How to write AI prompts that actually work
A practical way to write prompts for ChatGPT, Claude and Gemini: give context, show an example, say what good looks like, and keep the prompts that work.
Short answer: a good prompt reads like a brief to a smart colleague who knows nothing about your project. Say who it is for, what you want, what good looks like, and show one example. Then keep the prompts that work, because the second time is where the real time saving is.
Why do most prompts fail?
Because they are too short. "Write a landing page for my app" leaves the model to guess the audience, the tone, the length and the goal. It guesses the average, and the average is bland.
The fix is not magic words. It is the same context you would give a person.
What should a good prompt include?
Use these five parts. Not every prompt needs all of them, but most bad prompts are missing at least two.
- The goal. What the output is for. "A landing page that gets founders to join a waitlist" beats "a landing page".
- The audience. Who reads it, and what they already know.
- The material. Paste the facts: product details, your notes, the data. Models invent less when they have something real to work from.
- What good looks like. Length, tone, format, and anything to avoid.
- An example. One sample of the style you want is worth a paragraph of description.
Here is the difference in practice:
Write a landing page for my app.
Write the hero section of a landing page for Lumen Voice, an AI receptionist
for dental clinics in the US. Readers are clinic owners, not technical.
Goal: get them to book a 15 minute demo.
Facts: answers calls 24/7, books into their calendar, costs less than a
part time receptionist.
Tone: plain, confident, no hype words like "revolutionary".
Format: a headline under 10 words, one supporting sentence, one button label.
Give me 3 options.
The second one takes a minute longer to write and saves twenty minutes of back and forth.
Do you need different prompts for ChatGPT, Claude and Gemini?
Mostly no. Clear context works everywhere. Differences show up at the edges: how long a document a model handles well, how it follows formatting, how cautious it is. If a prompt matters, run it on two models and keep the better result.
How do you write prompts for images, video and music?
The same principle, different vocabulary. Describe the subject, then the style, then the technical details.
- Images: subject, setting, lighting, lens or medium, mood. Keep a reference image for consistent characters.
- Video: the shot first, then camera movement, then what changes over time. Short, single action shots hold together better than long scenes.
- Music: genre, tempo, instruments, vocal style and mood, then the lyrics or structure.
Tools change fast, so the best prompts are the ones people are posting right now with the result attached.
How do you keep and share prompts that work?
Save every prompt that gave you a result you used. A note app is enough. When a prompt becomes a process you repeat, turn it into a Claude skill so you never paste it again.
Sharing is where you learn fastest. On AfterPrompt, every post can include the exact prompt behind the result, with a copy button, so anyone can try it. Browse lessons on prompting to see what is working for other people.
Frequently asked questions
How long should an AI prompt be?
Long enough to include the goal, the audience, the facts and what good looks like. For real work that is usually a short paragraph, not one line.
Do prompt templates work?
They help you remember the parts of a good brief. The value comes from filling them with your real context, not from the template itself.
What is the most common prompting mistake?
Not giving the model your actual material. Paste the facts, notes or data, and the output becomes specific instead of generic.
Where can I find good AI prompts?
Look for prompts shared next to the result they produced, so you can judge them. On AfterPrompt, posts show the prompt alongside the work.