ChatGPT가 6번이나 똑같은 답만 하길래 단어 하나 바꿨더니, 결과가 완전히 달라짐
ChatGPT gave me the same answer 6 times. then i changed one word. everything shifted.
핵심 요약
프롬프트에서 단어 하나만 바꿔도 AI의 답변 방식과 깊이가 완전히 달라질 수 있다는 경험담.
- 프롬프트 단어 선택 — '어떻게'와 '왜'처럼 질문의 의도를 바꾸는 단어 하나가 답변의 깊이를 결정함.
- 답변 방식의 차이 — '해줘'와 '같이 생각해보자'를 구분하면 AI가 단순히 결과물만 내놓을지, 추론 과정을 공유할지 결정됨.
- 피드백의 질 — '이거 좋아?'와 '뭐가 문제야?'를 통해 칭찬 위주의 답변 대신 구체적인 비판과 개선점을 얻을 수 있음.
- 학습 효과 극대화 — '설명해줘' 대신 '내일 남 가르쳐야 하니까 설명해줘'라고 하면 핵심 위주의 명확한 답변을 얻음.
was stuck on a problem for three days.
kept asking variations of the same question. kept getting variations of the same answer. confident. well structured. completely unhelpful.
on day three i changed one word in the prompt.
changed "how" to "why."
"how do i fix this" → same answer sixth time.
"why is this broken in the first place" → completely different response. went three layers deeper. found the actual root cause i'd been circling for 72 hours without naming.
the answer was there the whole time.
i was asking the wrong word.
spent the rest of the week testing single word swaps. here's what i found:
"how" vs "why"
how: gives you the steps.
why: gives you the understanding underneath the steps.
use how when you know what you're solving.
use why when you're not sure you're solving the right thing.
"what should i do" vs "what would you do"
what should i do: produces generic advice optimised for the average person in your situation.
what would you do: produces a perspective. an actual position. something with reasoning behind it instead of balanced optionality dressed up as guidance.
the second one takes a stance. the first one hedges forever.
"give me" vs "help me think through"
give me: vending machine. input request. receive output. done.
help me think through: collaborative. the model shows its reasoning. asks clarifying questions. surfaces assumptions. treats the problem like something worth understanding rather than something worth answering quickly.
completely different experience of the same tool.
"is this good" vs "what's wrong with this"
is this good: yes. here are the strengths. here are some areas for potential improvement.
what's wrong with this: skips the validation entirely. goes straight to the problems. specific. named. no diplomatic cushioning.
one of these produces feedback. the other produces encouragement.
"write me" vs "show me how you'd approach writing"
write me: you get a draft.
