Most AI applications send text to an AI provider’s server and get an answer back. Google is building a different route into Chrome: AI features that run on the user’s own computer, with a small language model built into the browser. For small businesses with a website, online shop or their own web application that is interesting, but it does not replace everything. Below you will read what it is, what you can do with it and where to be careful.
What Chrome’s built-in AI is
Google offers a set of AI features and APIs in Chrome that let a website or extension have language tasks carried out by the browser itself. For some of these, Chrome uses Gemini Nano, a compact language model from Google that runs locally on the device. For those tasks the text does not need to go to an external AI service.
Which features exist
- Translating: converting text from one language into another.
- Detecting language: working out which language someone is typing in.
- Summarising: reducing a long text to its essentials.
- Writing and rewriting: producing a first draft or rephrasing a text.
- Open prompts: asking the local model your own question through the so-called Prompt API.
Not all of these features are equally mature. Translation, language detection and summarising were the first to reach the stable version of Chrome; other features were still experimental or only available in a trial phase when this article was written. The current status of each feature is in Chrome’s developer documentation.
Where it runs
The model is downloaded on first use and needs storage space, memory and a suitable device. Whether a feature works depends on the visitor’s Chrome version, operating system and hardware. So do not assume everyone has it, certainly not on phones or in other browsers.
Opportunities for small businesses
The built-in AI suits small, well-defined text tasks that help the visitor, without sensitive data leaving the device.
- Online shop: translating a product description or reviews for a visitor from abroad.
- Hotel: recognising which language a guest is typing in and choosing the right next step.
- Staffing agency: summarising a long job advert into its key points.
- Installer: helping a customer describe a fault clearly before they send the form.
The advantage: there is no separate bill per AI call and the text stays on the device. The drawback: you have no control over whether it works for the visitor who has just landed on your site.
Where the limits are
A small local model is no substitute for a large model on a server. Bear in mind:
- Varying availability: some of your visitors will not have the feature. Your website must work just as well without AI.
- Limited quality: fine for short, simple tasks, less reliable for long or complex texts.
- No knowledge of your business: the model knows nothing about your diary, stock or customers. For that you need your own software.
- No decisions: do not use it for financial, legal or medical advice, or for anything where mistakes cause harm.
Privacy: what local really means
The fact that the model runs on the device is the strongest argument. Text a visitor has translated or summarised does not need to go to an external AI provider for that. That does not automatically make an application compliant with the GDPR. Also look at what your website itself does with the input and the result: do you store it, do you send it on to your own server, who can access it? Those questions stay the same, with or without local AI.
How to use it sensibly
- Pick one task: start with something small, such as translating or summarising, where a mistake does little harm.
- Check availability: have your website first ask whether the feature is available on this device, and only show the button if it is.
- Build a fallback: without built-in AI the visitor gets the normal version, or you use an AI service on the server.
- Let a person decide: a suggested text is a suggestion; the visitor or your member of staff adjusts and sends it.
- Measure usage: check whether visitors actually use the feature before you develop it further.
Where it fits alongside your own software
For tasks that need to know your business, such as answering questions about appointments, moving a booking or preparing a quote, you need AI that can access your data. That is why the AI assistant is built into our software per industry, with human oversight for doubtful cases. Chrome’s built-in AI can be useful alongside that for small tasks in your visitor’s browser.
Frequently asked questions
Does Chrome’s built-in AI work for all my visitors?
No. It depends on the Chrome version, the device and the operating system. On phones and in other browsers you should assume it is not available.
Does using the built-in AI cost money per request?
There is no bill per call, because the model runs on the visitor’s device. Development, testing and maintenance do take time.
Is local AI automatically GDPR-compliant?
No. It limits which data goes to an AI provider, but you still need to manage what your website itself stores and sends on.
Can the built-in AI replace a chatbot on my site?
Usually not. The model does not know your business and is not available to every visitor. For customer questions you need an assistant that can access your own data.