Prompt library

AI prompts for real work

A curated library, organized by use case. Copy, fill in the placeholders, ship. Open straight in ChatGPT, Claude, or Gemini.

100prompts·10categories·One-click open in ChatGPT, Claude, or Gemini

What makes a good prompt

Good prompts aren't magic incantations. They're just specific. The difference between a 30-second answer and a useful one is usually 30 more seconds of input.

1. Role + task + format + constraints

"Write a follow-up email" → "Write a 3-paragraph follow-up to a prospect who didn't reply, opening with the specific demo concern they raised, ending with a clear next step."

2. Show your work

Include the audience, the goal, what's been tried, what you specifically don't want. The model can't read your mind — but it's great at using context you give it.

3. Constrain the output

Length, tone, structure. "Under 80 words." "In markdown with bold for key terms." "No bullet points." Constraints kill the generic LLM voice.

4. Iterate, don't redo

"Same email, but more direct" beats writing a new prompt. The model has the context already — use it.

How to use this library

  1. 1Browse by category below or search for the job-to-be-done. Each prompt has a use case and a fill-in template.
  2. 2Use the Copy button to grab the prompt, or Open in ChatGPT/Claude/Gemini to jump to a fresh chat with the prompt pre-filled.
  3. 3Replace any {{placeholders}} with your specifics before sending. The placeholders are usually the difference between a generic answer and a useful one.
  4. 4Treat the first response as a draft. Iterate with follow-up turns: "more direct," "cut by 30%," "rewrite for a skeptical CFO." Don't start over.

Doing this work every week instead of every now and then? Install the equivalent Claude Skill instead of copy-pasting.

Browse by category

Why we built this

Most prompt libraries online are SEO spam: 500 generic templates, no structure, no opinion. We curated these from real use, refined them based on what actually produces useful output, and organized them by job-to-be-done — so you can find the prompt you need in 10 seconds, not 10 minutes.

These are starting points, not final answers. The models change, the conventions shift, and so do these prompts. Send improvements: hello@ainews.tech.