How do you write and edit with ChatGPT canvas?

Before canvas, revising a document with ChatGPT meant asking for a change and receiving the whole text again, then scrolling to find what had moved. Canvas, which OpenAI introduced on October 3 as an early beta, opens the document in its own window beside the chat. You and ChatGPT work on the same page, and you can point at exactly the part you want changed.

Canvas is rolling out to ChatGPT Plus and Team users, with Enterprise and Edu users following, and OpenAI plans to offer it to free users once the beta ends. It runs on GPT-4o. This guide walks through a real editing session from first draft to final version, so you can see which tool fits which job.

Opening canvas

There are three ways in:

1. Choose GPT-4o with canvas in the model picker at the top of a chat.
2. Include the words "use canvas" in your prompt, which also works for pasting in an existing draft.
3. Let ChatGPT decide. It opens canvas on its own when it judges a separate workspace would help, for example for longer writing or a coding task.

Once the window is open, three controls do most of the work. You can type directly in the document, like any editor. You can highlight a sentence or paragraph and ask ChatGPT to work on that part only. And the shortcut menu offers one-click edits. The back button restores an earlier version.

A worked editing session

A community garden coordinator needs a letter asking local businesses to sponsor new raised beds. She starts with a short prompt:

Use canvas. Draft a one-page letter to local businesses asking them to sponsor raised garden beds at the Maple Street Community Garden. Each bed costs $150 to build. Sponsors get a small sign on their bed. Warm, professional tone.

Step 1: Fix the facts yourself. The draft appears in canvas. It invents a founding year for the garden. Rather than asking ChatGPT, she clicks into the text, deletes the sentence and types the correct detail herself. Direct editing is the fastest fix for facts only you know.

Step 2: Adjust the length. The letter runs a little long. She opens the shortcuts, chooses Adjust the length, and drags the slider toward shorter. The whole letter tightens while keeping its structure.

Step 3: Change the reading level. Many recipients are small shop owners who will skim it. She uses Reading level, which ranges from kindergarten to graduate school, and moves it down a few steps. Long sentences split and the vocabulary becomes plainer.

Step 4: Make a targeted edit. The closing paragraph sounds stiff. She highlights only that paragraph and types: "Make this friendlier and end with a clear next step: reply by May 15." Nothing else in the letter changes.

Step 5: Get a second opinion. She chooses Suggest edits. ChatGPT leaves comments on specific passages, such as a suggestion to explain where the sponsor signs will be placed, and she accepts the ones she agrees with. A final pass with Add final polish checks grammar, clarity and consistency.

If any step makes things worse, the back button returns to the previous version. The session above is illustrative; your drafts and suggestions will differ.

Code review works the same way

Canvas has a separate set of shortcuts for code: review code, add logs, add comments, fix bugs, and port code to another language, including JavaScript, TypeScript, Python, Java, C++ or PHP. In TechCrunch's demonstration, Add comments inserted plain-English explanations into a small Python web server, and Review code proposed specific changes that the user could accept, edit or reject one at a time. The same habits apply: highlight the function you care about, ask a precise question, and read each suggestion before accepting it.

How is canvas different from Claude Artifacts?

The two look similar at first glance. Anthropic introduced Artifacts on Claude.ai in June, alongside Claude 3.5 Sonnet: when Claude produces code, a text document or a website design, the result appears in a dedicated window next to the conversation, where you can view it and keep building on it. Canvas, launched four months later, follows the same idea of moving the work out of the chat stream.

The difference is mostly in emphasis. Canvas is built around editing a document together: typing directly in the text, highlighting a passage for a focused change, inline suggestions and a menu of ready-made edits like reading level and length. OpenAI trained a version of GPT-4o specifically for these actions, including when to open canvas and how to write useful inline comments. Artifacts, by contrast, shines at showing what Claude made, especially when the result is something visual, such as a web page design you can see in the side window. With Artifacts, most changes usually come from asking Claude in the chat.

In practice, choose by the job. For revising prose you have written, or reviewing code line by line, canvas's highlight-and-shortcut approach saves time. For quickly generating and viewing a small web page design, Artifacts is the more natural fit. Both are new, and both keep changing, so it is worth trying the same task in each.