Are you AI-enhancing your personal messages?
Magnus Söderlund, June 2026
The rapid adoption of generative AI has the potential to change one of the most intimate dimensions of human life, personal communication, because we humans have started to use AI to enhance our text-based messages. And so far, we rarely disclose what we have done. Yet, receivers are becoming increasingly capable of detecting signs of AI involvement through stylistic markers such as bullet points, text in bold here and there, and unusual words such as “delve”. So, when a sender has AI-enhanced his or her messages before sending them, without disclosure of how the messages have been produced, and when the receiver suspects that messages have been enhanced, what will happen? I have had a chance to conduct some experiments regarding this (no AI-enhancement vs. AI-enhancement of messages). The results are pretty clear: when the sender has used AI for language enhancement, the participants react with a more negative attitude towards the sender.
Why is this so? The answer is that personal communication is typically interpreted not merely in terms of informational content. Language provides evidence of the sender’s intentions, emotions, attention, and other mental states, which means that the suspicion that a message has been substantially mediated by AI can influence the receiver’s view of what is in the
sender’s mind. Quick background: we humans are more or less programmed to make mindreading attempts when we interact with other persons; information about their mental states is needed to explain, predict, and influence them. We never have full access to others’ minds (this is probably good), but what they say, do, and write provide clues about what is in there. When we realize that the sender has used AI, however, the inferential process can become disrupted and result in perceptions that the access to the sender’s inner world is restricted.
One reason is that the use of AI makes it harder to establish the intended meaning of message content – does it have to do with what is actually in the sender’s mind or with an algorithm? This uncertainty weakens evidence of the message as something that reflects the sender’s mind. And if the message no longer provides transparent access to the sender’s thoughts and emotions, but instead presents a curated version of them, how can one know who the sender actually is or what he or she wants? Rather than experiencing direct interpersonal contact, then, the receiver experiences contact filtered through an intermediary entity.
Another reason why AI-enhancement can produce a sense that the content of the sender’s mind is hidden is based on how LLMs work. They are trained on massive datasets scraped from books, newspapers, and the larger internet. Much of this material has been written by distinct authors, which opens the possibility that one particular AI-enhanced text is in fact somehow multi-authored and thereby it is less indicative of what is in the mind of a person who is the alleged sender. And if the suspected use of AI-enhancement comes with no disclosure, it can be seen as a type of deception signaling that even more is hidden.
Suspected AI use can also question the sender’s authenticity. We humans tend to associate authenticity with spontaneity and imperfection, but AI frequently suppresses these aspects. That is to say, in the typical interpersonal text-based communication, awkward wording, spelling mistakes and inconsistency often function as cues that a message is aligned with what is truly in the sender’s mind. And authenticity judgments concern not only whether the sender truly feels or thinks what he or she writes. Authenticity is also about who has written a text when it is signed by a specific individual. An AI-enhanced message may therefore be perceived as less authentic even if its content is sincere. The receiver may conclude that the sender outsourced not only the words in a message, but also the very content in terms of thoughts and emotions.
Then, in the next step in the receiver’s sense-making efforts, perceptions that some of the sender’s mind content is obscured can reduce trust in the sender, because interpersonal trust depends heavily upon transparency – particularly when it comes to other persons’ intentions. So, when undisclosed AI-enhancement of a message is detected, the receiver may feel that an implicit communicative norm has been violated. Even if the sender never explicitly claimed sole authorship, receivers have to date assumed that personal messages from a human are primarily human-created. And trust in a person is typically a strong determinant of the liking of a person, so it is not surprising that reduced trust reduces liking.
My recommendation to senders, then, is this: avoid using AI enhancement of messages (unless you are good at making AI markers disappear). However, this recommendation is unlikely to be followed. The allure of the enhanced text is very hard to resist. Why would I want to reveal my writing imperfections if there is easy help? This means that an ever increasing number of enhanced texts can be expected. Most likely, it will produce several unwelcome consequences in the aggregate. One of them is that in comparison with AI text, many human-written texts will be seen as less eloquent, so that we humans feel pressured to enhance our communication in order to remain socially effective. Another consequence is that we will have to devote more mental efforts to distinguishing genuine communication from technologically optimized expression. If this questions the baseline assumption that personal messages transparently reflect the sender’s mind, it could intensify social skepticism.
It is also possible that communication – particularly communication with emotional content – can become progressively aestheticized. After all, AI systems are effective at producing convincing and elegant (well, at less error-free) language. As its stylistic norms spread in a population, people may begin evaluating personal communication less according to sincerity
and more according to expressive sophistication. Relationships may thereby become increasingly mediated by aesthetic standards for language use.
To this comes that the meaning of effort may change. Traditionally, writing a personal message is an investment of time and cognitive energy. And we humans are typically sensitive to the effort others expend in relation to us – it is an indicator of how valuable we are to others. AI, however, dramatically lowers the cost of producing compelling communication. As a result, receivers may increasingly devalue those with whom they communicate. There is a cynicism aspect, too. If we increasingly assume that friendliness, empathy, enthusiasm and care, as expressed in personal messages, are merely computationally generated, emotional language itself may lose credibility because we no longer perceive it as sincere.
At the end of the day, we may have to accept that society in the age of AI will comprise a combination of increased communication quality and decreased perceived intimacy. Messages may become more articulate, empathetic, and emotionally intelligent on the surface, while simultaneously producing uncertainty about the extent to which another human mind has produced them. In this sense, the widespread use of AI-tools for message enhancement may alter fundamental assumptions about others’ mind content, authenticity and trust.