☕ Prompting & prompt engineering

Get better answers from AI.

Learn prompting, prompt engineering, and the techniques that get better results from ChatGPT, Claude, Gemini, and any AI. Not a thousand templates you’ll never remember, six short lessons on why AI gives bad answers and how to fix it, each paired with a tool you can try right now.

New to the lingo? The model reads your prompt as tokens and can only hold so many in its context window. When an LLM is unsure, it may hallucinate, answer confidently but wrong. Hover any underlined word for a quick definition, or click to learn more.

  1. 1

    Why AI misunderstands you

    A model can’t read your mind. Every detail you leave out, who it’s for, what it’s for, how long, it guesses, and guesses differently each time. Most “bad AI answers” are really just under-specified prompts. Start by seeing what your prompt leaves to chance.

    Try it: Prompt Simulator
  2. 2

    Start with a goal

    Before anything else, say what success looks like. “Write a homepage” has no goal; “write a homepage that gets visitors to start a free trial” does. A goal turns vague output into output that’s trying to do something.

    Try it: Prompt Builder
  3. 3

    Name your audience

    The same request needs opposite answers for a beginner and an expert. Tell the model who it’s writing for and the depth, vocabulary, and tone fall into place on their own.

    Try it: Prompt Builder
  4. 4

    Give it context

    The model only knows what’s in the prompt. Paste the article, the code, the data, the brand voice, whatever it needs to work from. Without context it fills the gap with generic filler. X-ray your prompt to see if the context is actually there.

    Try it: Prompt X-Ray
  5. 5

    Add constraints

    Constraints are where quality lives: length, what to avoid, what to include, the rules. “In 3 bullets, no jargon” beats “keep it short.” Watch out for conflicting ones (“detailed but brief”), the linter catches those.

    Try it: Prompt Linter
  6. 6

    Specify the output format

    Say exactly what you want back: prose, bullets, a table, JSON, a specific shape. Unspecified format is the number-one cause of “it almost worked.” The before/after gallery shows the difference a format line makes.

    Try it: Good vs Bad Prompts
Capstone · build it

Write and optimize a real prompt

Reading the method isn’t the same as owning it. Pick a task you actually need done this week, an email, a piece of code, a summary, and write the prompt for it using the five moves above. Then tighten it with the tools. Check off each step; your progress is saved on this device.

0/7

Tools for the job: Prompt BuilderPrompt OptimizerPrompt Diff

That’s the whole method

Goal, audience, context, constraints, format. Answer those and you’ve written a strong prompt, no formula required. When you’re iterating, compare versions with Prompt Diff, and when you’re ready to go deeper, the full learning paths and knowledge map cover how the models work underneath.