AI browsers are here. What they mean for your website
The next visitor to your website may not be a person. It may be a browser acting for one: reading your pages, comparing your prices with a competitor's, filling in your enquiry form, and reporting back. That is no longer a prototype. Perplexity launched its Comet browser in 2025, OpenAI followed with ChatGPT Atlas, Microsoft added Copilot Mode to Edge, and Google has been building Gemini's agent features into Chrome.
What an AI browser does differently
A search engine reads your page once and files it. An AI browser uses your page, live, for a specific person with a specific task: “find me a web designer who can start this month and book a call”. It reads the text, works out which buttons do what, and clicks them.
That makes some old problems newly expensive. A human visitor who cannot find your prices might scroll, squint and email you anyway. An agent that cannot find them moves on to the next site in its list, and the person never learns you were considered.
The surprising shortcut: accessibility
Here is the useful insight for anyone who has put off accessibility work. An AI agent navigating a web page relies on much of the same structure a screen reader does: real buttons that are buttons, form fields with proper labels, headings in a sensible order, links that say where they go, text that is actually text rather than a picture of words.
A site built to be accessible is, almost by accident, a site an agent can use. The work you might have done for the European Accessibility Act now pays twice.
Five things that make a site agent-friendly
- Say the important facts in plain text. Prices, opening hours, service area, lead times, what is included. Not in an image, not in a PDF, not behind a “contact us for pricing”.
- Label every form field properly. “Name”, “Email”, “What do you need?” — attached to their inputs, not floating nearby.
- Make errors explicit. “Please enter a valid email address” can be read and fixed. A red outline cannot.
- Keep the path short. Every extra step, pop-up and interstitial is a place an agent can stall.
- Use structured data where it is genuinely true — your business details, products, prices, FAQs. It is the machine-readable summary of what the page already says.
A tension worth knowing about
Bot protection and AI agents pull in opposite directions. The CAPTCHAs and bot filters that keep spam out of your forms can also stop a legitimate agent acting for a real customer. There is no perfect answer yet. For most small sites, a quiet honeypot field and server-side checks stop spam without putting a puzzle in front of every visitor, human or not.
And a warning for your own use
If you use an AI browser yourself, be careful what you let it do while logged in. A page can contain hidden instructions aimed at the agent rather than at you — the attack is called prompt injection — and the companies building these browsers have said openly that it is not a solved problem. Keep agent browsing away from your bank, your email admin and anything that moves money until that changes.
What has not changed
An agent can shortlist you, but a person still decides. The same things that win a human visitor — clear prices, real photographs, evidence you are a real business — are the things an agent reports back. Getting the basics right is still most of the job; five website mistakes is a good place to start.
Every site we build is designed to WCAG 2.2 AA with semantic, labelled markup — which is exactly what an agent needs. See how we build, or book a free call to check yours.
Sources
Researched, written and published by Aftab Ahmad, owner of Seven Layers.