An agentic browser is a web browser with an AI agent built into it that can carry out tasks on the user’s behalf: reading pages, clicking through menus, filling in forms and, with permission, completing a booking or a purchase. Instead of showing a website to a person, it uses the website for them and reports back.
How an agentic browser works
The user types an instruction in plain language, such as “find a plumber in Croydon who can come on Saturday and ask for a quote”. The browser’s AI agent breaks that into steps, searches, opens pages, reads the content and the page structure, and interacts with buttons and fields much as a person would. Most products pause for confirmation before anything irreversible, such as paying or sending a message.
To understand a page, the agent draws on what is rendered on screen and on the underlying structure: headings, link text, form labels and the accessibility tree that screen readers also use. Pages built with clear semantic HTML are far easier for it to operate than pages assembled from unlabelled custom elements.
At the time of writing (October 2026), several products work this way, including Perplexity’s Comet and OpenAI’s ChatGPT Atlas, and the major browser makers are adding agent features to Chrome and Edge. Capabilities, regional availability and permissions change quickly, so check the current position before planning around any one of them.
Why it matters
For a UK business, the practical question is whether an agent acting for a customer can complete your most valuable journey. If a booking form needs a custom date picker that only responds to a mouse, or your prices appear only inside an image, the agent may give up and move to a competitor whose site it can read. The customer may never know your site was considered.
It also changes measurement. Agent visits can look like ordinary sessions, arrive with unusual timing, or fill in forms in seconds. Some will be filtered as bot traffic and some will not, so conversion rates and engagement figures may drift without any change in real demand.
Much of the work overlaps with obligations you already have. A site that meets accessibility standards under the Equality Act 2010, with labelled fields and keyboard-operable controls, is usually one an agent can use too.
Common mistakes
- Treating agentic browsing as a distant trend and never testing a single journey with one.
- Placing key facts, such as delivery charges, service areas or opening hours, in images, PDFs or pop-ups the agent cannot reliably read.
- Forms with placeholder text instead of proper labels, or error messages that appear only as a red outline.
- Blocking all automated traffic at the firewall without checking whether legitimate agents acting for customers are caught too.
- Assuming a fall in on-site engagement means less interest, when agents may be completing tasks faster than people do.
How to act on it
Pick your two or three most valuable journeys, such as requesting a quote, booking an appointment or buying a best-selling product, and try each one with an agentic browser. Note where it stalls. The fixes are usually ordinary good practice: labelled form fields, real buttons, plain-text prices and availability, and a page structure where the main content is easy to find.
Then look at your server logs and analytics for agent traffic so you can separate it in reports. Keep structured data accurate and consistent with what the page shows. If you would like this reviewed alongside how your business appears in AI answers more broadly, it forms part of my AI search optimisation service.