Editing long drafts with ChatGPT

The problem: long documents in a chat stream

In an ordinary chat, a long document lives inside the conversation as plain text. Ask for a change and ChatGPT reprints the entire thing below your message, so a ten-paragraph report scrolls past every single time you adjust one line. You lose your place, you can't easily see what changed, and comparing versions means scrolling up and down. It works, but it's fiddly for anything substantial.

ChatGPT fixes this by giving a draft its own home. When you ask for something you're likely to revise (an email, a document, a chunk of code) it appears in a writing block or a code block rather than as loose text in the stream. That block is a proper editing surface: you can click into it and type directly, and it updates in place when you ask for a change, instead of a fresh copy appearing lower down. This is the idea the older Canvas side panel introduced; OpenAI's help now describes it as writing blocks and code blocks inside the chat (see Keeping current).

The block usually appears on its own when ChatGPT writes something you'd revise. You can also ask for one directly ("Draft a reply email to my manager") or paste existing text and ask to work on it as an editable draft.

Editing a document in place

Say you've drafted a two-page project update and it needs work. You've got two ways to change it, and the skill is knowing which to reach for.

Direct editing: click into the block and type, exactly like a normal document. Fix a typo, tweak a sentence, delete a line. ChatGPT sees these edits and works around them.

Targeted instructions: ask in the chat for a specific change to a named part. This is where the editable draft earns its keep: instead of "rewrite the whole thing", you can say "make this paragraph warmer" and only that paragraph changes.

In the draft above, rewrite just the second paragraph (the one about the timeline slipping) so it's more direct and doesn't sound defensive. Leave everything else exactly as it is.

The writing block also gives you a few controls: copy it, ask for revisions, open it full screen, undo or redo recent AI edits if a change made things worse, and (for an email) send it on. Exactly which controls you see varies by plan, device, model and rollout. And because your own typed changes and ChatGPT's edits share one document, you can co-write properly: you draft the hard paragraph, it polishes the rest.

The same idea for code

Editing in place helps just as much with code and structured text: a formula, a script, a chunk of config. Even if you're not a developer, you may hit this: a spreadsheet formula that won't work, a snippet someone sent you, a bit of website text with formatting. In its own block, the code stays put and you can ask for a fix to one function or line without the whole file being reprinted. Code blocks also let you edit supported languages directly, preview output, run Python and see the console result, and share a read-only link.

In the code above, the function that totals the monthly figures is returning the wrong number when a cell is blank. Fix just that function to treat blanks as zero, explain in one line what was wrong, and leave the rest of the script unchanged.

When an editable draft beats plain chat, and when it doesn't

An editable draft is worth it when the thing you're working on is long, iterative, and something you'll keep revising: a report, a policy rewrite, a proposal, a script you're debugging. The dedicated block, in-place edits and undo all pay off precisely because you're going round several times.

It's not worth it for the quick stuff. A one-line email, a fast question, a short summary you'll copy out immediately: plain chat is faster. If you find yourself with a giant document in an editable block but you're really only asking one-off questions about it, you probably wanted a normal chat, or a file upload (last-but-one lesson) instead. A rough rule: reach for the editable draft when you'll edit the thing, not merely read or receive it.

Try it now

Common mistakes

• Using an editable draft for tiny tasks. A one-line rewrite doesn't need its own document block. Save it for things you'll revise repeatedly.
• Asking for full rewrites. "Redo the whole thing" throws away the in-place advantage. Point at the specific section and change only that; it's the entire reason to work this way.
• Forgetting your own typed edits are part of the document. If you hand-edit the block and then ask for a big change, ChatGPT works from your edited version. Use undo if a change buries something you'd typed, and keep your own copy of anything important.
• Trusting the polished final draft without a read-through (over-trust). An editable draft makes a document look finished and authoritative, and editing in place can quietly drop a figure or soften a claim between versions. A neat final block is not a checked one: read the whole thing against your source before it goes out, especially any number or name ChatGPT touched.

Check your judgement

A tone edit unexpectedly removes your caveat. What would a useful revision check include?

Keeping current

This feature has changed shape more than once. It launched as Canvas, a separate panel beside the chat; OpenAI's help centre now describes writing blocks and code blocks inside the conversation instead, and the old Canvas help article is no longer published. If a colleague or an older guide talks about Canvas, this is what they mean. Which controls you get varies by plan, device, workspace settings and model. For current behaviour, see OpenAI's help article on writing blocks and code blocks and the ChatGPT release notes. Accurate as of 26 September 2026.