Why your prompts might not be working
If you have ever typed a question into Claude or ChatGPT, gotten something technically correct but completely unhelpful, and quietly concluded that AI is overhyped, then I want to suggest that the tool probably was not the problem. The most common reason AI gives mediocre output is that it was given a mediocre brief.
Think about how you would describe a task to a new employee on their first day. You would not say “write me an email.” You would say who it is going to, what you need it to accomplish, what tone to strike, what not to include, and roughly how long it should be. AI needs the same thing. The difference is that a new employee will ask follow-up questions if you are vague. AI will just fill in the gaps with its best guess, and its best guess is, by definition, generic.
The good news: once you understand what information AI is missing, the fix is usually just adding a sentence or two to what you were already writing. It is not a technical skill. It is a communication skill, and you already have the building blocks.
The four things every useful prompt includes
You do not need a framework or a course or a special syntax. You need to answer four questions before you hit send. Here they are.
1. What is the job?
Be specific about the task. “Write an email” is a job. “Write a follow-up email to a client who attended our webinar last Tuesday but has not responded to my initial outreach” is a much better description of the same job. The more precisely you describe what you need, the less AI has to guess.
A useful test: could two different people read your prompt and picture two completely different outputs? If yes, add more specificity. If the task is unambiguous, you are close enough.
2. Who is it for?
Audience shapes everything: tone, vocabulary, level of detail, what to explain and what to assume. “My executive team” and “a first-time customer who has never heard of us” are two completely different audiences, and a good prompt tells AI which one it is writing for. If you have relevant details about the specific person: their name, their role, their history with your business, then include those too. The more AI knows about the reader, the better it can calibrate what it writes.
3. What does good look like?
This is the question most people skip, and it is the one that makes the biggest difference for longer or more important outputs. Tell AI what success looks like. How long should the output be? What tone are you aiming for: formal, conversational, direct, warm? Are there things it should definitely include, or things it should definitely avoid? If you have an example of something similar that worked well, share it.
You do not need to answer every one of these for every prompt. For a quick task, a sentence or two of guidance is enough. For something that matters: a client proposal, a staff communication, a policy document, spending an extra two minutes on this section will save you ten minutes of editing on the back end.
4. What role should AI play?
This one is optional but powerful. Telling AI what perspective or expertise to bring to a task often improves the output significantly. “You are a plain-language editor helping me simplify a technical document for a non-technical audience” will get you a different result than just handing over the document. “You are a skeptical investor reviewing this business plan for weaknesses” will surface problems your own review would miss. “You are a patient tutor explaining this concept to someone who finds it confusing” shifts the register of the explanation entirely.
The role does not need to be elaborate or formal. One sentence setting the perspective is usually enough to change the quality of the response meaningfully.
Seeing it in practice
Here is the same request written three ways, from vague to specific, so you can see the difference these four elements make in real terms.
Version 1 - vague:
Write a performance review for an employee.Version 2 - better:
Write a mid-year performance review for a customer service
representative. She has been with us for 18 months. Her
strengths are responsiveness and handling difficult customers
well. Her main development area is documentation — she resolves
issues well but does not always log them properly in our CRM.
Keep it balanced, specific, and professional.Version 3 - best:
You are an experienced people manager writing a mid-year
performance review. The employee is a customer service
representative who has been with the company for 18 months.
Strengths to highlight:
- Consistently fast response times, often below our 2-hour target
- Exceptionally skilled at de-escalating frustrated customers
- Positive feedback from three separate customers in Q1
Development area:
- CRM documentation is inconsistent — she resolves issues but
does not always log them, which creates problems for the team
when following up
Tone: warm and constructive. This is someone we want to retain
and develop, not someone being managed out.
Format: 3 short paragraphs — strengths, development area,
overall summary. Under 250 words total.Version 1 will produce something generic enough to apply to any employee in any company. Version 3 will produce something you can actually use with light editing. The difference in time to write the prompt is about two minutes. The difference in the output is the difference between starting over and being almost done.
Five habits that will improve every prompt you write
Give examples when you have them
If you have a previous version of something that worked well: an email in your voice, a summary in your preferred format, a report structured the way your leadership team likes it, then paste it in and say “match this style.” Showing beats telling almost every time. AI is very good at picking up on patterns in examples and replicating them.
Say what you do not want
Negative constraints are just as useful as positive ones. “Do not use bullet points,” “do not open with a compliment,” “do not include legal language,” “do not make it sound like a press release” - these are all specific instructions that steer the output away from things AI will do by default if you do not tell it otherwise. Think about the things that annoy you in AI-generated text and name them explicitly.
Ask for a specific format
AI will choose a format if you do not specify one. Sometimes the default is fine. Often it is not. If you need bullet points, ask for bullet points. If you need prose, ask for prose. If you need exactly three sections with specific headings, say so. The more explicit you are about structure, the less editing you will do afterward.
Iterate rather than restart
The first output is rarely the final output, and that is fine. What most people do not realize is that you can keep refining within the same conversation. “That is good but the second paragraph is too formal, then make it warmer.” “Cut the last section, it is not relevant.” “Give me a version that is half as long.” Each refinement builds on what came before rather than starting from scratch. Treat the first output as a draft, not a finished product, and the conversation as your editing session.
Save the prompts that work
When you write a prompt that produces something genuinely useful, save it. A text file, a note, a folder in Drive, wherever is easiest to find later. Over time you will build a small library of prompts that work for your specific tasks and your specific voice. That library becomes a real asset. I wrote an entire post about how to build and organize one, and it is on the roadmap for later this year, but the habit of saving what works is worth starting today, before the library exists.
The one mistake that undermines everything else
Accepting the first output without reading it critically. AI produces confident-sounding text. Confident-sounding is not the same as correct, accurate, or appropriate for your specific situation. Every output needs a human read before it goes anywhere, not a skim, an actual read. Check the facts. Check the tone. Check whether it actually says what you needed it to say, or just a plausible-sounding version of it.
I have said this in almost every post in this series because it matters that much. The skills in this post will make your prompts better. They will not make AI infallible. You are the editor, the fact-checker, and the person whose name goes on the output. Act like it.
A great prompt is just a clear brief. The same skills that make you good at delegating to a person make you good at delegating to AI. If you can explain a task clearly enough that a capable new employee could do it without asking questions, you can write a prompt that works.
A practical exercise to try this week
Take a task you have already used AI for and got a result you were only halfway happy with. Go back and rewrite the prompt using the four elements from this post: job, audience, what good looks like, and the role you want AI to play. Run it again and compare the two outputs side by side.
For most people, the difference is significant enough that the exercise is worth doing once just to feel it. After that, the habits tend to stick on their own because the feedback is immediate and the payoff is obvious.
What is coming in March
Next month we are moving into phishing awareness, specifically, how to build a program that actually changes employee behavior rather than just checking a compliance box. If you manage a team of any size, this one is worth bookmarking. It sits right at the intersection of AI and security, which is where I spend most of my time professionally, and I have a lot of practical experience to draw from.
This is post six of a two-year series on AI for real people doing real work. Post one covers what AI actually is. Post two is a look at how I use these tools day to day. Post three covers the five free tools worth trying first. Post four is about taming email with AI. Post five covers what AI means for your security posture right now. A prompt that stumped you, or something specific you want me to cover? Send a note.
