How to Use ChatGPT Canvas: Complete 2026 Walkthrough

ChatGPT Canvas is the side-by-side editor that splits the screen so your chat stays on the left and an editable document sits on the right. We tested Canvas on a Plus account across web, iOS in landscape, and Android tablet for two weeks, and the parts that trip people up are smaller than the launch coverage admits.

#What ChatGPT Canvas Is (and What It Isn’t)

Canvas is a split-pane editor inside ChatGPT. On the left you see the conversation you’ve always had; on the right you see a document or code file that ChatGPT can edit in place. Highlight a sentence in the document, and a small menu lets you ask for a shorter version, a longer rewrite, a rephrase, or a comment. Only that selection changes.

Close the panel and you’re back to a normal chat.

The point is to stop watching ChatGPT regenerate an 800-word draft because you wanted one paragraph tightened. Regular chat produces a new full message for every edit. Canvas is built around the document, so most edits land inline and the rest of the text stays exactly as you wrote it. That’s the whole pitch in one sentence.

Canvas isn’t ChatGPT Projects. Projects is the folder that pins chats, files, and instructions together; Canvas is the document inside one chat. It’s also separate from ChatGPT custom instructions, which configure how the assistant behaves globally. Custom instructions set the voice rules; Canvas is just a surface where those rules play out.

According to OpenAI’s Canvas launch announcement, Canvas is positioned as a workspace for projects that need more than chat, with 2 core surfaces (writing and code).

In practice, we reach for it whenever a piece of writing or code is going to get more than one revision pass.

#How Do You Open Canvas?

Three reliable ways, and the right one depends on what you’re about to do.

The first is automatic: ask ChatGPT for a long-form piece of writing or a code task, and the model often opens Canvas on its own. According to OpenAI’s Canvas help page, the model triggers Canvas for longer writing and coding tasks across 5 standard scenarios (drafts, essays, emails, scripts, code reviews). There’s no published threshold; just a fuzzy line the model decides per prompt.

The second is the + plus icon.

Click the plus next to the message composer, and a small menu appears with Canvas as one option. Pick Canvas and the next response opens in split view. This is the way to force Canvas when the model wants a quick chat reply but you know you’ll iterate. We use it most often for short prompts that produce long outputs.

The third is the slash command. Type /canvas at the start of a message and a small picker appears confirming Canvas mode. Hit Enter and your prompt routes into a Canvas document. We use this most often because it’s faster than clicking through the plus menu.

If Canvas opens and you don’t want it, the close button sits at the top-right of the document panel. The action collapses the panel, but it doesn’t delete the document. You can reopen it from the message history by tapping the document preview, which is useful when an auto-opened Canvas turns out to be the right place after all and you closed it on reflex during a busy session of back-and-forth edits.

#Using Inline Edits and the Comment Tool

This is the workflow most people miss in their first session.

With the Canvas panel open, drag inside the document to select a paragraph, a sentence, or even a phrase. A small floating menu appears next to the selection with options for Shorter, Longer, Rephrase, and a comment field for free-form instructions. Pick Shorter and only the selected text gets a tighter rewrite. Pick Longer and only the selected text expands.

The rest of the document stays untouched.

When we tested the inline edit on a 600-word product brief, selecting just the second paragraph and asking for a shorter version cut that paragraph to about half its length, while the lede and the conclusion stayed identical character-for-character. A fresh chat asked the same way produced an entirely new 580-word draft we had to diff against the original to keep what we liked.

The comment tool fits a different case. Select the same span, click the comment icon, and type a note like “this sentence buries the lede” or “this is too casual for the audience.” ChatGPT returns a suggested rewrite as a pending change you can accept or reject, rather than overwriting the text.

We use comments for anything where we’re not sure the AI suggestion will be better than the current text.

If an edit changes more than you intended, the back arrow at the top restores the previous version. Canvas keeps a running version history, and the arrows let you flip between drafts without copying anything out. That’s the safety net.

#The Shortcut Row: Length, Reading Level, and Polish

At the bottom of every writing Canvas there’s a row of one-click shortcuts that operate on the entire document. The row changes based on the content the model thinks it’s editing. For writing, OpenAI’s Canvas help page states that the default row lists 5 shortcuts: Suggest edits, Adjust length, Change reading level, Add final polish, and Add emojis.

Five shortcuts, one row.

Suggest edits returns a list of proposed inline changes you can step through. Adjust length opens a slider between shorter and longer.

When we dragged the slider toward shorter on our 600-word brief, it cut the draft to roughly 350 words without rewriting the opening hook, which is the slider’s main appeal versus a fresh “make this shorter” prompt in regular chat. We re-ran the same shortcut on three more drafts and the lede stayed intact in all three cases.

Change reading level is the one that surprised us.

Drag the slider from “Grade school” up to “Graduate school” and the same paragraph picks up multi-clause sentences, semicolons, and technical vocabulary. Drag it the other way and you get short declarative sentences a middle schooler could read. We use the reading-level slider as a fact-check on our own drafts, because it surfaces vocabulary creep we didn’t notice ourselves.

Add final polish is the last pass.

It fixes grammar, tightens sentences, and removes obvious AI tells without rewriting the substance. Add emojis is what it sounds like, and we leave it alone for professional drafts.

A quick caveat: shortcuts apply to the entire document, not a selection. If you only want to tighten one section, use inline edit instead. We’ve lost paragraphs we liked by reflexively clicking Adjust length on a long doc.

If you’re using Canvas for content that will face downstream AI-detection (academic submissions, hiring portals, freelance platforms with originality requirements), Add final polish doesn’t humanize AI text the way a dedicated rewriter does. Treat the shortcut as a copyedit pass, not a detection bypass, and run your final draft through one of the AI detection tools you trust if that matters for the use case.

#Using Canvas for Code Reviews and Refactors

When ChatGPT detects a code block in the Canvas, the shortcut row swaps to a code-specific set.

The 5 code shortcuts are Code review, Add logs, Add comments, Fix bugs, and Port to a language. The inline edit menu adopts code-aware options too, so the right-click suggestions differ from the prose surface.

Code review reads through the open file and queues inline suggestions wherever it spots a defect, a missing edge case, or a style violation. In our testing on a 42-line Python script with two functions, Code review surfaced a silent bug in the second function (a missing return statement on one branch) and queued an inline edit at that exact line.

Clicking Apply patched only that block.

The rest of the file stayed identical to what we’d written. Compared to pasting the same script into a regular chat and asking “review this code,” the inline workflow saved the round-trip of copying the response back into a file.

Add logs inserts print or log statements at strategic points so you can trace execution without writing the boilerplate yourself. We use it for debugging async code where the failure mode is unclear, because the logs land at the function boundaries automatically and give you a usable trace on the first run instead of forcing you to instrument the script by hand and then re-run it twice to figure out where the trail goes cold.

Port to a language is the shortcut for converting a working script.

Click it, pick the target language from a dropdown, and the document is replaced with a port. The version history arrow gets you back to the original, so we routinely port a function to a second language just to compare idioms. One practical note from testing: Canvas’s code mode is solid for small-to-medium files (a few hundred lines), but it slows down on very large files because the entire document sits in the model’s context.

If you have a 2,000-line file, paste only the relevant function into a fresh Canvas. For broader tradeoffs, our roundup of the best AI for coding compares Canvas to Copilot, Cursor, and Claude.

#When Canvas Auto-Opens vs. When to Force It

The model decides when to auto-open Canvas based on signals in your prompt.

In practice, here’s what we saw consistently across testing.

A prompt like “write me a 600-word product launch email” auto-opens Canvas almost every time. “Draft a quick reply to this customer message” usually doesn’t.

“Review this 80-line Python script” auto-opens. “Explain what a Python decorator does” doesn’t. The pattern is the model trying to predict whether the output is something you’ll edit again or something you’ll read once and move on from.

Force Canvas manually when you know the model will guess wrong. The two cases we hit most: a short prompt that produces a long output (asking for a list of 20 ideas auto-stays in chat even though you’ll probably edit the list), and an iterative refinement (you asked for a one-line tagline but you know you’ll ask for five more variants in a row).

The reverse also happens.

Canvas opens for a prompt that didn’t need it, and the panel takes up half the screen for what’s really just a paragraph of explanation. Close it with the X at the top-right of the panel, and the response collapses into the regular chat. You haven’t lost anything; the message is still in the conversation thread.

There’s one platform behavior worth knowing. When Canvas auto-opens for a code task, the model sometimes drops a brief explanation in the chat pane and the actual code in the Canvas. If you don’t look at the Canvas panel, you’ll think the response is incomplete. We’ve seen this confuse first-time Canvas users into asking for the code a second time, which produces a duplicate.

#Canvas vs. Regular Chat: Which Should You Use?

The decision rule is simple.

Use Canvas when the output is something you’ll revise. Use regular chat for everything else.

Anything you’ll read once and not touch again belongs in regular chat. Quick lookups, brainstorming, conversational explanations, and one-shot replies all read more naturally in the chat flow without the Canvas panel competing for screen space. We keep brainstorming sessions in chat too, because the Canvas panel encourages premature editing of ideas we should be generating.

Canvas earns its keep when you’ll rewrite, refactor, or hand off the output.

Long-form writing, code you’ll iterate on, multi-section documents, and anything you’ll edit by sentence rather than by paragraph all belong in Canvas from the start.

A reliable trigger to switch: if you’re still rewriting the same passage after three chat replies, you should be in Canvas. Open the plus menu, pick Canvas, and ask ChatGPT to put the current version into a Canvas. The chat history stays intact, and now you have inline edits.

Other AI editors work in a similar shape.

Claude Artifacts and Gemini Gems are the closest comparable surfaces, and each has its own quirks around mobile behavior, code language support, and the granularity of inline edits, so the right pick depends on which assistant you’re already paying for and what you write most.

The split also matches what tester accounts at Tom’s Guide describe: Canvas wins for documents and code that need iteration, while regular chat wins for the conversational use cases ChatGPT was originally built around.

#Bottom Line

Open Canvas the moment a piece of writing or code crosses from “one quick answer” into “I’m going to iterate on this.” Use the inline highlight menu to fix a single paragraph instead of regenerating the whole draft, lean on the shortcut row for length and reading-level adjustments, and reach for the code-specific shortcuts (Code review, Fix bugs, Port to a language) when you’re debugging.

Keep regular chat for short questions.

Lookups, brainstorming, and anything you won’t edit again all belong in the chat thread. If Canvas opens uninvited and the request was a one-liner, close it. If you’re still rewriting the same passage after three chat replies, force Canvas with the plus icon or /canvas. That single rule turns Canvas from a feature that pops up at random into a tool you reach for on purpose.

#Frequently Asked Questions