26 Sep 2026 · 8 min readPrompting

How to Use AI Prompts That Actually Work

The difference between a weak prompt and a strong one is almost never model choice — it's structure. A good prompt tells the model what it is, what it's given, what a great answer looks like, and what it must not do. This guide gives you that frame, shows why an example outperforms an instruction, and explains why iterating beats rewriting from scratch.

The four-part frame

Every good prompt answers four questions. If yours answers all four, you rarely need to change it.

frame
# 1. ROLE        - who should the model be?
# 2. INPUT       - what material does it get?
# 3. GOAL        - what does a great result look like?
# 4. CONSTRAINTS - tone, length, format, and what to avoid
filled in
You are a senior technical editor. Below is a rough draft of a blog post.

Goal: a 600-word post a smart non-expert would finish, with one concrete example per section.
Constraints: plain English, no jargon, no hype words, end with a single question.
💡 Pro Tip: You don't need all four every time. A summarisation task rarely needs a role. A writing task almost always does.

Show, don't just tell

Models imitate patterns far more reliably than they obey abstract instructions. One concrete example is usually worth a paragraph of description.

  • Weak: 'Write headlines that are punchy.' Strong: 'Write headlines under 60 characters, no clickbait, like these: "The 4-minute fix for a slow morning"'.
  • Weak: 'Be concise.' Strong: 'Answer in exactly three sentences, then stop.'
  • Weak: 'Use a friendly tone.' Strong: paste one paragraph that already sounds like you and say 'match this'.

This is why a saved, reusable prompt beats a freshly typed one every time. Store the version that worked — that's exactly what the SlashAI command vault is for.

Iterate, don't restart

The most common waste in AI use is throwing away a mediocre answer and typing the whole request again. You lose the good parts.

  • Keep it and cut: 'Keep paragraph 1 and 4, delete 2 and 3, tighten the ending.'
  • Keep it and expand: 'Expand section 3 with a real example and a number.'
  • Change the angle: 'Rewrite this for a 12-year-old, same length.'
  • Change the format: 'Turn this into a 6-slide outline with one bullet per slide.'
⚠️ Note: Restarting from scratch every time is why 'AI gives generic output' is such a common complaint. It usually gave a good output, and you asked for a new one.

Commands for each step

Use these to generate the prompt, refine the answer, and check it before you ship it.

Explain Clearly

Explain any concept at the depth and reading level you ask for.

Use it in: ChatGPT / Gemini / Claude · copy → paste → replace bracketed placeholders with your details

Simplify Text

Translate into simple, learner-friendly language a text passage — e.g. a 300-word blog excerpt about remote work burnout.

Use it in: ChatGPT / Gemini / Claude · copy → paste → replace bracketed placeholders with your details

Rewrite Essay

Rewrite for clarity and rhythm an essay — e.g. college-essay-draft.docx — 650 words on overcoming a setback.

Use it in: ChatGPT / Gemini / Claude · copy → paste → replace bracketed placeholders with your details

Fact Check Topic

Verify every claim and rate confidence for a topic — e.g. "the economics of vertical farming".

Use it in: ChatGPT / Gemini / Claude · copy → paste → replace bracketed placeholders with your details

Compare Options

Build a weighted comparison table and give a clear recommendation.

Use it in: ChatGPT / Gemini / Claude · copy → paste → replace bracketed placeholders with your details

Five prompt mistakes

  • Asking a question the model has no data on (today's live prices, your internal metrics). It will guess — confidently.
  • Dumping everything at once. Long undifferentiated context buries the instruction. Put the instruction last, where attention is strongest.
  • Never correcting. When it's wrong, say what's wrong and what you expected. The next answer will be measurably better.
  • Asking for certainty. Ask for confidence and sources instead: 'how confident are you, and what would I verify?'
  • Using one prompt for ten different jobs. Separate the extraction from the analysis from the writing.

Explore the Full SlashAI Library

Every prompt in our guides is part of our offline-ready vault of verified commands and instant browser tools. Free forever, no account required.

Browse All Commands