Why does every answer from your AI assistant arrive dotted with asterisks, hash signs and little dashes? And why, when you paste that answer into a WhatsApp chat or a customer email, does it suddenly look broken? The answer is one word: Markdown. This guide explains what it is, why every major AI tool speaks it, and how to make it work for your business instead of against it.
The timing matters. On 5 October 2026, Google began rolling out native Markdown editing in Google Docs and formatted Markdown previews in Google Drive, available to all Workspace customers and personal Google accounts. When the world's most widely used office suite adds a file format specifically because AI tools produce it, it is worth ten minutes of a business owner's time.
What is Markdown?
Markdown is a simple way of formatting plain text using everyday keyboard symbols. Two asterisks around a word make it bold, a hash sign at the start of a line makes a heading, and a dash makes a bullet point. The text stays readable as it is, and software can turn it into neat formatting.
It was created in 2004 by the American writer John Gruber, with input from the programmer Aaron Swartz. Their goal was modest: let people write for the web without typing messy HTML code.
Think of Markdown as shorthand notes for formatting. A secretary taking dictation does not draw a fancy title; she underlines it and moves on, and everyone in the office knows what the underline means. Markdown works the same way. The symbols are instructions, not decoration.
Files written this way usually end in .md, which is why you may have seen names such as README.md, SKILL.md or AGENTS.md appear in AI tools and software folders.
Why do AI tools like ChatGPT, Gemini and Claude write in Markdown?
AI chatbots write in Markdown because it gives structure with very few extra characters. A model can mark a heading with one symbol instead of a block of code, which keeps answers cheap to produce, quick to display and easy for the chat window to turn into bold text, lists and tables.
Compare the cost of one heading. In Markdown it is a single # followed by a space. In web code (HTML) the same heading needs an opening and a closing tag, about nine extra characters. Across a long answer with a dozen headings, bullets and bold phrases, that difference adds up, and AI services are priced by the volume of text they process.
There is a second reason. Large language models learned from huge amounts of public text, and a great deal of technical writing online, from software documentation to discussion forums, is written in Markdown. The models picked up the habit the way a child picks up an accent.
Google said as much in its October announcement, describing Markdown as "a standard text format widely used by Large Language Models (LLMs) to preserve structured content, like tables, headers, and bulleted lists, without the bulk of heavy file formats."
Why does AI text show asterisks and hash signs when you paste it?
Stray symbols appear because the place you paste into does not understand Markdown. The chat window converts the symbols into formatting on screen, but copying often takes the raw symbols with it. WhatsApp, SMS, many booking systems and plain email fields then display them literally.
WhatsApp is the classic Hong Kong example, because it has its own, different formatting rules:
--- Bold: WhatsApp uses one asterisk on each side (*text*), while Markdown uses two (**text**)
--- Italic: WhatsApp uses underscores (_text_)
--- Strikethrough: WhatsApp uses a tilde on each side (~text~)
--- Headings: WhatsApp has no headings, so a Markdown "#" simply shows as a hash sign
The result is a promotional message that reads "## Weekend Offer" with doubled asterisks around the price. Customers notice. It looks like a copy-and-paste job, because it is one.
Tables are the worst case. A neat price table in the chat window becomes rows of vertical bars and dashes once it lands somewhere that cannot draw tables.
Which Markdown symbols are worth knowing?
You only need about eight symbols to read and edit almost any AI answer. Learning them takes five minutes, and it lets you fix a draft directly instead of asking the AI to start again.
--- # Heading: one hash for a main title, two (##) for a section, three (###) for a sub-section
--- **bold**: two asterisks on each side
--- *italic*: one asterisk or one underscore on each side
--- - item: a dash, asterisk or plus sign at the start of a line makes a bullet
--- 1. item: a number and a full stop make a numbered list
--- [text](link): words in square brackets followed by a web address in round brackets make a clickable link
--- > quote: a greater-than sign at the start of a line makes a quotation block
--- | table |: vertical bars separate the columns of a table
Two more patterns appear often in AI answers. Three backticks (```) on their own line open and close a block of fixed-width text, usually for code or exact wording to copy. A dash followed by square brackets (- [ ]) creates a checklist item, which some apps turn into a tick box.
A useful test: copy any AI answer into a plain text editor such as Notepad. Every symbol you see there is Markdown doing its job behind the scenes.
How can a small business use Markdown with AI every day?
Small businesses get the most from Markdown by controlling it rather than fighting it. Tell the AI which format the destination needs, keep master documents in Markdown where structure matters, and reuse the same file across tools. Three common Hong Kong situations show how.
Scenario 1: a Mong Kok phone accessories shop sending WhatsApp promotions. The owner asks AI for a weekend offer message. Instead of pasting the answer and deleting symbols by hand, she adds one line to her request: "Format for WhatsApp: single asterisks for bold, no headings, no tables." The message arrives ready to send. A small instruction removes five minutes of tidying from every broadcast.
Scenario 2: a ten-person trading company writing a staff handbook. The manager asks AI to draft the leave policy and expense rules in Markdown, saves the file as handbook.md and, with the new Google Docs support, opens it in Docs for colleagues to comment on. The same file can later be given to an AI assistant as background knowledge, because Markdown is the format these tools read most easily.
Scenario 3: a tutoring centre building reusable AI instructions. Tools such as Gemini skills store reusable instructions in a Markdown file called SKILL.md (we explain these in our guide to Gemini skills). The centre writes one file describing its tone, fee table and enrolment rules, then reuses it every time staff draft a parent message.
Getting started with the new Google support takes no setup. Google says there is no admin control or user setting to change: once the rollout reaches your account, which can take up to 15 days, a .md file in Drive shows a formatted preview, and opening it in Docs lets your team edit and comment without converting the file.
The common thread: Markdown is the language your business can use to speak to AI in a structured way, not just the language AI uses to speak to you.
What do people get wrong about Markdown?
The biggest misconception is that Markdown is a programming language for technical staff. It is not. It is a handful of typing conventions that any office worker can learn in an afternoon, and it carries no risk of breaking a system.
Misconception 1: "The symbols mean the AI made a mistake." The AI did exactly what it was built to do. The mismatch is between the format and the destination, and one instruction fixes it.
Misconception 2: "Markdown is a file type I need special software for." A .md file is plain text. Notepad on Windows and TextEdit on a Mac can open it, and from October 2026 Google Docs can edit it directly.
Misconception 3: "Every app reads Markdown the same way." Versions differ. A standard called CommonMark was published in 2014 to settle the details, and GitHub's popular variant, GitHub Flavored Markdown, adds extras such as tables. Basic bold, lists and headings work almost everywhere; tables and checklists do not.
Misconception 4: "Converting to Word is always better." Word files are heavier and harder for AI tools to read cleanly. For documents that both people and AI will use, such as policies, product notes and FAQ sheets, keeping a Markdown master copy is often the smarter choice.
How do you stop AI from adding Markdown symbols?
The simplest way is to state the destination in your request. Phrases such as "plain text only, no Markdown", "format for WhatsApp" or "write this as an email body without headings" reliably change the output in mainstream AI tools. Saving that instruction as a standing preference saves repeating it.
Most AI assistants let you store standing instructions, often called custom instructions, personal context or a saved skill. One line such as "When I ask for customer messages, never use Markdown symbols" applies to every future chat.
If you already have an answer full of symbols, ask the AI to "rewrite the same text as plain text". It takes seconds and avoids manual editing errors, such as deleting an asterisk that was part of a product code.
Is Markdown worth learning for a business owner?
Yes, at a basic level. Recognising eight symbols helps you clean AI drafts faster, brief AI tools more precisely and keep company knowledge in a format that both staff and future AI systems can read. It costs nothing, needs no software and takes less time than a lunch break.
The bigger picture is that AI tools are converging on Markdown as their shared working format. Google adding native support in Docs, AI skills stored as SKILL.md files and project rules stored as AGENTS.md files all point the same way. Owners who can read these files will find it easier to check what their AI tools are actually being told.
The bottom line
Markdown is the quiet formatting language behind almost every AI answer you read. The asterisks are not errors; they are instructions that only some apps understand. Learn the eight basic symbols, tell your AI where the text is going, and keep important documents in a format both people and machines can use.
Small habits like these are what separate businesses that tidy up after AI from businesses that make AI work cleanly from the first draft. We understand AI. UD stands with you.
Reviewed by the UD AI team.
Knowing how AI formats its answers is one small part of using it well. Take the free AI IQ Test to see where your team's AI skills stand today, and we will walk you through it step by step, from reading AI output with confidence to briefing AI tools so the first draft is ready to use.