system modifies, supplements or creates messages on behalf of a person for communication purposes. In 2020 this mainly meant smart reply suggestions and automatic sentence completion; as language models advanced, they are now often capable of producing full emotional exchanges.
Benefits and research findings
The advantages are clear: those who struggle to find words can get help organizing their thoughts, heated messages can be softened to remove unnecessarily hurtful phrases, and AI can suggest angles the sender had not considered. A 2025 study in the Journal of Computer‑Mediated Communication found that AI‑produced or AI‑modified supportive messages contained more emotional and informational support. Notably, participants who used AI as an adviser but wrote the final message themselves perceived those messages as more authentic than entirely machine‑generated texts.
When assistance becomes replacement
Other studies warn that gradual replacement of practice by AI may have downsides. When recipients learned that a message from a close contact had been produced with AI help, they perceived less effort invested by the sender; this correlated with lower relationship satisfaction and greater uncertainty. A similar reaction occurred when a message had been drafted with human help, suggesting that part of a personal message's value derives from the time and effort spent.
A carefully worded apology conveys more than the correct sequence of words: it signals thinking, uncertainty, and the effort to repair a relationship. A language model can perform the same task in seconds, potentially making this social signal less reliable.
New uncertainties in written communication
Historically, readers inferred a sender’s education, humor, empathy or attentiveness from written text. If algorithms regularly assist with phrasing, polished messages reveal less about the person who sent them. This can complicate interpretation of tone and intent and change expectations about social adaptability.
The developing generation and socialization risks
A crucial question is what happens to people who grow up in this environment. There are no long‑term, ten‑ to fifteen‑year datasets yet: the broad adoption of generative AI is too recent to draw definitive conclusions. Nevertheless, developmental psychology is increasingly studying the role AI systems may play in adolescents’ relationships, self‑expression and emotional life.
Usage data points to rapid change. The Common Sense Media 2025 US survey found that a large majority of 13–17‑year‑olds had tried some kind of AI companion; many use these systems for social interaction, emotional support or casual conversation, and some share important personal matters with them.
Developmental researchers warn that part of emotional support and social connection may migrate to technological systems that are always available, patient and accommodating, and that show far less resistance than human relationships. That resistance, however, is important: learning to communicate involves friction. Children and adolescents misinterpret one another, search for words, overreact, apologise awkwardly and sometimes say the wrong thing at the wrong time. Feedback from others after these moments helps build empathy, emotion regulation and awareness of how the same words can affect different listeners.
Generative AI is particularly effective in these uncomfortable moments: a teenager no longer needs to struggle to say they are disappointed— they can describe the situation and ask for a tactful, firm or empathetic formulation. Occasional use can be helpful, but regular reliance can hand over part of the practice of communication to the machine.
Two people, two AIs — conversational automation
A further development is when both parties route messages through their own AI assistants. One person receives a message and asks their assistant, "What do you think they mean?" The model analyzes tone and possible intent and drafts a reply; the recipient then sends it. The other side may do the same. Visually, two human names continue to appear in the conversation, but in interpretation and formulation two algorithms are participating. Hancock and colleagues have warned that such processes could, over time, change linguistic norms, patterns of social adaptation and even self‑concepts: if a model consistently makes someone appear more positive, funny, assertive or empathetic, it may shape that person’s self‑image.
Promising modes of use
Existing research suggests that outcomes depend heavily on how AI is used. There is a substantial difference between taking over a final text and offering thinking aids. AI‑driven suggestions that the user then rewrites tend to feel more authentic than purely machine‑written messages, pointing to a possible best practice. In that role, AI can become a "communication coach": it can surface alternative perspectives, flag potentially hurtful phrasing and propose multiple formulations, while leaving thought, choice and final wording to the human.
Why this matters for the future
For the next generation AI usage may be as natural as smartphones or search engines are to us. Therefore, what communication skills we transmit becomes especially important: if we substitute practice too early or too often, it may affect the long‑term development of empathy, self‑knowledge and social skills. At the same time, properly applied AI can be a valuable tool for learning better communication, provided its suggestions remain anchored in human reflection and decision.
Conclusion
Generative AI carries a dual potential: it can improve communication quality and support learning, yet regular substitution of communicative practice risks eroding the experiential processes that build empathy and social competence. Absent long‑term evidence, the way and extent to which AI is integrated into everyday exchanges will determine whether it serves primarily as an aid for growth or as a partial replacement of the practice that shapes how we relate to one another.
