Cheat sheets
Quick references you can scan in seconds.
Anatomy of a good prompt
Strong prompts almost always contain five parts:
- Role. Who the model should act as (“You are a financial analyst”).
- Context. The background it needs: audience, goal, constraints, source material.
- Task. The specific thing to do, stated as an instruction.
- Constraints. Length, tone, what to avoid, what not to invent.
- Output format. The exact shape you want back: a table, a list, JSON.
See the templates that put this into practice, or read the full guide: How to write a prompt that works.
Prompt Forge quick reference
- Your text. Paste anything messy; it does not need to be tidy.
- Execute in. “Any LLM” for chat tools, “Claude Code” for the coding agent.
- Template. The output shape; “General” fits most tasks.
- Model. Leave on the recommended default unless you need speed.
- Forge prompt, then Copy. Your structured prompt appears on the right.
Full walkthrough: How to use Prompt Forge.
Prompting do’s and don’ts
- Do tell the model what “good” looks like, and give an example if you have one.
- Do ask it to say when it is unsure instead of guessing.
- Do put the most important instruction first and last.
- Don’t bury three tasks in one paragraph; split them.
- Don’t assume it knows private or very recent facts. Give them, or use RAG.
Spotting AI mistakes
Before you trust an answer, check (full routine: How to fact-check AI answers):
- Names, numbers, and dates. These are the first things a model gets wrong.
- Confident citations. If it names a source, open it; models can invent plausible references.
- Anything recent. Models can be out of date; verify current events against a primary source.
- Unasked-for certainty. If it never hedges, be more skeptical, not less.
See also
- Glossary and Prompt templates.