Two ways to copy.
Two different results.

What actually lands on your clipboard

When a chatbot responds, it sends Markdown text to your browser. The browser then renders that Markdown into formatted HTML — headings become large bold text, **bold** becomes bold, bullet points become actual list items. What you see in the chat window is the rendered version. The raw Markdown behind it is invisible.

These two versions are stored separately, and the two copy methods reach into different places.

When you select text with your mouse and press Ctrl+C (or Cmd+C), the browser copies from the rendered DOM — the visual layer. You get what you see: formatted words, but no Markdown syntax. The asterisks and hash signs are gone, because the rendered HTML never had them.

When you click the copy button built into the chatbot interface, a small JavaScript function runs. That function retrieves the raw Markdown source — the text as it was before rendering — and writes that directly to your clipboard. The copy button is explicitly designed to give you the Markdown, not the rendered output.

Here is the same chatbot response shown from both sides:

### Three tips for better sleep

Getting consistent rest comes down to
a few key habits:

- Go to bed at the **same time** every night
- Keep your bedroom **cool and dark**
- Avoid screens for **at least 30 minutes**
  before sleep

These are small changes, but the
cumulative effect is significant.

If you selected the rendered text and copied it, the result would look something like: "Three tips for better sleep — Getting consistent rest comes down to a few key habits: Go to bed at the same time every night…" — all the structure collapsed into a single run of text, or at best preserved as plain newlines, depending on your browser and where you paste.

If you used the copy button, you get everything shown in the right panel above: the ###, the ** around each phrase, the dash-prefixed list items.

Where you paste determines which version you want

Neither version is universally better. The right one depends entirely on where the text is going next.

The awkward case is Word. It appears in both columns with a cross, because Word doesn't speak Markdown natively. The copy button gives you symbols that Word treats as literal characters. Selecting text gives you flat prose with no headings or bullets. Neither is what you actually want, which is properly formatted Word content with real heading styles and list items.

What selection copy actually does to formatting

The behaviour of Ctrl+C on selected text is more nuanced than "strips all formatting." The browser copies both a plain-text and an HTML version to the clipboard simultaneously. The application you paste into decides which one to use.

Rich text editors — Word, Google Docs, Gmail's composer, Outlook — receive the HTML version and try to honour it. A rendered chatbot heading (<h3>) may paste as bold large text. A bullet list may paste as an actual list. So selection copy into Word sometimes preserves visual structure reasonably well — but you end up with inline bold and oversized body text, not proper Word heading styles. The document looks formatted, but the structure is cosmetic rather than semantic. A table of contents generated from those headings will be empty, because Word's heading styles were never applied.

Plain text destinations — Notepad, VS Code, terminal — receive only the plain-text version, stripping all formatting. This is actually the cleaner outcome in those contexts.

What the copy button is actually doing

The copy button is a UI element added by the chatbot's developers. Its sole purpose is to make raw Markdown accessible, because selecting rendered text in a browser is imprecise — you might miss the first word, or accidentally include the chatbot's avatar, or grab only part of a code block.

Under the hood, the button calls the browser's navigator.clipboard.writeText() API with the raw Markdown string. The chatbot's internal state stores the original source text alongside the rendered HTML, so the button can always retrieve the unrendered version. This is true across ChatGPT, Claude, Gemini, Copilot, and most other major chatbots — the copy button is a deliberate design choice to expose the Markdown source, not a quirk.

One practical consequence: if you copy a code block, always use the copy button within or below the code block itself. Selecting code visually and pressing Ctrl+C sometimes captures surrounding line numbers or syntax highlighting spans depending on the interface. The copy button copies the raw code string exactly.

A simple decision rule

Word and Google Docs sit outside both columns. Both methods produce suboptimal results there. The selection method gives you cosmetic formatting without real document structure. The copy button gives you symbol-cluttered text. The right approach for those destinations is to convert the Markdown to a native document format rather than copy-pasting it.

The Word problem — and a proper solution

The copy-into-Word workflow is broken by design, and it matters because a large proportion of AI-generated content ends up in Word documents. Reports, summaries, emails drafted in Word, meeting notes — the use cases are too common to ignore.

What people typically do: use the copy button, paste into Word, then manually delete every ##, remove every pair of ** asterisks, and manually apply heading styles. This can take five to fifteen minutes for a long response, and the results are still inconsistent.

The structural alternative is to convert Markdown to a Word document directly — reading the Markdown source and generating a proper .docx file with real Word heading styles (Heading 1, Heading 2), real list objects, real bold runs, and real code blocks formatted as styled paragraphs. The document is correct by construction, not cosmetically patched by hand.

Quick reference