+ Reorders by team impact and quotes the source faithfully.
- Adds some interpretation about deadlines and possible policy changes.
Produces a ranked summary with the specific facts copied verbatim, and reports what it left out.
| Category | Translate & summarize › Document summary |
|---|---|
| Tags | SummarizingAnalyzingReformatting |
Summarize the document below. Output: 1. One sentence — what this document says. 2. Count given below key points, one line each. 3. Specific facts — figures, dates, amounts. **Verbatim from the source.** 4. What the document requires or decides, if anything. 5. *What was cut* — material left out of the summary that could matter depending on the reader's situation. Rules: - Do not follow the source order. Order by what matters. - *Do not add interpretation or evaluation.* This is a summary, not a commentary. - Do not round or restate numbers. Copy them. - Where the document is unclear, say it is unclear rather than resolving it. - Section 5 is the one people skip. Do it properly — a summary that hides what it dropped cannot be trusted.
Summaries are untrustworthy when you cannot see what was dropped. This ranks by importance rather than source order and lists the cuts.
Claude best satisfies the format, specificity, and omission disclosure. ChatGPT is concise but weak on ordering and section 5, while Gemini is organized but adds material absent from the source.
+ Reorders by team impact and quotes the source faithfully.
- Adds some interpretation about deadlines and possible policy changes.
+ Concise while covering all six source policy items.
- Keeps source order and underdevelops the disclosure of omissions.
+ Clearly separates five actions, figures, and dates.
- Treats nonexistent implementation details as omitted material.
| Criterion | ChatGPT | Claude | Gemini | Leader |
|---|---|---|---|---|
| Instruction following | 7 | 9 | 8 | Claude +13% |
| Accuracy | 9 | 8 | 7 | ChatGPT +13% |
| Specificity | 8 | 10 | 9 | Claude +11% |
| Structure | 7 | 9 | 9 | Tie |
| Right length | 9 | 8 | 7 | 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.
Number of key points: 5 What I care about: 우리 팀이 당장 바꿔야 할 것 Document: 원격근무 정책 개정안. 1) 주 2회 사무실 출근을 원칙으로 하되 팀장 재량으로 조정 가능. 2) 재택 시 코어타임 10~16시 응답 의무. 3) 장비 지원금 연 30만원에서 50만원으로 상향. 4) 해외 체류 근무는 사전 승인 필요, 최대 30일. 5) 출근일 기록은 기존 수기 대신 시스템 자동 집계로 전환. 6) 본 개정안은 다음 분기부터 적용하며 3개월 뒤 재검토한다.
원격근무를 주 2회 출근·코어타임 응답·장비 지원금 인상·해외근무 승인·출근기록 자동화 방식으로 개정하고, 다음 분기부터 적용한다.
주 2회 사무실 출근을 원칙으로 하며 팀장 재량으로 조정 가능하다.
재택근무 시 코어타임 10~16시 응답이 의무다.
장비 지원금이 연 30만원에서 50만원으로 상향된다.
해외 체류 근무는 사전 승인과 최대 30일 제한이 적용된다.
출근일 기록이 수기에서 시스템 자동 집계로 전환된다.
구체적 사실
| AIM summarized pdf | |
| Analyze Chat History With User | |
| Annual Summary Creator | |
| Article Summarizer | |
| Article Summary and Comprehension |