how to write better ai prompts.
most prompts fail for the same three reasons, and none of them is that you are bad at AI.
the difference between a vague prompt and a good one is specificity, a real example of what you want, and a clear role for the AI. add those and the output goes from generic to usable. this is the "description" skill in anthropic's ai fluency framework, one of the four skills that decide whether AI works for you (see the 4Ds). here are six techniques, with before-and-afters you can copy.
why prompts fail
you told the AI what to make, but not who it is for, how long it should be, in whose voice, or what "good" looks like. so it filled every gap with the most generic default it had. the fix is not a magic phrase. it is removing the guesswork.
the 6 techniques
1. give it context
say the scope, the audience, and the constraints. do not make it guess.
before: "write me a welcome email."
after: "write a welcome email to someone who just bought a $147 AI starter pack. one goal: get them to open it today. friendly, direct, under 90 words, no exclamation points."
2. show an example of what "good" looks like
one or two examples teach a pattern faster than any list of adjectives. paste something in your real voice.
before: "make it sound like me."
after: "here are two subject lines i've written: 'the part nobody tells you' and 'this took me 4 years to learn.' now write five more in that voice for a post about AI adoption."
3. specify the output
state the format, the length, and the structure. this is the fastest quality jump for most people.
before: "give me some content ideas."
after: "give me 10 reel hooks as a numbered list. each under 12 words. each has to name a specific tool and a specific outcome. no questions."
4. break complex tasks into steps
number the steps you want it to follow. it stops rushing and gets methodical.
before: "analyze my sales."
after: "analyze last month's sales. step one, name the top three products. step two, compare to the prior month. step three, flag anything odd. step four, give me your best guess at why."
5. ask it to think first
tell it to reason through the problem before answering. thinking space produces more considered output, especially on anything with tradeoffs.
"before you answer, think through the constraints and a few different approaches, then recommend one and say why."
6. give it a role
a role sets tone, depth, and audience in one line. this is the highest-leverage single move.
"you are my skeptical head of sales. read this offer and tell me the three reasons a buyer would not click. be blunt."
the move most people never try
when you do not know how to ask, ask the AI to write the prompt with you. this is the one that changes how people work, and almost nobody uses it.
"i'm trying to get you to help me with [goal]. i'm not sure how to phrase it to get the best result. can you help me write an effective prompt for this, and ask me anything you need first?"
it will ask you the questions you forgot to answer. that is context, examples, and constraints, gathered for you.
before and after, at a glance
| weak prompt | what to add |
|---|---|
| "write a caption" | audience, goal, length, voice, one example |
| "summarize this" | for whom, how long, what to keep, what to cut |
| "is this good?" | a role and criteria: "as an editor, score clarity 1-10 and fix the weakest line" |
| "help me with my offer" | break into steps, or ask it to write the prompt with you |
one caution
a better prompt gets you a better draft. it does not make the draft correct. AI will hand you a confident, wrong answer in a beautiful format. that is why prompting is only one of the four AI fluency skills, and why you keep a human on the send. read the full framework in ai fluency: the 4 skills.
skip the prompt-of-the-day treadmill
the free system builder helps you set AI up once, in your voice, on your real work, so you stop rewriting the same prompt every time. built in claude, today.
get the free system builderor just follow along. new field notes most weeks on x, instagram, and tiktok.