+ Best follows the six-line format and itemizes every cut.
- Drops the thanks and ends awkwardly with an apology.
Surfaces the buried request, cuts background, and keeps a restorable list of what was removed.
| Category | Office work › Email |
|---|---|
| Tags | SummarizingRewritingOffice worker |
| Updated | · 1 change |
Shorten this email. Rules: 1. **First find what is actually being asked.** If it is buried mid-body, move it to the top. 2. Cut background by at least half. Delete entirely anything the recipient already knows. 3. Target the target line count. 4. **Never cut the request, the deadline, or the contact.** 5. Below the shortened version, list **what you cut**, so I can restore any of it. 6. If a section would be better as an attachment, say so. Then: whether the email still works if the reader stops after three lines. *Most do.* **Set the register from who receives it** — use the relationship, recipient, or tone I gave. ***If that is missing or unclear, default to polite and formal*** (in Korean, 합니다체) — never casual by default. Keep it a complete email: a greeting, the body, a closing line, and a sign-off. *When shortening, do not cut the greeting or the closing to save lines.*
Long emails get skimmed and the request is missed. This promotes the ask, halves the background, and never cuts the deadline.
Claude is the best-balanced and most compliant. ChatGPT is concise and strongest within three lines but misses format requirements, while Gemini is specific yet overly long.
+ Best follows the six-line format and itemizes every cut.
- Drops the thanks and ends awkwardly with an apology.
+ Fits both questions and the deadline into line two.
- Misses six lines and does not list the removed content.
+ Puts the request and deadline first with specific cut notes.
- Overexplains and misjudges the three-line test.
| Criterion | ChatGPT | Claude | Gemini | Leader |
|---|---|---|---|---|
| Instruction following | 7 | 9 | 8 | Claude +13% |
| Accuracy | 9 | 9 | 7 | Tie |
| Specificity | 9 | 9 | 9 | Tie |
| Structure | 8 | 9 | 8 | Claude +13% |
| Right length | 9 | 8 | 5 | ChatGPT +13% |
Scored 1–10 by gpt-5.6-sol with model names hidden (2026-09-24). This is an AI review, not a measurement.
We gave three models the same input and copied their answers unedited. Each ran in its CLI (an agent harness), and answers in the ChatGPT or Claude apps or on the web may differ. Outputs are in Korean.
Target lines: 6줄 Source text: 안녕하세요 김대리님. 지난번 회의에서 말씀 주신 건 관련하여 연락드립니다. 아시다시피 저희 쪽에서 준비하던 자료가 있는데요, 그 자료를 준비하는 과정에서 몇 가지 확인이 필요한 부분이 생겼습니다. 우선 지난달 실적 데이터를 어느 기준으로 집계하는지가 명확하지 않았고, 또 하나는 부서별 구분을 기존 방식대로 할지 새 방식으로 할지입니다. 바쁘신 와중에 번거롭게 해드려 죄송하지만 이번 주 목요일까지는 회신을 주셔야 자료를 마감할 수 있을 것 같습니다. 감사합니다.
안녕하세요, 김대리님.
자료 마감을 위해 ① 지난달 실적 데이터 집계 기준과 ② 부서별 구분 방식(기존/새 방식)을 이번 주 목요일까지 확인 부탁드립니다.
회신 주시면 자료를 마감하겠습니다.
감사합니다.
[보내는 사람] 드림
잘라낸 내용: 지난번 회의 언급, 자료 준비 배경, “아시다시피”, 확인이 필요하게 된 경위, 사과 표현, 반복적인 완곡어법을 삭제했습니다. 첨부가 필요한 내용은 없습니다.
세 줄에서 멈춰도 수신자, 요청사항, 마감일, 회신 목적이 모두 전달됩니다.
| Date | What changed and why |
|---|---|
| Register now follows the recipient, defaulting to polite when unknown. |
| Deliver bad news to a client by email | |
| Write a business email in English | |
| Chase an unanswered request by email | |
| Write an email that declines clearly | |
| Write a chase email at three escalation levels |