Chrome is no longer just a window on the web. It now carries two different kinds of artificial intelligence: Gemini in Chrome, a cloud assistant that sits in a side panel and works with the pages you have open, and Gemini Nano, a small model that runs entirely on your own computer and powers built-in features for websites. This guide covers both, what each one actually does, and where each one still falls short.
The distinction matters because the two behave differently. The side-panel assistant can reach across your tabs, your history and your Google apps, but it sends page content to Google’s servers. The on-device model never sends anything anywhere, but it is smaller, needs capable hardware, and cannot do everything the cloud model can. Most of what people call “Chrome AI” is one of these two, and knowing which one you are using explains its limits.
Two Kinds of Chrome AI
The first kind is the assistant you meet as a user. Click the Ask Gemini icon at the top of Chrome and a side panel opens, letting you chat about the page you are reading without leaving it. It can summarise, explain, compare and answer questions, and it can reach into your other Google apps. It is powered by Gemini in the cloud.
The second kind is infrastructure for websites. Chrome ships with Gemini Nano, a lightweight model of around 1.5 GB, that runs locally on your machine. Websites can call it through built-in browser APIs to summarise text, translate languages, detect languages, draft writing and rewrite it — all without a server round-trip, an API key or a per-token bill.
The two overlap in places. Both can summarise a page. But they are built for different jobs: the assistant helps you browse, while the on-device model helps websites build features that work offline and keep data private. A guide that only covers one of them leaves half the story out.
Gemini in Chrome: The Assistant in the Side Panel
Gemini in Chrome starts with an opt-in. The first time you click Ask Gemini, Chrome asks you to turn the feature on, and from then on the panel is one click away on desktop and built into the Chrome app on iOS. On Android, Gemini can be called up from the power button and works with what is on your screen, including Chrome.
By default, when you start a new chat, Chrome shares your current tab with Gemini so it can answer about what you are reading. You can change that in Settings, under AI innovations, by turning off sharing the current tab by default, and you can add or remove individual tabs from any conversation. That control is worth knowing before you open anything sensitive in front of the assistant.
What you can ask is broad: summarise the key takeaways from an article, explain a complex topic in simpler terms, test your knowledge of something you are learning, adapt a recipe to a dietary need, or make recommendations based on your preferences. The panel stays open while you browse, and you can pop the conversation out into its own window if you want more room.
Availability has limits. You need to be 18 or over, signed in to Chrome, running the latest version, and using a supported language and region. It does not work in Incognito mode. Google first rolled it out in the United States, then expanded to India, New Zealand and Canada in March 2026, with support for more than 50 additional languages.
Working Across Tabs
The side panel’s most useful trick is that it is not limited to one page. On desktop, you can share additional open tabs with a conversation and ask Gemini to connect information across them. Research that used to mean juggling tabs and copying notes becomes a single question.
Planning a trip is the classic example: flights on one tab, hotels on another, activities on a third, and Gemini pulls them into one itinerary. Comparing products works the same way — open the competing pages, share them, and ask for the differences that matter, such as price, warranty or battery life, instead of re-reading each page yourself.
Google also says Gemini can turn page content into a practice quiz, which suits students and anyone onboarding into a new subject. Read a chapter, ask for a quiz on it, and answer from memory before checking. It is a small feature, but it shows the direction: the browser is becoming a place where you process what you read, not just display it.
Cross-tab answers are only as good as the tabs you share. If a tab is missing, the answer will be too, and Gemini will not always tell you what it did not see. For decisions that matter, check the answer against the source tab before acting on it.
Talking to Chrome: Gemini Live and Voice Navigation
Beyond typing, you can go Live with Gemini in Chrome and talk naturally while you browse. You click Ask Gemini, then Go Live, and the conversation becomes spoken: you can interrupt mid-response to add detail or change topic, the way you would with a person.
Live can also navigate the page by voice. Ask it to find specific information, scroll to a recipe section, or highlight the relevant part of a long page, and it moves the page for you. That is genuinely useful when your hands are busy, such as following a craft project or cooking from a long recipe list.
Voice navigation needs two permissions: page content sharing and microphone access. It is also rolling out gradually, so it may not be available on your machine yet, and it cannot navigate PDFs by voice. Live is not available to users under 18.
Finding Things You Already Saw
One of the quietest useful features is history search. Instead of keeping tabs open or remembering exactly when you saw something, you can describe it in everyday language: the chicken parmesan recipe from last week, the hotel you looked at in Palm Springs, the article about the stock market from this morning.
Gemini in Chrome searches your synced Chrome history and brings the page back, so you can resume where you left off. It works because you describe the thing, not the address or the date, and it removes one of the oldest browser chores: the tab you kept open for three weeks because you were afraid of losing it.
There are conditions. You need to have opted in, updated Chrome, signed in with a personal Google Account, and synced your history. It is not available in Incognito mode, not with a work or school account, and not inside Live chats. If your history is not synced, there is nothing for Gemini to search.
Chrome and Your Other Google Apps
Gemini in Chrome also connects to your Google apps, which is where it stops being a page reader and starts being an assistant. From the panel, you can create events in your Calendar, summarise a YouTube video without watching all of it, draft messages to send through Gmail, and look up past photos and albums.
The YouTube case is worth a closer look. Asking for the key points of a long video, or jumping to a specific moment, saves real time, but a summary is a compression: it drops detail by design. For a tutorial where a single step matters, watch the step itself rather than trusting the summary’s version of it.
When you share a Workspace page, such as a Google Doc, Gemini may access your Workspace account directly so it can refer to the whole document rather than only what is visible in the tab. That is convenient and also a reason to check which tabs you have shared before asking about work documents.
Nano Banana in Chrome
Chrome also carries Google’s latest image model, Nano Banana, inside the assistant. Ask Gemini in Chrome to visualise a webpage as an infographic, or to transform an image you find on the web, and it creates the image without you leaving the page.
This is best thought of as a brainstorming tool rather than a design tool. An infographic of a dense article makes the structure visible at a glance, and a transformed image gives you a starting point for a presentation or a post. For anything you will publish, check the result closely: generated images can garble text, invent details and misrepresent the source page.
Auto Browse: When Chrome Does the Task
The most ambitious Chrome AI feature is Auto Browse, which moves from answering questions to carrying out multi-step tasks on your behalf, such as filling a shopping cart or making a reservation. Instead of guiding you through the steps, it performs them while you watch.
Auto Browse is a separate capability from the side panel, and its availability is narrower: Google has limited it to AI Pro and Ultra subscribers in the United States and India, in English, on Android and desktop. Agentic browsing is also the feature where verification matters most. A booking made by an agent is still your booking, so check the dates, the price and the details before confirming anything it did.
The Other Chrome AI: Gemini Nano on Your Device
Separate from the assistant, Chrome ships with Gemini Nano, a small model that runs entirely on your computer. Once its roughly 1.5 GB of weights are downloaded through Chrome’s Optimization Guide component, the model is shared across all websites and extensions on that machine, so it is not downloaded again per site.
Websites reach it through a set of built-in APIs. The Prompt API offers general-purpose local inference for custom tasks. The Summarizer API condenses articles, tickets or transcripts into key points, a short summary, a teaser or a headline.
The Translator and Language Detector APIs translate between languages and identify the language of a piece of text, both offline. The Writer and Rewriter APIs draft original text and reshape its tone or length inside web forms.
When a site checks whether the model is available, Chrome answers with one of three states: readily, when the hardware qualifies and the weights are already on disk; after-download, when the hardware qualifies and the weights are downloading; or no, when the hardware does not meet the bar.
Chrome’s published requirements for on-device AI include a desktop operating system, at least 22 GB of free disk space on the profile’s volume, and a graphics processor with at least 4 GB of video memory.
The privacy advantage is real: summarising a support ticket or a medical form locally means the text never leaves the machine, which is why privacy-first extensions use the Summarizer API for sensitive content. The latency advantage is real too, since there is no network round-trip.
But the model is small, so it is not suited to deep factual recall, and very long documents can overwhelm its context window. Developers work around that by splitting long pages into smaller chunks locally, and by falling back to cloud models when the device cannot run Nano.
For the consumer side of translation, including live spoken translation, see our guide to how Google Translate and Gemini Live Translate compare. The on-device Translator API and the streaming Live Translate engine solve different problems: one translates text inside a website without a server, the other translates speech in real time.
Privacy Controls Worth Setting
Chrome AI’s controls are simple, but only if you set them deliberately. Turn off sharing the current tab by default if you do not want every new chat to see your page. Remove tabs from a conversation when the topic changes. Pause the assistant’s activity when you do not need it, and delete your Gemini history when you want a clean slate.
The same discipline applies to the on-device side. Just because Nano runs locally does not mean every website should get access to it: a site can only use the APIs when the browser grants them, and keeping Chrome updated keeps those grants under the latest rules. For work machines, administrators can manage Gemini in Chrome through policy, including whether it is enabled at all.
Getting the Most From Gemini in Chrome: Five Habits
First, share deliberately. Before asking a question, look at which tabs are shared with the conversation. A comparison across three product pages is only meaningful if all three pages are actually in the conversation, and a question about a private document should never be asked with unrelated tabs still shared.
Second, ask for structure. “Summarise this” gives you a paragraph; “give me a table of price, warranty and battery life for each product” gives you something you can act on. Comparisons, tables and short lists are where the panel earns its place, because they turn reading into a decision.
Third, use it to learn, not just to read. After a long article, ask for a quiz on the content and answer before looking. The quiz feature exists for exactly this, and it converts passive reading into something you will remember next week.
Fourth, verify the load-bearing facts. If a summary gives you a price, a date or a statistic that you will repeat to someone else, open the source tab and check it. Summaries compress, and compression drops caveats. The panel is fastest for the first pass; the source page remains the authority.
Fifth, clean up after yourself. When a research session ends, start a new chat for the next topic rather than letting tabs and context accumulate. Old tabs in a conversation are how a new question gets answered from last week’s pages, and a fresh chat is the simplest way to prevent it.
What Chrome AI Cannot Do Yet
Chrome AI is a strong reading and research companion, but it is not a reliable source of truth. Summaries can drop caveats, comparisons can miss the decisive difference, and history search can surface the wrong page with confidence. For anything where an error costs money, health or reputation, verify against the original page.
Availability is uneven by region, language, plan and device, and features arrive at different times: Live voice navigation, Auto Browse and some connected-app features may not be on your machine yet. The assistant also cannot see Incognito tabs, cannot search history inside Live chats, and cannot navigate PDFs by voice.
The on-device model has its own ceiling. It is small by design, so it will not match the cloud model’s knowledge or reasoning, and it needs capable hardware that many older machines lack. When a site reports the model as unavailable, the honest answer is usually a cloud fallback, not a settings fix.
Used with those limits in mind, Chrome AI genuinely changes browsing: fewer tabs kept open out of fear, faster first passes over long pages, and a browser that can answer, compare and act instead of only displaying. The skill is knowing which of the two AIs you are talking to, and what it can and cannot see.
Background
Chrome’s AI story began with on-device experiments in 2024, when early prototypes exposed a single text session API behind testing flags. Google then replaced that prototype with task-specific APIs and, separately, built Gemini into the browser as a side-panel assistant, expanding it from the United States to more countries and languages through 2026. You can browse the subjects we cover on our topics index.
Sources: Google Chrome Help: “Go Live help”; Google Chrome: “Gemini in Chrome”.
0 Comments