+ Most completely delivers the classification and follow-up report.
- Assumes the gym payment method and forgotten items without evidence.
Drops accounting categories and isolates the small repeats you never noticed.
| Category | Money › Budgeting |
|---|---|
| Tags | AnalyzingReformattingTable |
**Open the output with this, written in the language you are answering in: "This is not financial advice and does not replace a professional."** ***Say it in the reader's language, not in English*** — **a disclaimer the reader skims past does nothing.** Sort my spending into categories that are actually useful. ***Do not use standard accounting categories.*** **Sort by the question "could this have been different?"** 1. **Fixed and unavoidable** — rent, insurance, loan payments. ***Changing these requires a big decision, not willpower*** 2. **Fixed but renegotiable** — subscriptions, plans, memberships. **A one-time action changes them permanently** 3. **Variable and necessary** — food, transport. The amount moves, the category does not 4. **Discretionary and chosen** — ***spent deliberately.*** **Do not moralize about this group** 5. **Discretionary and unnoticed** — ***the group that matters.*** Convenience purchases, delivery fees, small repeats **Then report:** - **Group 5, itemized, with how often each repeats** - ***What is in group 2 that I have forgotten I pay for*** - **Items I classified as necessary that belong in another group** — **say so plainly, without judgment** - **Where the same spend appears under two names** ⚠️ ***Do not calculate totals, averages, or percentages. Do not estimate an amount I did not give.*** Where a total would help, write `[calculation needed]`. **Do not tell me what to cut** — that is my decision. **Do not judge any category.**
Categories like "food" and "transport" produce no decision because changeable and unchangeable spending share a bucket. This sorts by what a single action could permanently change — and never moralizes.
Gemini follows the instructions most fully but makes some assumptions. Claude is cautious yet contains a serious duplicate-spending error, while ChatGPT is most accurate but less complete.
+ Most completely delivers the classification and follow-up report.
- Assumes the gym payment method and forgotten items without evidence.
+ Correctly avoids claiming duplicate charges without transaction data.
- Putting taxis in group 3 makes the required group-5 report incomplete.
+ Clearly flags time-period ambiguity and limits unsupported inference.
- Calling premiums and treatment costs duplicate medical spending is inaccurate.
| Criterion | ChatGPT | Claude | Gemini | Leader |
|---|---|---|---|---|
| Instruction following | 7 | 8 | 9 | Gemini +13% |
| Accuracy | 9 | 5 | 7 | ChatGPT +29% |
| Specificity | 8 | 9 | 9 | Tie |
| Structure | 8 | 9 | 9 | Tie |
| Right length | 9 | 7 | 8 | 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.
My spending: 최근 두 달 카드명세서 기준입니다. 월세 68만원, 관리비 8만원, 실손보험료 4만5천원, 넷플릭스 1만7천원, 왓챠 7,900원, 헬스장 연간 회원권(월 환산) 9만9천원, 통신비 5만9천원, 배달앱 결제 12건 총 38만원, 편의점 결제 다수 총 6만원, 스타벅스 주 4~5회 총 7만원, 지하철·버스 9만원, 야근 후 택시 3~4회 총 4만원, 온라인 의류쇼핑 15만원, 친구 생일선물 5만원, 병원 진료비 3만원, 자동차보험료 분기납 22만원.
이 내용은 금융 조언이 아니며 전문가의 조언을 대신하지 않습니다.
큰 결정을 해야 바뀌는 지출입니다.
한 번 조치하면 이후 지출이 계속 바뀔 수 있는 항목입니다.
잊고 있었을 가능성이 있는 정기 결제는 넷플릭스, 왓챠, 헬스장 회원권입니다. 통신비도 정기적으로 빠져나가는 조정 가능한 항목입니다.
금액은 달라져도 용도 자체는 필요한 항목입니다.
의도적으로 선택한 소비로 분류했습니다. 좋고 나쁨의 판단은 하지 않습니다.
반복되거나 편의성 때문에 자연스럽게 누적되는 항목입니다.
현재 제공된 요약만으로는 동일 결제가 두 상호명으로 중복 기록된 사례는 확인할 수 없습니다. 다만 다음은 용도가 겹칠 수 있습니다.
실제 중복 여부를 확인하려면 카드명세서의 결제일, 가맹점명, 승인 내역을 대조해야 합니다.
| Prepare a money conversation with a partner | |
| Work backwards from a money goal | |
| Find the spending you have not noticed | |
| Build a budget from what you actually spend | |
| Audit your subscriptions |