Google has pushed a set of five new capabilities into Gemini, the company's AI assistant, extending it from a chat window into a tool that builds websites, fills PDFs, searches Gmail, drafts documents by voice and assembles slide decks. The features, reviewed by Your AI Workflow, mark a further push by the search giant to embed generative AI directly into the daily routines of its Workspace customers.
The most visually striking of the five is website creation. Users upload images, a logo and product details, and Gemini generates a storefront or landing page on a canvas in roughly a minute. Finished pages can be exported to Google Sites with a single click. In testing, the tool produced six product images of hoodies, t-shirts and backpacks in light and dark colorways, each carrying a minimal logo, and then built a "minimal premium design" storefront around them. The result was described as polished, with professional copy and smooth transitions.
There are limits. Because Gemini publishes its output to Google Sites as an embed, the top navigation bar cannot be stripped out, which may force users into manual adjustments. Uploaded images are hosted at unique links in webp format, and site owners can choose to make a Google Site public or restrict it to their organization.
A second feature tackles PDFs. Gemini can fill document fields inside the chat interface, pulling data from connected Workspace apps. When asked to complete an invoice, it retrieved the recipient's company name, address, itemized prices, total due and tax ID from Gmail threads, and recorded the payment method as a wire transfer. The catch is that completed documents are not attached to emails automatically; users must download the file and upload it to a Gmail draft themselves. A separate PDF filler tool offers sharing options, adjustable access permissions and notifications when a document is opened.
Gmail itself gets an "Ask Gmail" bar that replaces the standard search field and returns AI-generated overviews. Asked about the status of an invoice, Gemini combed multiple threads and notifications to summarize billing details, scope and current status, and identified when the recipient had opened the shared invoice. The feature performs well on narrow, person-specific queries but fails on broader ones: a question about who needs to be followed up with to complete invoice payments produced no AI overview, and users had to ask about a specific individual instead.
Voice-based document creation arrives in Google Docs through a "create with Gemini" button on iPhone. Users dictate a prompt and receive a detailed draft. In one test, Gemini produced five video concepts for a merchandise launch, complete with style suggestions, audio ideas, hooks and timelines. A source settings tab lets users specify which apps or web sources the model should draw on, so output can be aligned with the material at hand.
Google Slides rounds out the set. From a single prompt, Gemini generates a complete presentation, optionally using documents or existing decks in Google Drive as style and content references. Before generating, it confirms the audience, style, information focus, length and core topics. The test deck for a merch launch included an introduction, launch timeline, road map, product lineup and pricing. Images cannot be attached automatically, so users replace placeholders by clicking "replace image" and uploading files, and they can edit, remove or add slides in Gemini's plan before the deck is built.
The five features sit on top of a broader Gemini build-out. The platform now runs three main models, Flash-Lite for everyday summaries and brainstorming on free accounts, Flash for moderately complex reasoning, and Pro for math, coding and large files on paid tiers. Pro and Ultra subscribers also get Gemini 4 Argon, aimed at long multi-step tasks and tuned for concise answers, though it consumes plan limits faster. A larger model, Gemini 4 Carbon, remains limited to select security partners.
FeatureWhat it doesManual step that remainsWebsite creationBuilds storefronts and landing pages on a canvas, exports to Google SitesTop navigation bar cannot be removedPDF fillingPopulates invoice and form fields using data from Gmail and other Workspace appsCompleted file must be downloaded and attached to an email by handAsk GmailReplaces the search bar with AI-generated overviews of email threadsWorks only for narrow, specific queriesVoice document creationDrafts Google Docs from dictated prompts on iPhoneNone flagged in testingAutomated presentationsGenerates Google Slides decks from a prompt, with style and length controlsImages must be replaced manually
Note: Based on hands-on testing of the five features.
Beyond content creation, Gemini has been building out automation. Its Scheduled page lets users run up to 10 recurring or event-triggered actions, such as a weekly industry news digest or a spreadsheet update when a supplier email reports a shipment delay, with results delivered as notifications or marked unread. Task mode, available on Pro and Ultra plans, chains multi-step workflows across Gmail, Calendar, Docs, Sheets and Drive and can browse the web; in one demonstration it compiled a spreadsheet of cafe emails, filtered out irrelevant senders and drafted replies using a saved instruction set, flagging missing delivery details for review.
Reusable instructions, called skills, let users store preferences such as tone, length or a closing question, along with reference files. They fire automatically when a prompt matches their description, and Google warns that outputs should be checked: a skill-generated announcement in testing listed "250 g bags" instead of "1 lb bags." Notebooks, meanwhile, centralize project files, chats and instructions, with an optional memory setting that makes the model consider all notebook content when answering.
The productivity push coincides with Google's effort to move Gemini beyond the screen. XREAL AURA, an augmented-reality headset built with Google and Qualcomm, runs Android XR and Gemini and goes on sale in Japan on Nov. 27, priced at ¥229,800 (approximately $1,500) for a 12GB RAM and 256GB storage model and ¥269,800 (approximately $1,700) for a 16GB and 512GB version. Buyers who prepay in full through the official online store get a limited-time ¥29,900 (approximately $190) discount, bringing the two configurations to ¥199,900 (approximately $1,300) and ¥239,900 (approximately $1,500).
The device pairs glasses weighing under 95 grams with a 265-gram compute pack powered by Qualcomm's Snapdragon Reality Elite platform, connected by what XREAL calls a split computing architecture. Sony Micro-OLED displays deliver 1,920 by 1,200 pixels per eye at up to 120Hz, with a 70-degree field of view and five-step electrochromic dimming. Gemini can be asked about objects in the user's field of view, and Gemini Live supports real-time voice dialogue, including voice commands to rearrange windows in the space.
For Google, the strategic logic runs in both directions. Each new Gemini feature pushes deeper into Gmail, Docs, Slides and Drive, where switching costs are highest, while Android XR gives the assistant a physical presence that does not depend on a laptop or phone. The company has not disclosed how many Workspace seats have adopted the new tools, and the manual steps left in the website, PDF and Slides workflows suggest the automation is not yet fully hands-off.
For corporate buyers, the near-term question is governance as much as capability. Features that read Gmail threads to populate invoices and search email archives raise data-handling questions that IT departments will want answered before broad rollout, particularly for organizations that restrict Google Sites to internal access. For investors, the updates reinforce Google's position in the productivity software market, where rivals are racing to bundle similar assistants, and add a hardware channel through Android XR that could deepen Gemini's reach if the category gains traction.
Google's challenge is execution at scale. The tools are fast and the output is polished, but the gaps identified in testing, from generic Gmail queries that return nothing to skill-generated copy that gets product details wrong, mean users still need to review what the model produces before it reaches a customer or a colleague.
