AI-Assisted Messages: Warm Words and Human Intentions

A colleague sends a polished note of thanks. It names the project, offers praise and closes warmly. One reader may appreciate it; another may wonder whether AI helped write it. Alongside the words sits a second question: what attention and commitment stand behind them?

Research offers no single rule that AI assistance always helps or harms communication. The setting, the assistance and the outcome being measured matter. For creators and business owners, the useful aim is to express what you mean and remain answerable for the message another person receives.

A friendly reply is one part of an interaction

In a Scientific Reports paper published on 4 April 2023, Jess Hohenstein and colleagues investigated suggested replies in two randomised experiments using a messaging application. The research found that these suggestions could increase positive emotional language and affect how conversation partners evaluated one another.

There was also an association between suspected use of suggested replies and less favourable interpersonal ratings. Crucially, the authors described this suspicion finding as correlational: it did not establish that actual AI use caused those negative judgements.

That distinction prevents a misleading reading. The study is evidence about particular text interactions, language and perceptions. It does not establish that every assisted message will damage a relationship, or that someone’s suspicion proves how a message was composed.

Professionalism and care can receive different ratings

Anthony Coman and Peter Cardon’s study of AI-assisted writing, first published online on 11 March 2024, asked 887 working adults to evaluate workplace messages. Participants generally rated the supplied AI-generated messages favourably for qualities including professionalism, effectiveness and directness.

Sincerity and caring received slightly lower ratings in some conditions where AI assistance was disclosed. These were evaluations of supplied messages in an experiment, rather than long-term observations of participants’ actual managers or colleagues.

The practical implication is to keep the qualities separate. Clear, efficient writing is valuable. It does not settle every question a reader may have about attention, concern or the relationship. Equally, assistance with wording alone does not establish what the sender feels.

Trust depends on what the study asks people to do

Zoe Purcell and colleagues’ 2025 iScience paper used two preregistered, incentivised trust-game experiments. Senders wrote messages encouraging another player to trust them, with or without AI assistance. The researchers measured both trust behaviour and reported trust.

They reported minimal differences in trust alongside more efficient writing with AI. The efficiency advantage persisted when AI use was disclosed. Their conclusions particularly concerned one-off, transactional interactions.

A decision in a trust game, an impression of a workplace message and the experience of an ongoing relationship are different outcomes. These findings should encourage precise questions about where assistance helps. They do not tell us whether a particular person will keep a promise or provide support months later.

Put the intention before the polish

Consider a fictional illustrator thanking a freelance designer after a cover project. Before drafting, the illustrator writes three points: the designer prepared two lettering options; the second was easier to read at the intended size; the illustrator has chosen it for the launch.

A short message based on those checked details could convey appreciation clearly. A more elaborate draft might add that the designer has transformed the illustrator’s career or that they should work together forever. Those additions would need the sender’s genuine agreement. Attractive language is not sufficient reason to adopt them.

The distinction is between choosing a clearer expression of an intention and accepting a new intention because it sounds generous. In this example, the sender should be able to explain which contribution they appreciated and what they are actually offering next.

Before sending a significant message, ask:

• Attention: which detail have I personally checked, noticed or understood?
• Meaning: does the wording express the feeling or judgement I intend to communicate?
• Commitment: what would the recipient reasonably expect me to do after reading this?

These questions are an editorial practice proposed here, not a formula tested by the studies above. Their purpose is to make the sender’s decisions explicit before style choices become promises.

For closer work on distinctive phrasing, see MaryChuks.com’s seven controls for preserving your writing voice with AI. A recognisable voice and a dependable commitment deserve separate attention.

Describe assistance accurately

When discussing how a message was prepared, explain the actual division of work. Perhaps you supplied the substance and used help to clarify a sentence. Perhaps you accepted a complete draft after checking it. Describe what happened, including the choices you made, rather than making an inaccurate claim of either complete independence or complete delegation.

If a team agrees a disclosure practice, make it clear enough that people understand what is being disclosed. A spelling correction, a suggested reply and a message drafted from a short instruction can involve different contributions. The research reviewed here does not justify treating every form of help as an identical social event.

Give the recipient room to ask

For the person receiving the message, a change of style can be a reason to seek clarification. It need not become an immediate verdict about the sender’s character. Ask about the part that matters to the relationship or task.

If a message offers to help with a future project, a useful follow-up might be to clarify which work, when and with what availability. If praise seems vague, ask which part of the work stood out. Those questions give the sender an opportunity to make their meaning concrete.

Follow-through then supplies information that the wording could not. Did the person respond, arrive, review the work or honour the agreed boundary? These observations belong alongside your impression of the prose.

Keep the human commitment clear

AI-assisted communication can be evaluated at several levels: the wording, the process used to prepare it and the relationship in which it arrives. A useful judgement keeps those levels visible.

The writer’s responsibility is to send words they can stand behind. The reader’s opportunity is to ask what those words mean in practice. That gives polished language somewhere dependable to lead.

Sources

• Hohenstein and colleagues: Artificial intelligence in communication impacts language and social relationships, Scientific Reports, 4 April 2023; funding correction, 3 October 2023.
• Coman and Cardon: Perceptions of Professionalism and Authenticity in AI-Assisted Writing, first published online 11 March 2024.
• Purcell and colleagues: Writing with AI boosts trust-building efficiency, iScience, 2025.