MarkdownPaper Open reader

Guide

Markdown for AI prompts

A short question needs no formatting. A long prompt reads better to a model, and to you, when it is split up with Markdown: a heading for each part, a list for the rules, and a code block around anything the model should treat as material rather than instructions.

You write

## Task

Summarize the customer emails below for the support team.

## Rules

- One line per email
- Say if the customer is angry
- No more than 10 lines in total

## Emails

```text
Hi, my order arrived broken and nobody answers the phone...
```

You see

Task

Summarize the customer emails below for the support team.

Rules

  • One line per email
  • Say if the customer is angry
  • No more than 10 lines in total

Emails

Hi, my order arrived broken and nobody answers the phone...
On this page

Why it helps

  • The parts are clear. The model can tell the task from the rules and the rules from the material.
  • Lists are easy to follow. A rule on its own line is less likely to be missed than one in the middle of a paragraph.
  • Fences keep material separate. Text inside a code block reads as something to work on rather than more instructions. Use them for documents, data, logs and code you paste in. It helps, but it is not a lock: a model can still be swayed by instructions hidden in pasted text.
  • You can reuse it. A prompt with headings is easy to edit and keep in a file.

Markdown is not magic. A clear prompt in plain sentences beats a messy one with headings. It is the clarity that helps; Markdown is a tidy way to get it.

The parts worth using

UseFor
## headingsTask, context, rules, examples, the material
- listsRules, requirements, things to check
1. listsSteps that happen in order
Code fencesPasted documents, data, code, logs
**bold**The one rule that must not be missed, used rarely
TablesExamples of input and the output you want

Show the output you want

An example of the result is often clearer than a description of it.

You write

## Output format

Reply in this shape:

```markdown
### Summary
One paragraph.

### Risks
- One risk per line
```

You see

Output format

Reply in this shape:

### Summary
One paragraph.

### Risks
- One risk per line

Ask for Markdown back

Most chat tools already answer in Markdown and show it formatted. When you need the raw text, for a file or another tool, say so:

  • "Answer in Markdown, inside one code block." The chat shows the source, and the code block's copy button copies it exactly.
  • "Use only ## headings and - lists." Limits the answer to what your destination supports.
  • "No tables." Useful when the answer goes somewhere tables do not work, such as a plain email.

Keep long prompts in files

A prompt you use often is worth saving as a .md file. Coding agents go one step further: they read instruction files such as AGENTS.md at the start of every task, which is the same idea made permanent.

Questions

Does Markdown make AI prompts better?

It makes long prompts clearer, which usually makes answers better. For a short question it makes no difference.

Should I use XML tags or Markdown in prompts?

Both work for separating parts. Markdown headings and lists are easier to read and write; tags are useful when you need to name blocks of material precisely. Many prompts use both.

How do I get ChatGPT or Claude to give me raw Markdown?

Ask for the answer "in Markdown, inside one code block". The block shows the source, and its copy button copies it as written.

Next