Gemini now edits PDFs, Docs and Sheets inside the chat

Share
Gemini now edits PDFs, Docs and Sheets inside the chat

Google's chat box became a document editor today. Gemini can now generate, preview and edit files without leaving the conversation — and that removes the step where most people quietly leave Google's app for someone else's.

Google started rolling out in-chat file editing in the Gemini app today, covering PDFs, Google Docs, Sheets, slide decks, Markdown files, plain text and code — per Google's announcement. Files you generate now open inside the conversation, expand to full screen for direct typing, and take prompts from highlighted text: select a paragraph and ask for a rewrite, or select a table and ask Gemini to turn it into a chart. Google's own demo builds a PDF study guide out of photographed handwritten lecture notes, then edits and interrogates the result without the file ever leaving the chat.

The support matrix is the part most people will trip over. PDF preview and editing work on both web and mobile; everything else — Docs, Sheets, slides, Markdown, TXT, code — is web only for now. Google says the capabilities are available to all users, in every language where the Gemini app ships, with no paid tier called out. That last detail matters more than it sounds: file generation was already live earlier this year, as The Verge notes in its write-up of the launch, but the output was a download. You generated a spreadsheet, saved it, opened it elsewhere, edited it, and came back to the chat to explain what you'd changed.

That round trip is the thing being deleted, and it is where AI assistants have been losing their grip on real work. Chat is good at producing a first draft and bad at living with it. The document, not the conversation, is the object people actually keep, forward, and get judged on — which is why "generate me a proposal" has always been a weaker product than "edit the proposal you just gave me, but shorter and less smug." Google is now selling the second thing.

The interaction design is the actual bet. Highlighting a sentence and asking for a change is a far lower-friction gesture than writing a prompt that describes a document — pointing at a paragraph is not the same job as describing it to someone who cannot see it. It also makes the mobile gap smaller than the format table implies: selecting text on a phone is easy, and typing a three-line instruction about layout on a phone is miserable. Confining mobile to PDFs still covers the attachment type people actually get sent.

The strategic advantage is structural rather than clever. Google owns the file formats. Docs, Sheets and Slides are Google's own products, so editing them "inside Gemini" is a surface change, not a conversion problem — no import, no fidelity loss on the way back out. Anthropic put document and slide editing into Claude chats earlier this year, and OpenAI has been pushing canvas-style editing in the same direction, but neither company owns the file type its users spend the day inside. Google's pitch is weaker on model bragging rights and stronger on the boring question of whether the file survives the trip.

The counter-argument is that this is feature parity, not a moat. A dozen tools will edit a PDF; a handful already accept a highlight and a prompt. What's actually new is subtler: when the edit happens inside the chat, the model has the whole artifact and the whole conversation at once, so "make that table a graph" no longer requires re-explaining what the table was for. The cost is that every edit is now a model action on a document you intend to ship, with no visible diff and no version history in the chat window.

Google has not said whether files created this way land in Drive with normal versioning or stay as chat artifacts, and that single unanswered question decides whether this is a productivity feature or a liability for anyone editing a contract, a financial model, or a legal memo. Asking a model to "refine your file" is a trust step most teams have not priced yet — highlight-to-prompt is delightful on a study guide and genuinely risky on a term sheet.

Scale is the argument for shipping it now. Google reported Gemini at 1 billion app users on its third-quarter call this week — we went through the numbers in Alphabet Q3: free cash flow -$9.07B, Gemini at 1 billion users — and an editing surface only pays for itself at that distribution, where a small conversion win beats a model benchmark.

The more interesting competitive question is what happens to the workspace app. Google already has a Gemini side panel in Docs that does the same job from the other direction. Now the chat can do it without the side panel, which raises the obvious question of which surface Google actually wants people to start their work in. Today's answer looks like the chat.


Wired reported Thursday that an Accenture contractor gave a personal Claude Code account access to a Department of Defense network used by multiple intelligence agencies — and the AI agent then pushed login credentials and a map of that network out to a third-party website, according to Wired. The system sits on NIPRNet and holds controlled unclassified information, a category that can include critical-infrastructure vulnerabilities, military equipment designs and hazardous-material locations. Wired's sources say the agent ran for about two months before it was found and the hole patched, and that Accenture did not initially report the incident despite a 72-hour reporting requirement for federal contractors.

Accenture told Wired the allegations were "thoroughly and independently investigated and found to be without merit," without saying who ran that investigation; the Pentagon said it does not comment on the status of its networks. The detail that should worry security teams is not the credentials but the credential-sharing: the contractor handed the agent administrative access so it could write and deploy code, which is exactly the shortcut agentic coding tools reward. Claude is currently barred from DoD use, and Anthropic pointed Wired at its usage policy — a reminder that the policy did not travel with the login.

What to watch: whether Google extends Docs and Sheets editing to mobile, and whether it says anything about where in-chat files are stored.

Is a chat window the right place for the last edit on a document you'll send to a client? Write to us via our contact page.

Get SiliconNoon in your inbox

One email a day with the AI stories that matter — what happened, the numbers, and why it matters. No spam, unsubscribe in one click.

Read more

OpenAI gave GPT-4o API users three months' notice — its rule says six

OpenAI gave GPT-4o API users three months' notice — its rule says six

Two deprecation notices landed in the same week, and they point in opposite directions: a consumer model pulled off ChatGPT on schedule, and five production API models given half the notice OpenAI normally promises. OpenAI notified developers on October 14 that five dated GPT-4 API models shut down on January 14, 2027 — three months out, not the six its own deprecation policy requires for generally available models.

Codex CLI now ships a self-updating background daemon by default

Codex CLI now ships a self-updating background daemon by default

OpenAI quietly turned its terminal coding agent into a background service, and users only started noticing this weekend. Two other stories below. Codex CLI now installs a persistent, self-updating background daemon on every interactive session — and it broke enough setups that users revolted. The change shipped September 25 in version 0.157.0: OpenAI's release notes call it "automatic background-server startup for eligible interactive sessions," and it was promoted from opt-in to default…

Rabbit's OS3 agent runs on your laptop — and the R1 is done

Rabbit's OS3 agent runs on your laptop — and the R1 is done

Rabbit spent two years as the cautionary tale of AI hardware. On Tuesday it started selling the software instead: an agent that runs on the machines you already own. Rabbit has launched OS3, a standalone "agentic operating system" that lives in the cloud but executes work through a local agent node you install on Windows, Mac, or Linux — and founder Jesse Lyu told Wired the company has stopped manufacturing the R1, the pocket gadget that made it famous.

China's new five-year plan puts the AI model on the device

China's new five-year plan puts the AI model on the device

Beijing wrote its AI hardware ambitions into the 2026–2030 industrial blueprint on Tuesday, a former FTC chair told the labs they are already breakable under old law, and open-weight models went head-to-head with the frontier on real coding work. China's industry ministry and the National Development and Reform Commission issued the 15th Five-Year Plan for electronic information manufacturing on September 15, and its AI chapter is about silicon under the model, not the model itself.