Stop copying prompts. Learn to write them.
A copy-pasted prompt solves today's task and leaves you helpless the moment the situation is slightly different — a different client, a different tool, a different day. Learn the six parts of a prompt that actually works and you can write the exact right one, from a blank page, for anything a remote career throws at you.
“A prompt library makes you dependent on someone else's imagination. Learn the anatomy and you can write a better prompt than any library, for any situation you'll ever face.”
Eight parts. Learn them once, use them forever.
Every strong prompt is made of the same parts — most weak ones are just missing one. Study the difference below, then build with it.
Role
Sets which expertise and judgment the AI borrows before it writes a single word.
Rule — Name a specific role with something at stake — not a job title. 'A recruiter who rejects vague resumes' beats 'a helpful assistant.'
Weak
You are a helpful assistant.
Strong
You are a senior recruiter for remote ops roles who rejects any resume that buries results in duties.
Context
Gives the AI the raw material it needs. Without it, the AI invents facts to fill the gap.
Rule — Paste the actual source — the job post, your real experience, the client's email — never a summary of it from memory.
Weak
I have some experience in customer service.
Strong
Here is my resume: [PASTE FULL TEXT]. Here is the job post I'm targeting: [PASTE FULL TEXT].
Task
The single, concrete action you want performed — the verb that does the work.
Rule — One task, one verb, one object. Split anything with an 'and' in it into two prompts.
Weak
Help me with my resume.
Strong
Rewrite my resume summary and top 3 bullets to match this specific job post.
Constraints
The boundaries that keep the output usable — length, tone limits, and what it must not do.
Rule — State limits as numbers and explicit rules, never adjectives. 'Under 150 words' beats 'concise.'
Weak
Keep it short and professional.
Strong
Under 120 words. No buzzwords like 'synergy' or 'passionate.' Don't invent metrics I didn't give you.
Format
The exact shape the output must take, so you can use it without reformatting it yourself.
Rule — Specify the structure directly — bullets, headers, a table, a script — not just 'organized.'
Weak
Give me some ideas.
Strong
Output as 5 bullet points, each starting with a past-tense action verb, one line each.
Examples
Shows the AI the pattern instead of describing it — the fastest way to close an ambiguity gap.
Rule — Paste one real example of the output you want, even a rough one, and say 'match this.'
Weak
Write it in my style.
Strong
Here's an email I sent last week that landed well: [PASTE]. Match this tone and sentence length.
Voice & tone
Keeps the output sounding like a person — the difference a reader feels in the first sentence.
Rule — Describe your actual voice with specifics — sentence length, formality, phrases you never say — not just 'friendly.'
Weak
Sound friendly and professional.
Strong
Short sentences. No exclamation points. I say 'let's fix this,' not 'I would like to bring to your attention.'
Success criteria
The rubric the AI should judge its own draft against before it ever reaches you.
Rule — Tell it what a good answer does that a bad one doesn't, then ask it to check its own draft against that before handing it over.
Weak
Make it good.
Strong
A good version names one specific result with a number in it. Check your draft against that rule before giving it to me.
Six habits that separate a real prompt from a lucky one
Specificity beats politeness
'Please' and 'thank you' don't improve output. A precise noun does. Trade every 'nicely' for a number, a name, or a real example.
Give it the rubric you'll judge by
Tell the AI what 'good' looks like before it drafts, not after you've read three bad drafts and are guessing why.
Ask for options, not one answer
Request 3 versions and pick the strongest, or merge two. One draft is a guess dressed as an answer; three is an actual choice.
Make it critique its own output
After it drafts, ask 'what's weak about this, and fix it.' AI catches its own generic phrasing better than you'd expect.
Never trust a fact it produces
Names, numbers, dates, quotes — verify every one against a source you can point to. AI states wrong facts as confidently as right ones.
Paste real material instead of describing it
Summarizing your resume from memory loses the exact details that make output specific. Paste the actual text, every time.
Draft, run, diagnose, change one thing, re-run
No prompt works on the first try. What separates people who improve fast is a procedure for the second try — not luck.
- 1
Draft
Write your first version using every anatomy part you have real material for. Don't aim for perfect — aim for complete.
- 2
Run
Send it. Read the whole output once before you judge any single line of it.
- 3
Diagnose
Name exactly what's wrong using the signals below — 'this isn't it' isn't a diagnosis.
Too generic — could apply to anyone
→ Missing Context — you described your situation instead of pasting it.
Right content, wrong shape
→ Missing Format — you never specified structure, length, or template.
Sounds like AI wrote it
→ Missing Voice — and it had no real sample of you to match.
Confident, but wrong
→ Missing Constraints — you never told it what not to invent.
- 4
Change one thing
Add or fix the single part you diagnosed as missing. Resist changing three things at once — you won't know which fix worked.
- 5
Re-run
Send the revised prompt. Compare only the issue you targeted. Still off? Loop back to Diagnose.
What a remote career actually uses AI for
Not finished prompts — situations. Use the anatomy above to build the exact prompt each one needs.
Research
Ask it to summarize a client's industry, a tool's docs, or a company before a call — then verify the two facts that matter most, yourself.
Drafting
Get a fast first version of anything — an email, an SOP, a proposal — then edit it into your own voice before it goes out.
Critique
Paste your own draft and ask it to find the weakest paragraph, the vaguest claim, or the line a client would push back on.
Mock interviews
Have it play a skeptical hiring manager and grade your answers — practice handling pressure, not just reciting a script.
Learning a tool fast
Ask it to explain the 20% of a tool's features that cover 80% of real jobs, then have it quiz you until you can do it cold.
Automation design
Describe a repetitive task and ask it to map the trigger, the steps, and the edge cases — before you touch Zapier or Make.
Build yours now
Fill in what you know about your real situation. This assembles into your prompt, live — not ours.
Fill in what you actually know. Skip what you don't have yet — the meter below will tell you exactly what that gap costs you.
Who should the AI be, and what's the stake if it gets this wrong?
Paste your real material here — not a summary of it.
One verb, one object. What should it actually do?
Numbers and rules, not adjectives.
The exact shape you need back.
How you actually sound — sentence length, what you never say.
A prompt isn't done when it runs. It's done when it survives your own critique.
Judge it before anyone else does
Does it contain a fact I can't verify?
Names, numbers, claims — if you can't point to where it came from, it might be invented. Check before it goes out under your name.
Does it sound like me?
Read it out loud. If you'd never say it that way, it reads as fake to the person on the other end too.
Would I send this exactly as-is?
If you're already planning to edit half of it, the prompt was missing something — go back to the loop instead of hand-fixing forever.
Does it actually follow my constraints?
AI drifts past word limits and sneaks in the words you banned. Check the numbers, not just the vibe.
Is there a better version buried in a longer draft?
Sometimes the third paragraph is the real opening line. Read the whole thing before you extract anything.
Kill AI-slop before a client sees it
"In today's fast-paced world..." or any throat-clearing opener
Fix — Delete the first sentence. Almost every draft is stronger starting with the second one.
Padding three-item lists — 'reliable, efficient, and detail-oriented'
Fix — Cut two of the three. Keep only the one with a number attached to it.
Perfectly balanced sentences with no rough edges
Fix — Add the specific, slightly awkward detail a real person would actually say.
Em dashes doing the work a period should do
Fix — Split it into two sentences.
Words nobody says out loud — leverage, utilize, delve, robust, seamless
Fix — Swap for the plain verb: use, dig into, strong, works.
A closing line that just restates the opening — 'In conclusion...'
Fix — Delete it. If the piece needs a summary to be understood, the structure is the actual problem.
Specimens to dissect, not to copy
Four prompts pulled from an old copy-paste library, kept here only as specimens to dissect — not to paste. Read each for the pattern it uses, close the tab, then write your own version from the rule you found.
Tailor my resume to a job post
SpecimenRewrites your resume to match a specific job's keywords and priorities.
You are an expert resume writer for remote roles. Here is my current experience: [PASTE]. Here is the job post: [PASTE]. Rewrite my resume summary and 5 bullet points to match this role. Use action verbs, quantify results where possible, and mirror the exact keywords from the post so it passes ATS screening. Keep it truthful to my experience.
Dissect — It never says 'be concise' — it says 'quantify results' and 'mirror the exact keywords.' That's Constraints written as rules, not adjectives.
Run a mock interview
SpecimenPractice with realistic questions for your exact role.
You are a hiring manager for a [ROLE] working remotely with a US client. Interview me one question at a time. After each of my answers, give a quick score (1–5) and one concrete improvement, then ask the next question. Cover experience, tools, a scenario/problem-solving question, and 'why should we hire you'. Start now.
Dissect — Format here is procedural, not a document shape — 'one question at a time, score after each answer.' Format can structure an interaction, not just a page.
Turn a messy task into an SOP
SpecimenDocument any repeatable process so you never re-explain it.
Turn this rough process into a clean Standard Operating Procedure: [DESCRIBE THE STEPS]. Output a titled SOP with purpose, tools needed, numbered steps with sub-steps, common mistakes to avoid, and a definition of done. Format it so I can paste it into Notion.
Dissect — 'Definition of done' is Success criteria without naming it. Extract the rule it's using — don't copy the phrase.
Design an automation for a repetitive task
SpecimenLet automation handle the repetitive work.
I do this repetitive task manually: [DESCRIBE]. I have access to [Zapier / Make / n8n]. Design an automation: list the trigger, the steps, the apps involved, and any edge cases to handle. Explain it simply enough that I could build it today, and suggest what to keep manual.
Dissect — 'Suggest what to keep manual' is a Constraint framed as a scope boundary, not a word limit — a different flavor of the same part.