← sources

I'm not an expert. I learned this stuff on my own, Here are 10 prompts I actually use. Tell me if they're trash.

source: reddit (reddit:PromptEngineering)published 2026-10-21 14:18ingested 2026-10-22 02:42

retained text

​

so i got tired of asking AI "what should i do" and getting a polite "it depends" that makes me want to scream. over the last year, i ended up building a small library of specific prompts that actually force the model to be useful.

none of them are magic. they’re just guardrails.

so i got tired of asking chatgpt "what should i do" and getting a polite "it depends" that makes me want to scream. over the last year, i ended up building a small library of specific prompts that actually force the model to be useful.

none of them are magic. they are just guardrails.

here are 10 of them that i use on my phone when i actually need to get things done:

\# 1. the "two advisors" one

\* \*\*the prompt:\*\* \`Act as two advisors analyzing \[decision\]. Advisor A argues for \[option 1\] as a \[role with a stake\]. Advisor B argues for \[option 2\] as a different \[role with a stake\]. Each gives 3 arguments with reasoning. Then write a one-paragraph reconciliation stating which arguments survive scrutiny and what recommendation follows. Flag any claim that needs verification.\`

\* \*\*why i use it:\*\* because asking "what should i do" gets me a polite "it depends" and i want to scream. this one gives me two people arguing in my phone, like a tiny board meeting. i used it to decide whether to quit my job. it did not tell me to quit, it told me to ask for a raise first. i did, and i got it.

\* \*quick question:\* tell me if i am wrong  i feel like this one only works if the two advisors have real conflicting incentives. if they are both like "here are the pros and cons" it is useless. whats your take on that?

\# 2. the "roast my draft" one

\* \*\*the prompt:\*\* \`Act as a strict auditor with no incentive to be polite. Review \[document\] against: accuracy, logic, missing evidence, risk, clarity, feasibility. For each weakness, state the problem, rate it (Critical / Major / Minor), and give a specific fix. Do not praise. If something is fine, say "no issue." End with the 3 issues that most threaten the outcome.\`

\* \*\*why i use it:\*\*  AI will compliment a garbage draft if you let it. i learned this the hard way when i sent a proposal with a math error because i asked "does this look good" and it said yes. now i use this prompt. it does not say yes, it says "Critical: your numbers do not add up." which is the reality check i need.

\* \*wondering:\* does this work for everyone or does it just work for me because i need to be humbled? lmk.

\# 3. the "explain it to my mom" one

\* \*\*the prompt:\*\* \`Explain \[topic\] to a \[role\] who has \[knowledge level\] and needs it to \[make a decision or take an action\]. They already know: \[X\]. They do NOT know: \[Y\]. Avoid jargon unless you define it in one clause. Use one analogy from their world. End with the single action they should take.\`

\* \*\*why i use it:\*\* "explain like i am 5" is too vague. this tells it exactly how much the person knows and what they need to do. i used it to explain api rate limits to my friend in marketing, and now she thinks i am a developer. i am not. but the "they already know X, they do not know Y" part is the secret.

\# 4. the "just answer me" one

\* \*\*the prompt:\*\* \`Answer directly. No greetings, no praise, no "great question," no restating my prompt. Start with the answer in the first sentence. Then only the support it needs. If uncertain, say "uncertain:" and why. If the answer is "it depends," name the deciding variable. No closing summary unless I ask.\`

\* \*\*why i use it:\*\* sometimes i ask a yes/no question and get four paragraphs of preamble. when i am standing in line at the pharmacy, i do not need a friend, i need a fact. this prompt gives me a fact.

\# 5. the "argue against me" one

\* \*\*the prompt:\*\* \`Argue against this plan as a well-informed skeptic who wants it to succeed: \[plan\]. Give the 3 strongest objections, each with the mechanism by which it causes failure and the evidence behind it. Distinguish "likely to fail" from "could fail." Then state what would have to be true for the plan to work despite each objection. Do not invent facts.\`

\* \*\*why i use it:\*\* my friends will not tell me my idea is bad because they love me. chatgpt will if i ask it right. this one saved me from launching a notion template nobody asked for because it forced me to realize nobody actually needed it.

\* \*question for you:\* is this one too harsh, or do you need the reality check more than the optimism?

\# 6. the "what actually matters" one

\* \*\*the prompt:\*\* \`For \[topic\], identify the \~20% of things that produce \~80% of the outcome for \[goal\]. For each, state the item and one sentence of justification. Then list what you deliberately excluded and why it's low-impact. If the 80/20 split doesn't hold, say so.\`

\* \*\*why i use it:\*\* because i do not need the whole internet worth of advice, i need the 3 things that actually matter. the "what i excluded" part is where the magic is because knowing what not to do is 90% of life.

\# 7. the "assume im wrong" one

\* \*\*the prompt:\*\* \`Before answering \[question\], list the assumptions hidden in it — about scope, data, time horizon, and definitions. For each, say whether it's safe, risky, or needs checking. Then answer the question, labeling which assumptions you rely on. Flag any assumption that, if wrong, would change the answer.\`

\* \*\*why i use it:\*\* i once asked how to get more followers and it told me i was assuming i needed more followers when i actually just needed to talk to the ones i had. it challenges the premise of your question, and sometimes the premise is the problem.

\# 8. the "quote it or drop it" one

\* \*\*the prompt:\*\* \`Using only \[provided source\], extract \[fields/answers\]. For each, quote the exact line supporting it (with location if available), or write "Not stated." Separate source-supported facts from inference and label inference as "inference." Do not invent details.\`

\* \*\*why i use it:\*\* because sometimes you need proof and not vibes. if the source does not say it, the answer is "not stated." it refuses to make things up, which is a lifesaver for contracts and dense text.

\# 9. the "five why" one

\* \*\*the prompt:\*\* \`Apply 5 Whys to \[symptom\]. For each "why," state the proposed cause, whether it's EVIDENCE or HYPOTHESIS, and how to verify it. Stop early if the chain becomes speculative — say so. End with the most likely root cause and the single test that would confirm it.\`

\* \*\*why i use it:\*\* because the first "why" is never the real answer. the fifth "why" is the one that makes you rethink everything. the evidence vs hypothesis filter stops the model from just guessing five times.

\# 10. the "what would i regret" one

\* \*\*the prompt:\*\* \`Reframe \[decision\] three ways: (1) what would a version of me 5 years ago have missed about this? (2) what am I overweighting right now? (3) what would a version of me 5 years from now wish I had done? For each, give 3 bullets. End with the one insight that only the future perspective reveals.\`

\* \*\*why i use it:\*\* because i am bad at thinking long term. this prompt makes past me, present me, and future me all weigh in at once.

ok so thats  just things i learned by trying stuff and failing and trying again.

if this post doesnt get destroyed in the comments ill drop the link to the free preview next week. let me know which one of these you're going to steal first.

document fetched 2026-10-22 00:42 · canonical https://reddit.com/r/PromptEngineering/comments/1xbofrk/im_not_an_expert_i_learned_this_stuff_on_my_own