TL;DR
Canvas is ChatGPT’s split-view editor: the document or code file lives in a panel on the right, the chat stays on the left. Highlight a paragraph or function, prompt against the selection, and the model edits just that part instead of regenerating the whole reply. It keeps a per-document version history with a red/green “Show changes” diff. As of June 2026, Canvas is on every plan (Free, Go, Plus, Pro, Team, Enterprise, Edu) on web, the Windows/macOS desktop apps, and mobile. Draft in chat first, switch to Canvas once a draft is worth keeping, and always edit from a highlighted selection.
What this covers
Editing a 1,200-word essay inside a chat window is miserable. The model rewrites the whole thing every turn, you lose track of which paragraph changed, and the part you care about scrolls out of view. Canvas fixes the shape of the work: the document is a persistent object on the right, and you can tell the model “edit only this paragraph” without it regenerating the entire piece.
This guide is for people who write or refactor in ChatGPT and keep hitting the wall where the chat-only loop is the wrong tool. It covers exactly how to open Canvas, how to scope edits so the model stops touching paragraphs you didn’t select, how the version history works, and which of the built-in shortcuts are worth using.
Who this is for
Writers polishing a draft, engineers refactoring a 200-line function, PMs iterating on a one-pager, and anyone who has copy-pasted between ChatGPT and a Google Doc a dozen times because the chat keeps losing the prior version. The rule of thumb: if your output is a single paragraph, stay in chat. If it is more than half a page and will be revised three or more times, switch to Canvas.
When to reach for it
- Iterating on a draft longer than ~400 words where you want paragraph-level edits, not full rewrites.
- Refactoring code where you need the function signature to stay stable across turns.
- Any task where you’ll want to compare versions, because Canvas keeps a version history per document.
- Building structure with the model: outline first, then expand section by section without losing the outline.
How to open Canvas (four ways)
ChatGPT opens Canvas automatically when it detects a long document or coding task, but autodetection is inconsistent, so it’s worth knowing how to force it. As of June 2026 there are four reliable methods:
- Type
/canvasin the prompt composer, or simply ask: “open this in a canvas” / “open a coding canvas.” - Click the tools icon (the box-of-objects icon) in the message bar and pick Canvas.
- Paste a block of text or code into the composer and use the open in canvas shortcut that appears in the upper-right of the composer.
- Point at an existing chat reply and ask the model to “open your last message in canvas.”
Decide upfront whether this is a writing canvas or a code canvas. Code mode exposes the developer shortcuts (review, logs, comments, fix bugs, port); writing mode exposes length and reading-level controls. Have your source material ready before you open Canvas. Once you’re in the document view, pasting large reference blocks gets awkward.
Pick the right model
As of June 2026 the ChatGPT default is GPT-5.5, with a picker exposing Instant, Thinking, and Pro modes. For Canvas:
| Task | Model / mode | Why |
|---|---|---|
| Structural rewrite, long-doc reorganize | GPT-5.5 Thinking | Holds more of the surrounding context across a big edit |
| General paragraph and code editing | GPT-5.5 Instant | Fast, preserves the selection boundary well |
| Quick grammar/tone polish | GPT-5.5 Instant | No reason to spend reasoning tokens on a wording pass |
| Deep refactor across a long file | GPT-5.5 Pro (Pro plan) | Largest in-app context headroom for stubborn refactors |
In-app context on Plus is roughly 320 pages; the full 1M-token in-app context only lands on the $200 Pro tier. If a single Canvas document is so long that edits start leaking across boundaries, the document is the problem, not the model. Split it.
Step by step
-
Draft in chat first. Get the outline and rough draft done in the chat loop, where throwing away a version costs nothing.
-
Switch to Canvas once you have a draft worth keeping:
Open this draft in a canvas so I can edit it paragraph by paragraph. -
Highlight the paragraph you want to change, then prompt against the selection. You can also click the inline comment icon to grab a whole block:
Tighten just this paragraph by about 30%. Keep the example about the 1,200-word essay. Do not touch the surrounding paragraphs. -
For code Canvas, name the function boundary explicitly:
Refactor the parseTransactions function to use a single pass instead of two. Keep the signature and return type unchanged. Do not modify the helper functions below it. -
Compare with version history. The arrows at the top of the document step through previous versions; click the Show changes icon to see deletions in red and additions in green. If a revision dropped something important, roll back to the prior version instead of asking the model to recall it.
-
Copy the final version out to your real document. Canvas has no comments thread, no track-changes review flow, and no real-time collaboration, so don’t treat it as the source of truth for anything that ships.
The built-in shortcuts, and which ones earn their place
Canvas ships preset shortcut buttons. Most are situational; a few are genuinely useful.
Writing shortcuts
| Shortcut | What it does | Verdict |
|---|---|---|
| Suggest edits | Inline comment-style suggestions you can apply one by one | Useful for a review pass on someone else’s draft |
| Adjust length | Slider from Shortest to Longest | Handy, but a scoped prompt gives finer control |
| Reading level | Slider from Kindergarten to Graduate School | Good for retargeting an explainer to a different audience |
| Add final polish | Grammar, clarity, consistency sweep | Fine as a last step, never as the main edit |
| Add emojis | Sprinkles emojis through the text | Skip it for professional copy |
Code shortcuts
| Shortcut | What it does | Verdict |
|---|---|---|
| Review code | Inline suggestions on logic and performance | The most useful code shortcut |
| Add logs | Inserts print/log statements | Handy when you’re about to debug |
| Add comments | Annotates the code | Use sparingly; over-commenting is noise |
| Fix bugs | Detects and patches problem code | Always re-run after; trust nothing blindly |
| Port to a language | Translates to JavaScript, TypeScript, Python, Java, C++, or PHP | Solid for a first-pass translation, not a final port |
A prompt that keeps Canvas edits surgical
Edit only the highlighted text. Constraints:
- Keep the surrounding paragraphs untouched.
- Preserve the example, the numbers, and the structure of the
sentence after the highlighted one.
- If a change requires touching adjacent text, stop and ask
before doing it.This is the single biggest improvement to a Canvas hit rate. It sharply reduces the “helpful” rewrites of paragraphs you never asked it to touch.
Quality check
- Open Show changes and read the diff. If you can’t tell what changed, the edit was too broad.
- Re-read the paragraphs adjacent to the edited one. Canvas sometimes leaks edits across the boundary.
- For code, run it. The inline review is not a substitute for actually executing the function.
- Confirm links, citations, and numbers survived. Canvas drops these more often than plain chat does.
How to reuse this workflow
- Keep a
canvas-edit-prompts.mdwith your top five selection-scoped prompts: tighten, expand, simplify-tone, add-example, fact-check-this-paragraph. - For recurring document types (weekly updates, release notes, retros), keep a starter Canvas you duplicate and edit instead of starting from a blank document.
- For code refactors, keep a separate Canvas per file. Cross-file refactors are where Canvas’s context starts to leak.
Common mistakes
- Starting in Canvas before you have a draft. The empty-document mode produces a worse first draft than the chat loop does.
- Not highlighting before editing. An unscoped prompt rewrites the whole document and silently drops things you wanted kept.
- Treating Canvas as a word processor. No comments thread, no track changes, no real collaboration. Copy out to Docs or Notion when you’re done.
- Editing with a quick-polish mindset on a structural change. Use GPT-5.5 Thinking for big reorganizes; Instant preserves less context on heavy edits.
- Ignoring version history. If you lost a good sentence two edits ago, roll back. Don’t ask the model to “remember it.”
- Pasting a 4,000-word document and expecting clean paragraph edits. Canvas degrades on very long documents. Split the work.
FAQ
- Is Canvas free?: Yes. As of June 2026 Canvas is available on every ChatGPT plan, including Free, on web, the Windows and macOS desktop apps, and mobile. Higher tiers mainly help by giving you stronger models and more in-app context, not by unlocking Canvas itself.
- How do I open Canvas if it won’t switch automatically?: Type
/canvasin the composer, click the tools icon and choose Canvas, or paste your text and use the “open in canvas” shortcut. If none of those work, see Canvas not working. - Can I import a file directly into Canvas?: Pasting the text body and asking the model to format it is the reliable path. Direct file-to-Canvas is inconsistent, so paste the content rather than relying on an upload to open in the editor.
- Why did Canvas rewrite a paragraph I didn’t highlight?: Usually the prompt referenced the surrounding context, so the model treated it as fair game. Re-prompt with explicit “do not touch the other paragraphs” language and edit from a selection.
- How does version history work?: Each edit is a version. Use the arrows at the top to step back, and the Show changes icon to see a red/green diff. Rolling back is lossless; you can always step forward again.
- Is Canvas safe for confidential content?: It follows the same data policy as the rest of ChatGPT, so check your plan’s data controls. Don’t paste anything into Canvas you wouldn’t paste into a normal chat.