OpenAI
ChatGPT prompts
How to prompt ChatGPT: clear instructions, the right context, and standing preferences set once instead of repeated.
66 prompts
Prompts by AI model
The library is model-independent by design. These pages collect the prompts that suit each tool and explain how to adapt them.
Anthropic
Claude prompts
How to prompt Claude: explain the task as you would to a capable colleague, structure long inputs, and show examples.
66 prompts
Gemini prompts
How to prompt Gemini: specific instructions, stated constraints, a few examples, and the question after the context.
66 prompts
Perplexity AI
Perplexity prompts
How to prompt Perplexity: ask a precise research question, say which sources count, and check the citations it returns.
3 prompts
Anysphere
Cursor prompts
Cursor is an AI code editor, not a model. To prompt it well: reference the right files, set project rules, review diffs.
11 prompts
How model compatibility is labelled
Every prompt page has a compatibility table. Each model listed there carries one of three labels:
- Expected to work — the prompt uses plain structure that the vendor’s documentation supports, and it has been reviewed for that. It has not been run and recorded on that model.
- Tested — a test record is published on the page: model and version, date, inputs, the actual output and notes against the rubric.
- Model-specific version — a separate version of the prompt exists for that model because the difference improved results in testing.
Models change often. When a vendor ships a new version, tested prompts are flagged Needs retesting until they are run again. Read more in the editorial standards.