The Punctuation That Makes AI Listen

Without structure, the AI is guessing

Every prompt has parts — a task, some constraints, maybe an example or some context. Write them as a single paragraph and the AI has to figure out which is which. Sometimes it gets it right. Often it doesn’t.

That’s where markdown comes in. A heading tells the AI “this is a new section.” A bullet list says “these are separate points, treat them individually.” Bold says “this bit matters most.” These structural cues help the AI figure out what’s an instruction, what’s context, and where to focus its attention.

Markdown is very often used to add them because it is quick and simple. It uses ordinary punctuation (#, -, **) and because it’s just text, you can type it straight into any AI chat box.

Why Use It

AI models respond well to structure. A prompt with clear sections and labelled parts is easier for the model to parse than a wall of undifferentiated text — and markdown is the most accessible way to add that structure. You don’t need special tools or syntax knowledge. A # and a few - characters are enough to turn a vague paragraph into an organised brief.

Because models were trained on enormous amounts of internet text — documentation, technical articles, code repositories — and a large proportion of that text used markdown formatting, they already recognise what headings, bullet lists, and code blocks mean. You’re not teaching the model a new format. You’re using one it’s already seen millions of times.

The difference is like handing someone an organised brief versus reading instructions aloud over a noisy phone line. Same content, far less room for misinterpretation.

Five Essential Markdown Elements

1. Headings

Put # symbols at the start of a line to create headings. # is a top-level heading, ## is a section, ### is a subsection. Use them to separate distinct parts of your prompt — role, task, context, rules.

# Role
You are a marketing copywriter for a small coffee roastery.

## Task
Write a product description for a new single-origin blend.

## Rules
- Keep it under 100 words
- Use a warm, inviting tone
- Mention the tasting notes: chocolate, cherry, light smokiness

Headings remove ambiguity about where one instruction ends and the next begins.

2. Bullet lists

Start a line with - or * to create a bullet point. Use bullets for requirements, rules, or steps — anything where individual points matter and shouldn’t be buried in a paragraph.

## What to include
- Company name and founding year
- The three main services we offer
- A short quote from a satisfied client

## What to avoid
- Don't mention pricing
- Don't use jargon or technical terms
- Don't exceed 200 words

Indent to create sub-items when a point needs more detail:

- Write a welcome email for new subscribers
  - Keep the tone friendly and informal
  - Include a link to our most popular blog post
  - Mention the free guide they'll receive

I use bullet lists extensively in my own AI tooling — for flagging constraints (- IMPORTANT: Do not modify the existing schema), breaking complex instructions into steps, and separating “must do” from “must not do.”

3. Bold

Wrap text in double asterisks — **like this** — to mark it as critical. Use it for constraints the AI must not overlook. Use it sparingly; when everything is emphasised, nothing is.

## Rules
- Write in British English
- Keep it under 300 words
- **IMPORTANT: Do not include any pricing information**
- Use a professional but approachable tone

The **IMPORTANT:** pattern is worth adopting. I use it constantly when writing instructions for AI tools. The model recognises bold as emphasis and treats it accordingly.

4. Code fenced blocks

Three backticks (```) before and after a block of text create a code fence. In prompting, code fences separate examples and templates from instructions. Without them, the AI may treat your example as a command.

If you need output in a specific format, give the AI a fenced template:

Write a weekly status report following this template:

```
Project: [project name]
Week ending: [date]
Status: [on track / at risk / blocked]

## Completed this week
- [item 1]
- [item 2]

## Planned for next week
- [item 1]
- [item 2]

## Risks or blockers
- [item 1, or "None"]
```

The fence draws a clear line: everything inside is a pattern to follow, not an instruction to act on.

5. Inline code

A single backtick on each side — `like this` — marks a term as specific and literal. Use it for file names, field names, or any term the AI should treat exactly as written.

“Update the name field” is ambiguous. “Update the name field” is precise.

- Save the output as `weekly-report.csv`
- Use the `summary` field, not the `description` field
- The subject line should start with `[Action Required]`

Small difference in formatting, large difference in accuracy.

Start with headings and bullets

You don’t need all five at once. Headings and bullet lists alone will noticeably improve your results. Add bold for critical constraints, code fences when you need to show a template, and inline code when precision matters.

The formatting takes seconds. The improvement in output quality is immediate.

Things to try

Try these with any AI chat tool. The examples are taken from Five Prompting Basics That Will Transform Your AI Results and rewritten using markdown.

1. Customer service email. Paste this into any AI chat and compare the result to a plain-text version of the same prompt:

# Role
Experienced Customer Service Manager

## Task
Draft a response to a complaint about a late delivery

## Context
- Mrs Smith has been a loyal customer for five years
- Bad weather caused the delay

## Rules
- Be empathetic
- Don't offer a full refund
- Offer a 10% discount on the next order
- Under 150 words

2. Restaurant staff rota. Try the same approach with a scheduling task:

# Role
Restaurant Shift Supervisor

## Task
Create a rota for Friday to Sunday

## Context
- Three staff: Carla, Mike, Sam
- Opening hours 5pm–11pm

## Rules
- Everyone works Friday
- Mike can't work Sunday
- Minimum 4-hour shifts

## Format
A structured table

Want to know more?

Next time you’re writing a prompt, try adding some markdown. Put headings between sections, turn your constraints into bullet points, and see what changes. Then experiment — try different structures, different formatting, and notice what makes the results better or worse. The more you play with it, the quicker you’ll develop a feel for what works.