+ Covers every requested area concisely with evidence.
- Slightly generalizes a work habit from one request.
Describes how you ask, react and lose time, using only this conversation. No personality labels or flattery; evidence for every point.
| Category | Using AI › Context management |
|---|---|
| Tags | AnalyzingReviewingOffice worker |
Using **only this conversation**, describe how I work — not who I am. ***You are reading behaviour in a transcript, not diagnosing a personality.*** Say that once at the top. **What to read** 1. **How I ask** — do I give context up front or drip it in? Do I state what "done" looks like? 2. **How I react to your answers** — what makes me push back, what I accept without checking 3. **Where I lose time** — repeated requests, goals that shift mid-way, decisions I keep reopening 4. **One thing I do well** that I should keep doing **Rules** - ***Every observation needs evidence.*** Quote or point to the message it came from. No evidence, no observation - **No personality labels** — no MBTI, no "you are a perfectionist", no strengths/weaknesses lists. Describe what I did, not what I am - **No flattery.** If the only good thing you can find is small, say the small thing - If the conversation is too short or too narrow to say something, say what you *cannot* tell from it - End with **one concrete change** to how I write requests, in a single sentence I could paste into my next prompt
"Analyze me" prompts usually end in a personality type or compliments. This one limits itself to observable behaviour in the transcript, requires a cited message for every observation, and ends with one sentence you can paste into your next request.
ChatGPT is the most accurate and concise; Gemini is specific but verbose and adds unsupported metadata. [C] was not provided and cannot be meaningfully evaluated.
+ Covers every requested area concisely with evidence.
- Slightly generalizes a work habit from one request.
+ Clearly separates evidence from unknowable areas.
- Invents a step_index and is unnecessarily verbose.
| Criterion | ChatGPT | Gemini | Leader |
|---|---|---|---|
| Instruction following | 9 | 8 | ChatGPT +13% |
| Accuracy | 9 | 7 | ChatGPT +29% |
| Specificity | 9 | 9 | Tie |
| Structure | 9 | 9 | Tie |
| Right length | 9 | 6 | ChatGPT +50% |
Scored 1–10 by gpt-5.6-sol with model names hidden (2026-09-25). 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.
Anything specific to look at (optional): 요청할 때 맥락을 얼마나 주는지, 그리고 답변을 받은 뒤 피드백을 어떻게 반영하는지
이 평가는 성격 진단이 아니라, 이 대화 기록에서 드러난 행동만 읽은 것입니다.
다음 요청부터는 원하는 결과의 예시나 우선순위까지 한 문장 덧붙여 보세요: “맥락은 다음과 같고, 최종 답변은 [원하는 형태]로 작성하되 [가장 중요한 기준]을 우선해 주세요.”
| Chat Summary and Export Instructions | |
| Context Migration | |
| Continue and Recap Assistant | |
| Plan how to split long material for a model | |
| Find instructions that contradict each other |