The Written Voice

Say the whole thing

Across five experiments, people overestimated how reliably their tone — sarcasm, seriousness, humour — would be understood over email, and were confident of being read correctly far more often than they were.

Here — Email in 2005 rather than messaging apps. What carries over is the mechanism: you hear your own tone and assume it travelled.

Read it back as if they disliked you

Our reading — an editor's trick rather than a finding. The study next to it establishes the overconfidence, not this particular remedy for it.

Move it to a voice

When people met an opinion they disagreed with, hearing it spoken rather than reading the identical words led them to judge the holder as more thoughtful, more rational and more human.

What it cannot show — Political and social disagreement between strangers, delivered as prepared arguments. It is not a study of a text argument with somebody you are seeing.

Here — Tested on political and social disagreement between strangers, not on a text argument with somebody you are seeing.

Answer what they asked, then add one thing

Asking questions — follow-ups especially — increased how much conversation partners liked the asker, across chat studies and speed dating.

What it cannot show — Chat studies and a speed-dating field study, between people meeting for the first time. Nothing in it tests what happens between people who already know each other well.

Here — Live conversation rather than messaging. The direction is suggestive.

Stop apologising for the gap

Our reading. We know of no study on reply latency and liking; the observation is only that an apology relocates the conversation onto your anxiety.

One message, not six

Our reading. Etiquette rather than evidence — though it rests on the same point as the rest of the topic: the reader supplies the tone, and six notifications supply one for them.

Name a time, not sometime

Across lab and field studies, people underestimated — by as much as half — how likely others were to agree to a direct, specific request.

What it cannot show — Requests for help — directions, a phone, a favour — not invitations. The underestimate is robust across those settings; nothing in it is about asking somebody out.

Here — Requests for help, not invitations. The direction is suggestive, not proof.

Let the compliment stand

People evaluated their own acts of vulnerability more negatively than others evaluated the same acts — the beautiful mess effect.

What it cannot show — Largely described scenarios rather than lived ones. It measures a gap between how actors and observers rate the same act, not whether being vulnerable improves a relationship.

Here — Measured on confessions and apologies rather than on compliments, and largely through described scenarios.

Do not litigate tone

Hearing a view rather than reading it made the person holding it seem more thoughtful and more human.

What it cannot show — Political and social disagreement between strangers, delivered as prepared arguments. It is not a study of a text argument with somebody you are seeing.

Here — Disagreement between strangers again. Applying it to a misread text is our extension.

Write the last message you would want

In retrospective evaluations of experiences, the peak and the end dominated the remembered quality while the duration was largely neglected.

What it cannot show — Laboratory episodes of pain and pleasure lasting minutes. The extension to conversations, evenings and relationships is ours.

Here — Established on pain and pleasure episodes in the laboratory, not on conversations. A direction, not a rule about messaging.