Why not just copy and paste into your AI?

Copy-paste keeps the words but loses the shape. Here's what your AI actually receives, why structure matters, and the format that fixes it.

PASTED MARKDOWN

It's the most natural thing in the world. You find a page, a table or a long thread, select it, copy it and paste it into ChatGPT, Claude or Gemini. Most of the time it sort of works. But "sort of" is where a lot of wrong answers come from.

What copy-paste actually sends

Copying keeps the words. It loses the shape around them. Headings turn into ordinary lines, bullet points blur into one paragraph, and tables collapse into a row of values with nothing to say which number belongs to which column.

Take a simple comparison of two cards. On the page it's a neat table. Pasted into a chat box, it often arrives like this:

textCard Forex fee Annual fee Card A 0% S$0 Card B 1% S$192
CardForex feeAnnual fee
Card A0%S$0
Card B1%S$192

What you see in the browser.

Card Forex fee Annual fee Card A 0% S$0 Card B 1% S$192

What lands in the chat box. Which fee is the 1%?

| Card   | Forex fee | Annual fee |
|--------|-----------|------------|
| Card A | 0%        | S$0        |
| Card B | 1%        | S$192      |

Every value still sits under its column.

Tap through: the same table three ways.

Was 1% the forex fee or the annual fee? You can work it out because you saw the page. Your AI didn't. It has to guess, and it guesses confidently.

Markdown: plain text that keeps its shape

Markdown is a way of writing structure with ordinary characters. A # makes a heading, a dash makes a bullet, and pipes make a table. It's also the format AI tools write back to you: all that bold text and those neat lists in a ChatGPT or Claude answer are Markdown.

The same table in Markdown:

markdown| Card   | Forex fee | Annual fee |
|--------|-----------|------------|
| Card A | 0%        | S$0        |
| Card B | 1%        | S$192      |

Now there's no guessing. Every number sits under its column.

Structure is how AI understands what it reads

Good answers depend on things a human reader takes for granted:

  • which number belongs to which row
  • what's a heading and what's the detail under it
  • who said what in a long thread
  • where a piece of code starts and stops

Strip those out and the model fills the gaps itself. That's where made-up numbers and mixed-up quotes come from.

It's lighter too

When an AI tool opens a web page itself, it often reads the code behind the page as well: tags, class names, scripts. Here's that same two-row table as a typical website might build it:

html<div class="product-specs-wrapper"><table class="spec-table table--striped" data-component="specs"><thead><tr><th class="spec-head" scope="col">Card</th><th class="spec-head" scope="col">Forex fee</th><th class="spec-head" scope="col">Annual fee</th></tr></thead><tbody><tr class="spec-row"><td class="spec-cell"><span class="label">Card A</span></td><td class="spec-cell"><span class="value">0%</span></td><td class="spec-cell"><span class="value">S$0</span></td></tr><tr class="spec-row"><td class="spec-cell"><span class="label">Card B</span></td><td class="spec-cell"><span class="value">1%</span></td><td class="spec-cell"><span class="value">S$192</span></td></tr></tbody></table></div>

We counted both versions with a GPT tokenizer. The web code is 232 tokens. The Markdown table is 42 tokens, about 82% fewer, for exactly the same information.

Tokens for the same two-row table

As web code232Tags, class names and the data: 232 tokens
As Markdown42Just the data and its shape: 42 tokens, about 82% fewer
FormatTokens
As web code232
As Markdown42
Counted with a GPT tokenizer on the example above. Hover a bar for detail.

Tokens are what plan limits, message caps and API bills are counted in. On a free plan that means more questions before you hit the wall. For a team that runs a lot of work through AI, it adds up per person.

And some things you can't copy at all

A forum thread spread over ten pages. The captions of a two-hour video. The searches, scripts and sources hidden behind collapsed panels in an AI chat. None of these can be selected in one go.

The shortcut

This is why we made Page to Markdown, a free Chrome extension. One click turns any page, a whole multi-page thread, a video's captions or an AI chat into one clean .md file, with tables, headings, code and links kept and ads, menus and cookie banners left out. It runs in your browser, so nothing is uploaded anywhere.

Whatever you use, the takeaway is the same: give your AI the structure, not just the words.