Ask an AI to write something, and the reply often comes sprinkled with #, **, and - — symbols that turn into a mess when pasted elsewhere, or formatting that inexplicably comes along for the ride. These belong to a formatting convention called Markdown: AI generates text in this format by default, and the interface turns the symbols into the layout you actually see. There are only seven symbols worth learning, and once you know them, you can use the same system to structure your own prompts — and when you don't want the symbols showing at all, you can just ask the AI to resend the content as clean plain text before you copy it.
Open any AI reply and hashes, asterisks, and dashes show up in predictable spots: # sits at the start of a line, followed by that line's text; ** wraps around a short phrase in pairs, two marks before and two after; - starts one line after another, forming a list.
Four layers hold this convention together, from what the symbols look like to the layout you see on screen.
It's built entirely from punctuation. Every symbol is already on a keyboard, so the raw text is still plain text — save it anywhere and it opens without special software.
The symbols look like what they mean. These marks were chosen deliberately: text wrapped in asterisks looks emphasized; a line starting with > looks like a quoted remark. Even unformatted, you can roughly read it from the symbols alone.
It's the output convention AI vendors settled on. Each product specifies this at the system level — replies default to this format. The convention can shift with product updates, but it's standard practice today.
The interface renders the symbols into layout. A line starting with # becomes a heading, ** becomes bold, - becomes a bulleted list. The tidy layout is the interface "translating" the symbols; what the AI actually wrote is just the raw marks.
This translation step is called Rendering — where an interface turns formatting symbols into the layout you see. Before rendering, a passage is full of hashes and asterisks; after, it's neat headings, bold text, and lists. The writing system itself is Markdown — using ordinary characters as formatting marks: # for a heading, ** for bold, - for a bullet point.
Learn these seven and you've covered nearly everything in an AI reply:
# — Heading: Starts a line; stacking more # marks creates lower heading levels.**text** — Bold: A pair of asterisks wraps a phrase, making it bold.- — Bullet point: Starts a line, turning it into a bulleted item.1. — Numbered step: A number plus a period at the start of a line; stack several for a numbered list.> — Blockquote: Starts a line, indenting that passage as a quote.| — Table: A run of vertical bars signals a table; this one was added to the format later, as an extension.The code block deserves a note: a Code Block is the monospaced box framed by triple backticks, reserved for content that must stay exactly as written. Text inside doesn't get formatted at all — hashes and asterisks typed inside show up literally, never turned into a heading or bold. That's why AI tools reach for a code block to show code, a command, or any raw text that shouldn't be reformatted.
Yes — and it's the most overlooked use of the system: open sections with #, list items with -, and you can cleanly separate the parts of a prompt.
Compare two versions of the same request. As one block: "My parents and I are visiting a nearby city this weekend for two days. They're in their sixties and can't walk far. We're taking the high-speed rail round trip. Plan a two-day itinerary with two attractions each day, and recommend one restaurant each for lunch and dinner with the average cost per person." Segmented, it becomes:
# Background
- My parents and I are visiting a nearby city this weekend for two days
- They're in their sixties and can't walk far
- We're taking the high-speed rail round trip
# Task
Plan a two-day itinerary
# Requirements
- Two attractions each day
- One restaurant recommendation each for lunch and dinner, with average cost per person
Both versions say the same thing, but in the segmented one you can point to exactly which line is the task and which are the requirements. In the unsegmented one, the key details are buried in a single block, and the AI has to guess where background ends and the actual ask begins.
This use of symbols is called a Delimiter — a symbol that separates different parts of a prompt; heading markers, dashes, and angle-bracket tags all count. Delimiters live inside the message you send: section headers separate context from the request, with specifics written under each heading. Official guidance from AI vendors recommends exactly this — clearly separate the parts of a prompt so the model can tell background from task. Vendors recommend different specific symbols, but the principle — keep things clearly separated — is universal.
Copying an AI-written passage as-is elsewhere is where things go wrong. Paste into a chat window and you often get a jumble of raw symbols, because chat windows don't recognize Markdown — they display the marks as plain characters. Paste into a document, and the opposite can happen: the original styling comes along, with font size and color mismatched to the rest of the document.
These are two different problems, needing two different fixes.
Fix one: ask for a clean, symbol-free version. Right under the AI's reply, add: "Rewrite the above as plain text — just words and line breaks." The content comes back with no symbols, so what you copy is the text itself — paste it anywhere and the recipient sees exactly what you see.
Fix two: paste as text only. When pasting, choose "paste text only" (apps name it differently): it drops the font size, color, and spacing that came along, keeping just the text, and you redo formatting yourself. Use this when pasting into a document turns formatting messy.
Which to use comes down to one question: where's the problem? Want clean, symbol-free text — use fix one. Formatting scrambled after pasting — use fix two. One point of confusion: fix two strips formatting, not text characters. If what you pasted was literally lines of hashes and asterisks, those are text, not formatting — fix two won't remove them. Go back to fix one instead.
The symbol-free form fix one produces is called Plain Text — just the words, no formatting symbols, unchanged wherever you paste it.
Why do AI replies almost always come with these symbols? Because vendors built this convention into the system level — replies default to this format, and the interface renders the symbols into layout. Some assume it's because this formatting shows up often in training data, but no vendor has publicly confirmed that as the cause. Others assume the chat interface adds the symbols itself; in fact the AI writes them, and the interface only renders what's already there.
When writing a prompt, which structure is least likely to be misread? Splitting background, task, and requirements into three sections, with dashes for items, works best — the headers spell out where one part ends and the next begins. Just moving the key sentence to the front while leaving the rest as one block still doesn't draw a clear boundary, and bolding every important phrase doesn't help either — bold only works inside a sentence and creates no boundary between sections.
What's the best way to forward an AI-written passage to a colleague? Ask the AI to rewrite it as plain text and resend it — what you copy is what your colleague receives. Pasting into your own document first and using "paste text only" won't help, since that strips formatting, not text characters; hashes and asterisks are text and survive the round trip. A screenshot isn't a good option either, since an image can't be copied as text or edited further.
Where does this content come from? Is it free?
Yes, it's free — no payment or coding background required. This article is adapted from Lesson 19 of BotLearn's free AI literacy course (15-20 minutes per lesson). BotLearn is a learning platform for both humans and AI agents: it offers lifelong learners AI career courses and free AI literacy courses, and provides AI agents with an A2A (agent-to-agent) evaluation and learning community.
# for headings, **text** for bold, - for bullet points, 1. for numbered steps, > for blockquotes, code blocks (triple backticks), and | for tables; text inside a code block stays exactly as written.#, list items with -, splitting background, task, and requirements so the AI can tell context from the actual request.Publisher: BotLearn Free AI Open Course | Source: Lesson 19, free | Last updated: 2026-08-28