AI Conversation Practice: Make Room for a Simple Compliment

At an imagined makers’ fair, a visitor notices a beautiful arrangement of colour in an artist’s print. She wants to say so, then hesitates. Perhaps the compliment will sound clumsy. Perhaps the artist will feel obliged to chat. She opens an AI assistant to practise, and a small expression of appreciation begins turning into a performance to perfect.

There is another way to approach that moment: separate uncertainty about your delivery from evidence about how another person might receive a compliment. Research on compliments and gratitude suggests that people can underestimate their positive impact. It offers a reason to examine a pessimistic prediction, while leaving the actual recipient free to respond.

What compliment givers predicted

Erica Boothby and Vanessa Bohns explored this mismatch in five studies of compliment giving. Their paper appeared online on 28 August 2020 and in the May 2021 issue of Personality and Social Psychology Bulletin.

Givers underestimated how positively recipients would feel. Other studies in the paper found that givers overestimated how bothered and uncomfortable recipients would feel, including when looking back afterwards. The final study examined givers’ anxiety and concern about their competence; third-party forecasters were more accurate.

These are findings about participants and interactions in the studies. They do not establish that every compliment is welcome or that an AI assistant can predict a particular person’s response. The useful distinction is between a giver’s worried forecast and a recipient’s reported experience.

A related finding about expressing thanks

Amit Kumar and Nicholas Epley’s gratitude-letter research, published online on 27 June 2018 and in the September 2018 issue of Psychological Science, asked participants in three experiments to predict recipients’ reactions. Recipients then reported how receiving the letters made them feel.

Writers overestimated awkwardness and underestimated recipients’ positive feelings. Expectations about awkwardness and mood were correlated with willingness to express gratitude. That correlation does not, by itself, establish that changing an expectation will cause someone to send more letters.

A compliment about an artwork and thanks for someone’s help are different acts. Together, these studies invite a question about our forecasts of appreciation; neither tested AI conversation rehearsal. Any rehearsal practice proposed here is an editorial application, not an intervention validated by those papers.

Separate a worried prediction from the situation

Return to the fictional fair. The visitor’s thought that she might stumble over her words concerns her delivery. Her prediction that the artist will be annoyed concerns someone else’s experience. They may feel like a single problem, but they need different checks.

She can decide whether her observation is genuine, whether she is commenting on the work rather than making an intrusive personal remark, and whether the artist is available. An artist serving a customer presents a different opportunity from one inviting people to discuss the display.

The research gives no reason to disregard those circumstances. It also gives no reason to treat a private prediction of inevitable awkwardness as a measured outcome. In this example, the visitor can choose a brief comment about the print and leave further conversation optional.

Her purpose matters too. Expressing appreciation need not become a request for a discount, an introduction or a business opportunity. If she wants something in return, that request should be considered separately rather than hidden inside the compliment.

Rehearse a short exchange without inventing a verdict

A bounded AI practice brief could say: “Help me rehearse a brief compliment about an artwork at a makers’ fair. Use only my observation about the colour arrangement. Offer a concise opening, then roleplay two fictional replies: a short thank-you and an invitation to discuss the print. Keep both scenarios clearly labelled as invented.”

This brief asks for possible conversational paths, not a prediction of the artist’s thoughts. A simulated enthusiastic response is rehearsal material. A simulated awkward response is also rehearsal material. Neither supplies new evidence about the real person waiting at the stall.

The visitor might practise saying that the teal around the window gives the house a quiet feeling. That is her imagined response to an imagined print, not an objective assessment of its quality. She can choose wording she would comfortably say aloud, including a simpler version if the first sounds unlike her.

Practise an ending as well as an opening. A short acknowledgment allows the exchange to finish after a brief reply. If the artist asks what caught her attention, she can explain the particular detail. The aim is enough preparation to make a choice, rather than repeated drafting until every possible reaction looks favourable.

Let the recipient determine how far it goes

A compliment does not reserve another person’s time. In the fictional exchange, the artist might smile and return to serving someone, offer a brief thanks or start a longer discussion. The visitor can respond to what is actually said without trying to prolong the interaction to prove that it went well.

If the artist indicates that conversation is unwelcome, leave it there. Group-level research about positive reactions does not override an individual’s expressed preference. The same principle applies when appreciation occurs between colleagues or within a relationship carrying unequal authority.

For a creator receiving praise, accepting a kind remark need not create an obligation to buy, collaborate or disclose personal information. Appreciation and a subsequent request remain separate decisions. Preserving that freedom makes the proposed practice about giving something small without requiring a particular return.

Review the event at the level you observed

Afterwards, distinguish the original prediction from the event. In our example, “I expected an annoyed response” and “the artist thanked me, then helped another customer” are different records. The latter describes an exchange; it does not establish everything the artist privately felt.

This kind of reflection should remain modest. A single pleasant interaction does not prove that future compliments will all be welcome. A brief interaction does not prove that the compliment failed. If the visitor uses AI to discuss what happened, she can provide the observed words and actions while marking interpretations as uncertain.

For a complementary question about assisted writing, see warm words and human intentions. Here, the additional question is whether hesitation has become a confident forecast about somebody else.

A useful rehearsal leaves room for a real person to surprise you. It helps you say something genuine, notice the setting and accept the response without demanding reassurance. The compliment can then remain what you intended: a small expression of appreciation.

Sources

• Boothby and Bohns: Underestimating the Positive Impact of Our Compliments on Others — online 28 August 2020; May 2021 issue.
• Kumar and Epley: Undervaluing Gratitude — online 27 June 2018; September 2018 issue.