Canvas is Gemini’s answer to “stop scrolling back through chat to find your draft.” Open it and the document sits in a panel beside the conversation: you edit text inline, highlight a paragraph and request a targeted rewrite with Select & ask, and keep the chat focused on change requests instead of full re-pastes. As of June 2026 Canvas is free for every Gemini user, and Google AI Pro and Ultra subscribers run it on Gemini 3.1 Pro with a 1M-token context window. This guide is the working loop that turns Canvas from a demo into the place a draft actually lives.
TL;DR
- Get a rough draft into chat first, then open Canvas (tools menu / “Add Files” → Canvas). Never open Canvas on an empty intent.
- For scoped edits, highlight the text and use Select & ask — the inline “Ask Gemini” box edits only the selection, which is the single biggest fix for “Gemini rewrote my whole doc.”
- Use the Previous Version / Next Version buttons to step back through the active session; it is not a permanent version archive.
- Export every few iterations: Export to Docs, Export to Slides, or Export to Colab for Python. The chat is the working copy, the export is the deliverable.
- Free tier works; Gemini 3.1 Pro and the 1M-token window are gated behind Google AI Pro ($19.99/mo) and Ultra ($99.99/mo) as of June 2026.
What Canvas actually is
Canvas is a split-screen surface inside the Gemini app. The left side stays a normal chat; the right side holds one live artifact — a document, a slide deck, or a working code app — that auto-saves as you go. You can type directly into the doc, use the formatting toolbar, or drive changes through the prompt box. The point is a single source of truth that both you and Gemini edit, instead of a draft buried 40 messages up the chat.
Two things make it worth the context switch:
- Select & ask — highlight any chunk of text (or, in an app, click a UI element) and an inline “Ask Gemini” box appears. Your instruction applies to that selection only. In a plain chat, asking for a change can re-roll the entire response; in Canvas the edit is scoped to what you selected.
- Version stepping — Previous Version and Next Version buttons let you walk back through edits in the current session, the same way undo/redo works, so a bad rewrite is one click from being reverted.
Who this is for
People who already draft in Gemini but lose track of versions, or who keep copying half a doc back into chat to say “edit this part.” It is most useful for writers iterating on one artifact across several days and engineers refining a single script or small app prototype.
When to reach for it (and when not)
Reach for Canvas when you expect at least three iterations on one document, want to see the doc and the conversation side by side, or plan to export the final version into Google Docs, Slides, or Colab. Skip it for one-shot questions and free-form brainstorming where structure does not matter yet.
It is the wrong tool for multi-file software projects, live multi-author co-editing of prose, or anything needing fresh external data mid-edit. For deep research syntheses, run the Gemini Deep Research workflow first, then pull the verified result into Canvas (the Create button can turn a Deep Research report straight into a Canvas page).
Free vs paid: what you actually get
Canvas itself costs nothing — text docs, code, slides, Select & ask, and exports all work on the free Gemini tier. The paid tiers change the engine, not the feature set.
| Tier (June 2026) | Price/mo | Model in Canvas | Context window |
|---|---|---|---|
| Gemini (free) | $0 | Gemini 3 (standard) | Standard |
| Google AI Pro | $19.99 | Gemini 3.1 Pro | 1M tokens |
| Google AI Ultra | $99.99 | Gemini 3.1 Pro | 1M tokens, higher limits |
Google AI Pro is the tier formerly sold as “Gemini Advanced” / “Google One AI Premium”; it was renamed in early 2026 and bundles the 1M-token context plus Gemini in Workspace. If you are pasting long source material into a single Canvas session, the larger window on a paid tier is the difference between Gemini holding the whole document and silently truncating it.
Before you start
- Pick the artifact: one doc, one script, one outline. Canvas works on one thing at a time.
- Decide where the final version lives — Google Docs, GitHub, Notion. Canvas auto-saves but is tied to the chat, so treat it as a working surface, not durable storage.
- Get your source material into chat before opening Canvas: notes, transcripts, the prior draft. Once Canvas is open, the prompt box is for change requests, not context dumps.
- Write a one-sentence spec — audience, voice, must-haves — and pin it as the first message.
Step by step
- Start a normal Gemini chat, paste your spec plus source material, and get a rough draft in chat. Do not open Canvas on a blank intent.
- Once the draft has shape, open Canvas from the tools menu (Add Files → Canvas) or ask Gemini to “put this in Canvas.” The artifact moves to the right panel and the chat becomes the change channel.
- Make small fixes directly in the doc — typos, single-line tweaks — using inline editing and the formatting toolbar. Changes auto-save and Gemini sees them on the next turn.
- For a scoped change, highlight the exact text and click Select & ask (bottom right), then type the instruction in the inline box: “Expand this bullet into a paragraph with two concrete examples.” The edit stays on the selection.
- For quick tone or length passes, use the Change tone and Change length buttons instead of a full prompt.
- If a chat-box prompt rewrites more than you asked, hit Previous Version to revert, then re-issue the edit through Select & ask so it can only touch the selection.
- Building an app instead of a doc? Toggle Code / Preview to switch back-end and front-end views, and click Show console to read errors. Canvas auto-attempts to fix runtime errors; Show recent changes shows what it altered.
- Export every few iterations: Export to Docs, Export to Slides, or Export to Colab for Python. Canvas is per-chat — lose the chat, lose the history.
First-run exercise
- Take a piece you have already drafted once — a post, a memo, a script — and paste it into chat as the starting state.
- Open Canvas and run three edits in a row, each on one section, using Select & ask for the scoped ones.
- After each, step back with Previous Version and compare. Did the selection-scoped edits stay contained while the chat-box edits drifted into neighboring text?
- Note which method held. In practice, Select & ask is the reliable scoping mechanism; chat-box prompts still benefit from naming the section explicitly.
Quality check
- Did only the requested edit happen? Step back one version and eyeball the diff.
- Are facts, names, and figures intact across rewrites? Models silently paraphrase numbers when they restructure paragraphs, so re-check any specific data after a big rewrite.
- Does the voice still match the spec, or has Gemini drifted toward its default tone? Three edits in, drift is common.
- Is the exported version current? If Canvas is two iterations ahead of your Google Doc, that is technical debt.
How to reuse this workflow
- Save your spec template as a snippet — audience, voice, must-haves, do-nots — and paste it as the first message of every Canvas session.
- Default to Select & ask for anything scoped; reserve chat-box prompts for whole-document passes.
- Export to the source-of-truth tool at the end of every session. Treat Canvas as a working copy, never the original.
- Re-run a small Canvas session every month or so on the same kind of artifact. Canvas ships changes often — the Select & ask, vibe-coding, and Create menu features all arrived through 2025–2026 updates — so the buttons move.
Recommended workflow
Spec in chat → rough draft in chat → open Canvas → scoped edits via Select & ask → step back with Previous Version if anything overreaches → Export to Docs → close the chat. The chat is scratch; the export is the deliverable.
Common mistakes
- Opening Canvas before any draft exists in chat — you get a blank panel with no anchor.
- Typing “make it better” into the chat box instead of highlighting and using Select & ask. Vague whole-doc prompts are what trigger full rewrites.
- Treating Canvas as durable storage. It auto-saves, but the version steps live in the active session and the chat itself can be pruned.
- Forgetting to export. Two weeks later the chat is gone and the final draft with it.
- Expecting shared Canvas links to open in the mobile app — they only open on gemini.google.com on desktop.
- Mixing two artifacts in one Canvas. It is single-document by design; open a second chat for the second artifact.
FAQ
- Is Gemini Canvas free?: Yes. Documents, code, slides, Select & ask, and exports all work on the free Gemini tier as of June 2026. Google AI Pro ($19.99/mo) and Ultra ($99.99/mo) upgrade the underlying model to Gemini 3.1 Pro and the context window to 1M tokens.
- How do I edit only one section without Gemini rewriting everything?: Highlight the text and use Select & ask. The inline “Ask Gemini” box applies your instruction to the selection only, which is the native fix for runaway rewrites.
- Can I undo a bad edit?: Yes. Use the Previous Version / Next Version buttons to step through the active session, or ask Gemini to revert. Build the habit of exporting state externally anyway, since version steps are session-scoped.
- Can I share a Canvas with a teammate?: Use Share & export for a link, or export to Google Docs. Shared Canvas links open only on gemini.google.com on desktop, not inside the mobile app.
- Does Canvas work on mobile?: Partly. The mobile app can view and prompt, but the formatting toolbar and serious editing are desktop-web only, and shared links don’t open in the app.
- Can Canvas build a working app, not just a doc?: Yes. Describe the app and Canvas generates the code, with Code / Preview toggles, a console, and auto-fixing of runtime errors. Apps can save data between sessions and share data between users, and Python can go straight to Colab via Export to Colab.
Related
External references: Gemini Canvas overview and the Create docs, apps & more with Canvas help page.