# ChatGPT Can Now Fill In Forms From Your Open Chrome Tabs. Check This First.

Published 19 October 2026 by Prad Patel. Tags: AI Tools, Small Business.

Page: https://kaizenaiconsulting.com/chatgpt-chrome-tabs-fill-forms-check-first/

ChatGPT can now read your open Chrome tabs, and in its Work mode it can fill in forms, send messages and download files for you. That is the headline from OpenAI's 15 October update to the ChatGPT Chrome extension. We have not run it on a client's account yet, so this is not a review. It is what we would tell a client to check before they switch it on.

## What actually changed

The extension used to be a side panel. Now, according to [PPC Land's write-up](https://ppc.land/chatgpt-gains-control-of-chrome-tabs-to-fill-forms-and-send-messages/), you can mention Chrome inside the ChatGPT web or desktop app and it pulls in whatever is open in your tabs. In ordinary Chat mode that means comparing products, summarising research or pulling details out of several pages at once.

The bigger change is in Work, OpenAI's agent for longer jobs. With your permission it can take control of tabs to send messages, fill out forms and download files. OpenAI's own example is reading an event description on one page and using it to complete a registration form on another, then stopping for you to review. It can also group and tidy your tabs, which is the least risky thing on the list.

Coverage from [EM360Tech](https://em360tech.com/tech-articles/openai-lets-chatgpt-read-open-chrome-tabs-and-act-them-work-mode) and others says it works in Chrome and several Chromium browsers, though Opera lacks the side chat. Which ChatGPT plans get the tab-control part is unclear in the coverage, so check your own plan before planning around it.

## Why a small business would care

Admin is where the hours go. The American Express and Small Business Saturday UK SME Business Barometer, published in July 2026, found owners spend about 11 hours a week on administrative or finance tasks, as [Retail Times reported](https://retailtimes.co.uk/amex-sme-barometer-small-business-owners-spend-nearly-twice-the-time-on-admin-than-growing-their-business/). That is nearly twice the time they spend on sales and growing the business. A lot of that is the same boring pattern: copy details from one page, paste them into another, click submit.

That pattern is exactly what a tab-controlling assistant is built for. Supplier portals, booking forms, tender registrations, expense claims, event sign-ups. If you can describe the job in a sentence, the pitch is that ChatGPT will do the clicking.

## The catch is where it runs

The extension works inside your normal, signed-in browser profile. So it can read or act on any site you are already logged into, whether that is your email, your accounting software or your bank. Several reports make the same point: giving an assistant access to your tabs means giving it access to everything in them, including billing pages, open emails and internal documents.

The install prompt is broad too. The permissions can include reading and changing data on all websites, and OpenAI says it only stores browser activity when it becomes part of the ChatGPT context. The default, according to the same coverage, is that ChatGPT asks before it interacts with a new website. That is sensible. But the reporting we found does not make clear whether it asks again before every form submission, so do not assume it does.

Then there is prompt injection. A web page can carry hidden text written to look like an instruction, and an agent that reads the page can mistake it for one. The risk is higher when the agent sits inside a logged-in session, because there is more to steal and more it can do. Most of the security discussion we found comes from general commentary rather than OpenAI's documentation, so treat it as a reason for care, not a proven attack on this product.

## What we would tell a client to do first

### 1. Use a clean browser profile

Make a separate Chrome profile just for the assistant, signed in only to the sites the job needs. Keep your email, banking and accounting tabs in a different profile. This is the single cheapest control, and it shrinks what the assistant can see.

### 2. Allow one site, once

If you are offered a choice between allowing a site once, for that site, or for all sites, pick once. Widen it later when you have seen how it behaves, not before.

### 3. Start with a low-stakes form

Pick a job where a mistake costs you a minute, such as a newsletter sign-up or an internal booking form. Do not start with a payment, a tax filing or anything that goes to a customer.

### 4. Read it, then submit yourself

OpenAI says Work stops for review at the end of a task. Use that. Read every field before anything is sent, and keep the final click as your own. A form filled in wrongly at speed is still wrong.

### 5. Write down what it is allowed to touch

One line is enough: which profile, which sites, which kinds of job. If a member of staff uses it, they should know the same line. It is the cheapest policy you will ever write.

## Is it worth trying?

For the right jobs, yes. Repetitive form-filling is real time, and 11 hours a week of admin is a lot to give away. But this is a tool that acts as you, inside accounts that matter, and the sensible way in is small: one profile, one site, one low-stakes job, a human on the final click.

We have not tested it on a client's setup, so we are not claiming it saves any particular amount of time. What we can say is that the safeguards above cost nothing and apply to any browser agent, not just this one.

If you want help working out which of your admin jobs are safe to hand to an assistant and which are not, we run a free 30 minute AI audit to go through it with you.

**Prad Patel** ran Glass Tailors for twelve years, growing it to 12 staff and around £1m turnover. He now builds AI systems, quote tools and marketing for owner-run businesses, and shows them how they work. [About Prad](https://kaizenaiconsulting.com/about/)