Markdown basics
Structure beats paragraphs when precision matters. Markdown is a simple way to write that structure in plain text.
All views are my own. Examples are generalized or anonymized and do not reflect any single employer's confidential data, systems, or metrics.
Prompt Console uses it because ChatGPT, Claude, Gemini, and other chat LLMs follow structured instructions more reliably than a single paragraph.
What changes in LLM outputs when you add structure
Chat LLMs interpret text. A paragraph prompt blends goals, context, constraints, and formatting into one blob.
Structured prompts separate those pieces. That reduces missed constraints and makes output format easier to follow.
Expect fewer "reasonable but wrong" assumptions, fewer dropped requirements, and more consistent formatting.
Why Prompt Console uses Markdown
Prompt Console generates a prompt shape on purpose: Task, Context, Constraints, Output format. Markdown is the simplest way to represent that shape in plain text.
It's not about looking pretty. It's about making instructions explicit so models miss less.
The only Markdown you need
Default pattern: Use ## for prompt sections. Use # only for an optional title.
Use ## to label sections. The label is the instruction.
Use - list items for requirements. One line = one instruction.
Use triple backticks when you paste text that must stay intact.
Examples
Unstructured prompts bury constraints in paragraphs. Structured prompts force constraints into labeled slots.
Less guessing. Fewer dropped requirements.
Help me reply to this message. I can't meet this week.
I want to be polite but direct.
"Can we jump on a call tomorrow to review the project?"
## Task
Draft a short reply declining the meeting request and proposing alternatives.
## Audience
A colleague
## Tone
Direct, polite
## Context
They asked to meet tomorrow. I can't meet this week.
## Content to Process
"Can we jump on a call tomorrow to review the project?"
## Constraints
Under 120 words. Offer two times next week.
## Avoid
Don't over-explain. Don't promise availability you can't keep.
## Output Structure
Return: subject line + message body.
## Important
If any information is missing or unclear, indicate this with [MISSING] or [UNCLEAR] rather than making assumptions.
Turn these notes into action items:
- Launch moved to a later window
- Need legal review
- Someone owns comms plan
- Open question: pricing tiers
## Task
Extract the key decisions and convert notes into an action plan.
## Audience
Project team
## Tone
Clear, operational
## Content to Process
- Launch moved to a later window
- Need legal review
- Someone owns comms plan
- Open question: pricing tiers
## Constraints
Don't invent owners or dates. Use [MISSING] for unknowns.
## Output Structure
Provide:
1) Decisions
2) Action items (Owner, Due date, Action) - use [MISSING] placeholders
3) Open questions
## Important
If any information is missing or unclear, indicate this with [MISSING] or [UNCLEAR] rather than making assumptions.
Which option is better?
Option A: cheaper, fewer features
Option B: more expensive, saves time
I care most about time and reliability.
## Task
Compare the options and recommend the best choice based on my priorities.
## Audience
Me (non-expert)
## Tone
Direct, practical
## Content to Process
Option A: cheaper, fewer features
Option B: more expensive, saves time
Priority: time and reliability
## Constraints
Don't assume prices or specs not stated.
## Output Structure
1) Side-by-side comparison table
2) Recommendation
3) Key tradeoffs + what would change the decision
## Important
If any information is missing or unclear, indicate this with [MISSING] or [UNCLEAR] rather than making assumptions.
Practical tips
- Name sections with headings: Task, Context, Constraints, Output format.
- Keep constraints together, and write them as one-line list items.
- Use triple backticks when you paste text that must stay intact (quotes, logs, drafts).
- Tell the model what to do with unknowns: "Use [MISSING] instead of guessing."
Common symbols (reference)
Below are common Markdown symbols. Use them to label sections, isolate requirements, and keep verbatim text intact.
| You type | Meaning | Why/when to use it in a prompt |
|---|---|---|
# Heading |
Big heading | Optional: name the whole prompt (e.g., "# Prompt"). Use once. |
## Subheading |
Section heading | Use for the main prompt sections (Task, Context, Constraints, Output format). |
- List item |
List | Write requirements as separate lines so they don't get buried or blended. |
1. Step |
Numbered list | When order matters. Helps the model produce a sequence instead of a summary. |
**bold** |
Bold emphasis | Emphasize the one thing that must not be missed. Use sparingly. |
*italic* |
Italic emphasis | Light emphasis or labels. Optional. |
`inline code` |
Inline code / literal text | Pin exact terms (variables, filenames, commands) so the model copies them literally. |
``` (triple backticks) |
Code block / preserved formatting | Wrap source text you don't want rewritten (quotes, logs, drafts). Reduces accidental edits and preserves formatting. |
[text](https://example.com) |
Link | Provide a reference the model should use, follow, or cite (when relevant). |
> Quote |
Block quote | Separate quoted material from your instructions so the model doesn't absorb it as requirements. |
Learn more
Learn more (optional): CommonMark Markdown reference