AI-Assisted Follow-Ups: Leave Room for the Liking Gap

The conversation ends, and the editing begins in your head. You remember the sentence that wandered, the answer you gave too quickly and the pause before the other person left. By the time you consider a follow-up, an ordinary meeting has become a case against contacting them again.

An AI assistant offers an inviting next step: describe the encounter and ask whether they liked you. Yet a more useful task may be to prepare a considerate message while leaving that question unanswered. Research on the “liking gap” gives us a reason to examine negative assumptions, without replacing them with a machine-generated promise of approval.

What the liking gap research found

In research first published online on 5 September 2018, Erica Boothby, Gus Cooney, Gillian Sandstrom and Margaret Clark compared people’s estimates of how much conversation partners liked them with those partners’ own reports. Across the studied settings, participants underestimated how much they were liked and how much others enjoyed their company.

The work included strangers meeting in a laboratory, members of the public at a personal development workshop and students getting to know dorm mates. The researchers called the discrepancy the liking gap. It is evidence that a person’s negative estimate can be mistaken; it cannot establish what any particular acquaintance thinks of you.

A later paper by Adam Mastroianni and colleagues, available online on 27 November 2020 and published in a January 2021 journal issue, extended the investigation to groups and teams. It reported a liking gap after small-group conversations and in engineering project teams. A study of working adults linked the gap with workplace relationship outcomes. Those associations should not be read as proof that correcting a belief will improve team performance.

Neither paper tests the AI drafting approach proposed here. Applying their findings to a creator’s follow-up routine is an editorial suggestion, with a modest purpose: make room for uncertainty before deciding what to do.

Separate the encounter from the verdict

Consider Nneka, a fictional illustrator who meets a workshop organiser at a creative event. They discuss accessible drawing materials. The organiser asks which paper she uses, mentions a forthcoming family session and leaves when the next talk begins. Nneka remembers stumbling over her introduction and concludes that the organiser wanted to escape.

The stumble belongs in her account of what happened. The claim about wanting to escape is an interpretation. The organiser’s departure at the start of a talk is compatible with several explanations. Listing those possibilities does not prove the most favourable one; it stops one explanation from masquerading as the only available fact.

Before involving AI, Nneka could make a short note with three parts: what was said or done, what she inferred and what she wants to do next. Her aim might simply be to share the paper supplier she mentioned. She does not need a judgement about her overall likeability to fulfil that small intention.

Give AI a task it can be checked against

An unhelpful request would ask the assistant to decide whether the organiser secretly disliked her. In this example, the supplied account contains no direct report of the organiser’s feelings. A confident answer would add certainty that the account cannot support.

A more bounded instruction could read: “Using these notes, separate observations from interpretations. Do not infer the organiser’s private feelings. Draft a short follow-up offering the resource we discussed, without assuming a future collaboration.” This is a proposed prompt, rather than a validated psychological intervention.

Nneka can then check the output against her notes. Did the organiser actually ask for a proposal? Was a workshop booking discussed, or only a type of material? Has the draft converted a friendly exchange into an agreement? Editing those details is more concrete than asking the assistant to certify that the meeting went well.

Write from shared context, with a small next step

A possible message for this fictional encounter is: “It was good to meet you at the drawing event. You asked about the paper I use for beginner sessions; I’m happy to send the supplier details if useful. I hope the family workshop goes well.”

The message names the encounter, refers to something actually discussed and offers one manageable next step. It leaves the recipient free to decide whether the resource would help. It does not require the organiser to reassure Nneka about the impression she made.

This wording is an example to adapt, not a claim that this style produces more replies. A different exchange may justify attaching the promised link immediately, asking a specific practical question or sending no message at all. The relevant issue is whether the action fits the conversation and the relationship.

Do not turn uncertainty into a chase

The liking gap is not permission to disregard an explicit refusal or a request for no further contact. A research pattern cannot cancel a person’s stated boundary. Nor does being liked imply interest in a commission, available time or willingness to collaborate.

If the organiser does not reply, Nneka still lacks a reliable explanation for the silence. She can decide whether any practical obligation needs resolving and otherwise leave the exchange there. Repeatedly asking AI for more persuasive versions would not supply the missing information about the recipient’s wishes.

Where a project genuinely requires a response, keep the follow-up tied to that requirement: the decision needed, the agreed deadline and the next operational step. A social impression and an outstanding work decision should not be treated as the same question.

Review the action, rather than scoring your personality

• Did the draft represent the original exchange accurately?
• Did it make one clear, proportionate offer or request?
• Did it avoid inventing enthusiasm, commitments or private thoughts?
• Did it respect any boundary already expressed?
• Can you stand behind the wording without an AI prediction of the reply?

MaryChuks.com’s guide to asking for help without predicting the answer explores a related distinction between making a request and anticipating someone else’s decision. Here, the same practical restraint applies after a first conversation.

For an independent creator, there is value in being able to take a small, appropriate social step without first solving every uncertainty about how it will be received. A follow-up can be accurate, warm and easy to decline while the other person’s opinion remains unknown.

The useful lesson from the liking gap is a reason to question an overly certain negative verdict. Let AI help organise the account and improve the message. Leave the other person room to supply their own response.

Sources

• Boothby and colleagues: The Liking Gap in Conversations: Do People Like Us More Than We Think? Psychological Science, first published online 5 September 2018.
• Mastroianni and colleagues: The liking gap in groups and teams. Organizational Behavior and Human Decision Processes, January 2021; available online 27 November 2020.

Sources checked on 4 November 2026. Nneka, the encounter, prompt, message and review routine are illustrative.