Chrome's Gemini autofill: your enquiry form now has an AI filling it in

The short answer

We have argued that what an agent can find, read and act on is becoming the conversion layer. Our guide to WebMCP covers the new standard that lets a page hand an agent a list of tools. Chrome's Gemini autofill is the version of that future that is already shipping to ordinary users, with no new standard at all.

What is Chrome's Gemini autofill?

Chrome's enhanced autofill is an opt-in layer on Chrome autofill that uses Gemini with Personal Intelligence to suggest information from the person's Gmail when they fill in an online form. Suggestions appear in the normal autofill drop-down, marked "Suggested by Gemini". Google announced it on 28 October 2026.

Google's announcement says Chrome can now "fill out forms with suggestions pulled from Gmail", including "order numbers, tracking information, and flight bookings" (Google). Google says it is starting in the US on desktop, Android and iOS, for Gemini in Chrome users who have Personal Intelligence switched on. People can turn it off, or limit what it can reach, in Chrome's settings.

Reporting on the launch adds the fine print. The feature needs a personal Google account, and work, school and supervised accounts are excluded. Personal Intelligence is not offered in the European Economic Area, Switzerland or the UK for now (Notebookcheck). Google's post names only the US, and gives no date for other countries, including Australia.

How is this different from normal Chrome autofill?

Normal Chrome autofill fills in data the person saved: their name, email, addresses, passwords and cards. Enhanced autofill reaches into data the person never typed into Chrome at all. An order confirmation, a shipping email or a booking reference in their inbox becomes a suggested answer.

The step before this came in December 2025. Google said signed-in users could fill their name, email and saved home and work addresses from their Google Account, and fill loyalty cards, travel details and vehicle information from Google Wallet. Those Wallet fills rely on "smarter form understanding", where Chrome works out what kind of form it is looking at and matches the data to it (Google).

Then there is the agent. In January 2026 Google introduced auto browse in Gemini in Chrome for AI Pro and Ultra subscribers in the US. Google's examples include "filling out online forms", using details from a PDF, scheduling appointments and getting quotes from plumbers and electricians. It is designed to "explicitly ask for your confirmation" before sensitive actions such as a purchase (Google).

Read down the table and the direction is plain. Each year the browser fills in more of the form, with data from further away, and needs less of the person's attention to do it.

Why does this matter for a business website?

Because the form is where a visit becomes an enquiry, and the visitor is handing more of the form to software. Every field the browser fills correctly is a field the person does not abandon. Every field it cannot read is a field the person has to fill by hand, on a phone, often at the moment they were about to leave.

The fields enhanced autofill targets are the fields that matter after a sale. A returns form asks for an order number. A warranty claim asks for a purchase date. A travel business asks for a booking reference. A support form asks for a tracking number. Those details used to mean switching to the inbox, searching and copying. Now the browser can offer them in place, if it can tell which field wants them.

The wider agent market is moving the same way. OpenAI began letting US ChatGPT Pro users text their Dot agent on 28 October 2026 (Digital Trends). Anthropic's Claude Duet, announced on 21 October, runs on its own virtual machine and checks with the person before it acts (Engadget). Our piece on AI personal assistants and websites covers what these assistants do when they reach a site. Not all of them say who they are when they arrive, as our guide to AI agent detection shows. When an assistant fills in a form, it has the same problem Chrome autofill has: it needs the page to say what each field is.

That sentence from Google's Lighthouse scoring notes is the practical test. Lighthouse's Agentic Browsing category includes "Accessibility for agents", which checks names and labels: "Ensuring every interactive element has a programmatic name" (Chrome for Developers). A field with no label is invisible to that tree, whoever is filling it.

How do browsers know which field is which?

Mostly from what the page tells them. The clearest signal is the HTML autocomplete attribute, which names the kind of data a field wants. Google's own forms course says the right values "ensure that the browser fills in the correct values for a form" (web.dev). Without it, browsers fall back on the field's label, name and position, and guess.

MDN documents the standard values (MDN). The ones a typical enquiry form needs are short:

The last row is the one enhanced autofill changes. There is no standard autocomplete value for an order number, so Gemini has to work it out from the visible label and the field's name. "Order number" next to a field named order_number is easy. A field labelled "Ref" with a placeholder that disappears on typing, named field_7, is a guess.

A caution on what we can claim. Google has not published how enhanced autofill decides which field gets which suggestion. We cannot tell you that one attribute triggers a Gemini suggestion. What is documented is that browsers use autocomplete and labels to fill forms, and that agents read the accessibility tree. Building to both is the safe bet.

This is also an accessibility requirement, not only a convenience. WCAG 2.2 success criterion 1.3.5, Identify Input Purpose, is level AA. It asks that the purpose of each field collecting information about the user "can be programmatically determined" so browsers and assistive technology can help fill it (W3C). The work that helps a person with a memory or motor disability is the same work that helps an AI fill the form.

What breaks AI form filling?

The same things that break forms for people, only faster. Our contact form best practices cover the fields and errors that lose human enquiries. For browsers and assistants, five patterns cause most of the damage:

• Placeholders instead of labels. A grey hint inside the box is not a label. Once typing starts it disappears, and some form builders never connect it to the field in the accessibility tree.
• Custom widgets. Drop-downs, date pickers and phone inputs built from styled divs rather than real form controls. Autofill cannot fill what it cannot recognise as a field.
• autocomplete="off" everywhere. Some themes and plugins switch it off for the whole form, often to stop spam. MDN notes browsers may ignore it on login fields, but on an enquiry form it simply tells the browser not to help.
• Fields that load late. Forms injected by script after the page loads, or inside a cross-origin iframe, can be missed by autofill and by agents reading the page.
• A CAPTCHA before the form. An agent cannot pass a challenge built to stop software. Put spam protection behind the submit, not in front of the fields.

On WordPress, most enquiry forms come from a form plugin, and whether it outputs proper labels and autocomplete values depends on the plugin and how each field was set up. Checking that is quick, and it is the kind of routine fix a WordPress specialist such as WP Creative would make in the same pass as tidying the form's error messages, rather than as a separate project.

Where is AI form filling heading?

Our view: the form is turning from a page element into an interface between software. Today the browser suggests and the person clicks. Over the next few years, more of the form will arrive filled in, and the person's job will shrink to checking and pressing submit.

We expect three things. These are forecasts, not facts.

• Personal context will fill more fields. Gmail is the first source. We expect order history, bookings and documents the person already has will appear in forms without copying and pasting. The businesses that benefit will be the ones whose forms ask for those details in plain, labelled fields.
• Suggestion will slide into action. Enhanced autofill suggests. Auto browse fills and asks for confirmation. Standards such as WebMCP let a page hand the whole form to an agent as a tool. The human keeps the final click for now, as every vendor's safeguards show.
• Blocking agents will cost enquiries. A site that blocks AI agents at the firewall, or hides its form behind a challenge, will lose the visits where the agent was doing the work. Our guide to whether to block AI agents sets out how to keep the protection without shutting out customers.

The risk is not that Gemini autofill passes your site by this month. It is a slow drift: the share of form fills done by software grows, and forms that software cannot read quietly convert less, with nothing in analytics to say why.

What to do next

1. Run can visitors contact you on your main enquiry page and fix any field without a visible, connected label.
2. Add autocomplete values to every field that asks for the visitor's own details: name, email, tel, organization, street-address and postal-code are the usual set.
3. Label the fields Gemini now targets in plain words. "Order number", "Booking reference", "Tracking number", with matching field names, on returns, support and warranty forms.
4. Remove autocomplete="off" from enquiry forms unless a field genuinely must not be filled, such as a one-time code or a spam honeypot.
5. Run the agent readiness check to see whether an agent can read the page and reach the form without running scripts, then run Lighthouse's Agentic Browsing category in Chrome 150 or later.
6. Test the form the way visitors now use it. Fill it with Chrome autofill on a phone. Count the fields you still had to type. Each one is a place an enquiry can stall.

Frequently asked questions

What is "Suggested by Gemini" in Chrome?

It is the label Chrome shows on enhanced autofill suggestions that come from Gemini with Personal Intelligence rather than from saved autofill data. They can include order numbers, tracking information and flight bookings found in the person's Gmail. It began reaching US users on 28 October 2026 for people who have switched Personal Intelligence on.

Can I stop Chrome's Gemini autofill on my website?

Not completely, and you should not try. Visitors control enhanced autofill in their own Chrome settings. A site can set autocomplete="off" on a field, but that only stops ordinary autofill helping on fields that should never be filled, and browsers may ignore it on login fields. Blocking help on an enquiry form costs completions.

Does the autocomplete attribute help SEO or AI search?

Not directly. Autocomplete helps a browser or agent fill a form once someone is on the page. It does not help a page rank or get cited. It does support accessibility under WCAG 2.2 criterion 1.3.5, and Lighthouse's experimental agent audits check that form fields have names.

Is Chrome's Gemini autofill available in Australia?

Not yet. Google's announcement names only the US, and gives no date for other countries. Ordinary Chrome autofill already helps Australian visitors, so the same form fixes pay off for Australian visitors today, whatever the timing of the Gemini layer.