Gemini moves file editing into the chat box: PDFs can be previewed and edited on web and mobile, Docs, Sheets, slides, Markdown and code stay web-only, and it is open to all users in all languages

The source. The primary material is Google's own blog post, You can now generate, preview, and edit documents, PDFs, presentations, spreadsheets, and more directly in Gemini, published on blog.google on October 29, 2026 under the Gemini App category. Every capability, format and availability statement below comes from that page. Anything the page does not address, including which model powers it, how faithful the editing is, whether there are file-size limits, and whether Workspace or enterprise tiers get the same access, is not filled in here.

What it actually does. Four things: generate, preview, edit, follow up, all inside one conversation. Generating means asking Gemini for a file rather than a paragraph. Previewing means the file is visible in the chat instead of being downloaded and opened elsewhere. Editing takes two routes: expand to full screen and type the changes yourself, or highlight a section and hand it to Gemini to revise the text or the visuals. Following up means asking questions while you edit. Google's example strings all four together: upload handwritten lecture notes, generate a PDF study guide, edit it by hand, highlight a section for the model to change, then keep asking. Not one step in that sequence requires leaving the chat window.

There is one explicit line in the format support, and it matters more than the feature list. PDF preview and editing work on both web and mobile. Every other format, Google Docs, Google Sheets, slide decks, Markdown files, TXT files and code, is web only. In other words you can edit a PDF on your phone but not a Doc or a Sheet. If you were planning to treat this as a mobile workflow, that single limit decides whether you can finish the work on a commute. Google does not explain the split, and it is reasonable to read it as a difference in client-side layout capability, but for a user it is a boundary worth knowing before you rely on it.

Availability is stated flatly: all users, in all languages where the Gemini app is available. Those two alls are the only part of the announcement you can verify immediately, because nothing here requires an application, a waitlist or a paid tier. Compare Google's usual pattern when it ships a new model, where access starts with a small group or a paid tier, and the cadence is clearly different. That says something about what this is: not a capability demo but a distribution move for something that already works.

What Google did not say is exactly what you should ask. First, the model is unnamed, which means you cannot infer a quality ceiling from the announcement or tell whether regional versions differ. Second, editing fidelity is unaddressed: hand it a PDF with multi-level headings, running headers, footnotes, a table of contents and complex tables, ask it to change one sentence, and whether the original layout survives is not mentioned once. Third, file size and page limits are absent. Fourth, whether enterprise and education tiers get the same access is absent. None of that is nitpicking. Anyone who has shipped a document feature knows those four items decide whether it is usable in real work.

What actually changed is the entry point, not the capability. Generating documents, rewriting text, editing visuals from a picture — standalone AI products have done these for a while. The difference is that they now live in the chat window, which is where users already are. Product design has an old rule: every extra jump costs you half your usage. Before this, editing a slide deck with AI meant export, upload, download, re-import, and if any one of those four steps felt like a hassle the feature died in the user's head. Now four steps collapse to zero. That will move usage frequency far more than any model improvement would.

For people who produce content there is a specific reading here: a document is turning from a file into an object inside a session. A PDF used to be a bounded thing on a disk with a version number, mailed back and forth. Now it may only exist inside one conversation, exported when done, with the edit history living in that conversation's context. The upside is speed. The cost is weaker traceability, because answering who changed that sentence and when becomes hard. If your work needs an audit trail, note that nothing in the announced capability provides version history or a change log, and think about that before you move a process onto it.

One sober note to finish. Shipping a feature to all users is good news, but there is a gap between can use and dare use. When it turns handwritten notes into a study guide, you will spot a mistake at a glance. On a contract, a quotation or a spreadsheet with formulas, one wrong character is money. Google publishes no accuracy figure and no fidelity promise, so use it the conservative way: let it produce the first draft, keep the last pass human. That advice has nothing to do with how capable the model is and everything to do with who signs off.