Is your website agent-ready? A practical checklist

Your next visitor might not be a person. AI agents can now open a browser, read your pages, compare you with competitors and report back, or even fill in a form, on someone's behalf. OpenAI's new dots, for example, each come with their own cloud computer and browser, and OpenAI's Atlas browser has an agent mode built in. The person still makes the decision. But the agent decides what they see first.

The good news: an agent-ready website is mostly just a well-built website. Here is a practical checklist.

1. Serve real content in the first response

Many modern sites send a nearly empty page and build it in the browser with JavaScript. People with fast phones rarely notice. Automated readers often do. Google notes plainly that not all bots can run JavaScript, and it still recommends server-side rendering or pre-rendering.

Check: View the raw HTML of your key pages (or ask your developer to). Are your headline, prices, service list and contact details in it? If the page only fills in after scripts run, some agents and crawlers may see a blank page.

Also check speed. Agents work on timeouts, and so do people. Fast hosting, sensible caching and compressed images help both.

2. Make prices and key facts explicit

Agents are often sent to answer simple questions: "How much does it cost?" "Do they serve my area?" "Are they open Saturday?" If the answer is buried in a PDF, an image or "contact us for a quote," the agent may report that it could not find it, and move on to a competitor who said it clearly.

• State prices, or at least starting prices or ranges, in plain text.
• Keep hours, locations, service areas, delivery times and return policies on clearly labeled pages.
• Use the same facts everywhere. Conflicting prices on two pages confuse people and machines alike.

3. Add structured data that matches the page

Structured data is a small block of code, usually JSON-LD using the schema.org vocabulary, that labels what is on a page: this is a product, this is its price, these are our opening hours. Google recommends JSON-LD where your setup allows it.

One rule matters more than the rest: Google's guidelines say not to mark up information that is not visible to users, even if it is accurate. Structured data should describe the page, not add hidden claims.

Be realistic about what it does. Google also says there is no special schema you need to add to appear in its AI features. Structured data makes your facts easier to read reliably; it is not a ranking shortcut.

4. Build forms and buttons that explain themselves

When an agent operates a website, it has to work out what each button and field does. OpenAI has said website owners can add ARIA labels to improve how ChatGPT's agent works on their sites in Atlas. ARIA is part of the same accessibility toolkit that screen readers rely on.

In practice, that means ordinary accessibility work:

• Use real buttons and links, not clickable boxes.
• Give every form field a visible, specific label ("Business email," not "Field 3").
• Show clear error messages next to the field that caused them.
• Avoid flows that only work with a mouse hover or a drag gesture.

This helps agents, and it helps your human visitors who use screen readers, keyboards or simply a small phone screen.

5. Keep checkout and booking clean

If you sell online, the fewer surprises the better: a clear cart, visible totals including shipping and tax before the last step, guest checkout where it makes sense, and no pop-ups that block the page.

Agent payments are still early. OpenAI's help documentation for ChatGPT Work's cloud browser says that at launch it cannot sign in to websites or complete payments. Dots can use saved passwords for sign-in, with your confirmation for new sign-ins. Protocols such as the Agentic Commerce Protocol, an open standard developed by Stripe and OpenAI, aim to let businesses accept orders through AI agents while remaining the merchant of record. For most small businesses, the practical step today is a clean, standard checkout, not a rush to adopt a new protocol.

6. Keep URLs stable

Agents, AI answers and search engines remember links. If you move a page, set up a permanent redirect from the old address. Avoid putting important content behind URLs that change every session. A broken link in an AI answer is a lost customer.

7. Decide who gets in, on purpose

Robots.txt lets you tell well-behaved crawlers what they may access, and the big AI companies now run several different bots. OpenAI, for example, documents:

• OAI-SearchBot, for ChatGPT search. Sites that block it will not be shown in ChatGPT search answers.
• GPTBot, which crawls content that may be used to train models. Blocking it is a separate choice.
• ChatGPT-User, for visits a person asks for in a chat. OpenAI notes that robots.txt rules may not apply to these user-initiated visits.

Make a deliberate decision for each bot instead of copying a blanket block or allow list.

There is also a newer way to recognize genuine agents. OpenAI's cloud browser signs its requests using a web standard for HTTP message signatures (RFC 9421), and major providers such as Cloudflare, Akamai and Vercel can verify them. If your firewall or bot protection blocks everything automated, check whether it can tell verified agents apart from scrapers.

8. Treat llms.txt as optional

You may have heard that every site needs an llms.txt file. It is a proposal, published in 2024, for a Markdown file that gives AI tools a short guide to a site. It is not a formal standard. Google says you do not need to create AI text files or other new machine-readable files to appear in its AI features.

Our advice: add one only if you have something genuinely useful to put in it, such as documentation or a product catalog that benefits from a clean map. Never use it to say anything that differs from your real pages.

9. Don't trade security for access

Agent traffic is still traffic. Keep the basics in place:

• Rate limits and bot protection on login, contact and checkout forms.
• Spam protection that does not rely only on puzzles that also block assistive tech.
• No sensitive data in URLs.
• A patched, monitored server, because a site that gets more automated visitors also gets more attackers testing it.

What this means for you

You don't need a separate "AI website." You need a fast, clear, accessible site that states its facts plainly and runs on well-maintained hosting. A quick self-audit:

• Do my key pages show content without JavaScript?
• Are my prices and policies written in plain text, and consistent?
• Does my structured data match what's on the page?
• Can someone complete my main form with only a keyboard?
• Have I decided which AI bots to allow?

If you answered "not sure" to more than one, that is a good place to start. SEVA Systems builds fast, server-rendered websites and runs them on managed hosting designed for exactly this. Contact us for a review of your site.

Sources

• https://openai.com/index/introducing-dots/
• https://openai.com/index/introducing-chatgpt-atlas/
• https://developers.openai.com/api/docs/bots
• https://help.openai.com/en/articles/11845367-chatgpt-agent-allowlisting
• https://learn.chatgpt.com/docs/dots
• https://www.agenticcommerce.dev/
• https://llmstxt.org/
• https://developers.google.com/search/docs/crawling-indexing/javascript/javascript-seo-basics
• https://developers.google.com/search/docs/appearance/structured-data/intro-structured-data
• https://developers.google.com/search/docs/appearance/ai-features