Making your website agent-friendly

0:00 Making your website agent-friendly

Your next website visitor might not be a person. It might be an AI assistant comparing your prices, reading your return policy, or trying to book an appointment for someone. In this lesson you will learn the seven layers that make a site agent-friendly, and you'll see that most of them also improve accessibility, search and conversion.

0:25 Layers 1 and 2

Layer one is semantic, accessible HTML. Agents read the accessibility tree, so use real buttons, links, labels and selects. Give controls specific names, book a consultation, not click here. Put a visible label on every form field and make the whole task work with the keyboard. Follow WCAG two point two. Layer two is structured data. Use schema.org JSON-LD to say what things are: your organization, products with prices and availability, services, events, and policies. Keep it consistent with what's visible on the page.

1:02 Why it matters

Why does this matter? Because the assistants your customers use are increasingly reading your site on their behalf, answering questions about your prices, hours and policies, and sometimes filling your forms. If they get it wrong, customers blame you, not the assistant. The good news is that almost everything that makes a site agent-friendly also makes it more accessible, better for search and easier to convert. It's one of the rare investments that pays off three ways.

1:35 The personal shopper

Here's an analogy. Imagine your shop is visited by two customers at once. One is a regular shopper who looks at the window display. The other is a personal shopper sent by a busy client, with a list, a budget and ten other shops to visit today. The personal shopper wants clear price tags, readable signs and an easy checkout, and they'll leave quickly if they can't find what they need. AI agents are that personal shopper. Make your shop easy for them, and you'll usually make it easier for everyone.

2:15 Simple example: a bakery

A simple example. A small bakery's site shows its opening hours in a pretty image, and its menu as a PDF. Ask an assistant when does this bakery open on Friday, and it may guess or quote a directory with old hours. The fix takes an afternoon: put the hours in plain text on the page, add local business structured data with opening hours, and publish the menu as a simple web page. Now the answer is unambiguous, for assistants, search engines and customers on slow phones.

2:53 Layers 3 and 4

Layer three is clear, forgiving forms. Label every field, explain the format, accept sensible variations, and show errors as text next to the field. Decide which forms you want agents to complete, like inquiries and bookings, and which you don't, like mass account creation, and protect those with rate limits and abuse controls. Layer four is machine-readable facts. Prices, delivery areas, returns, warranties, hours and contact routes belong in text and structured data, not just in images or PDFs.

3:27 Layers 5 and 6

Layer five is optional: an llms.txt file. It's a community proposal from 2024, a Markdown file at your site root that gives language models a short overview and links to your most useful pages. It's cheap and harmless, but major AI search providers haven't committed to relying on it, so treat it as a helper, not a ranking trick. Layer six is APIs and protocol endpoints. For high-value actions like booking or quoting, a documented API or MCP server is far more reliable than an agent clicking your interface.

4:06 Layer 7: access policy

Layer seven is your access policy. Decide which AI crawlers and agents can reach what. Use robots.txt for the crawlers that honor it and bot management at your CDN for the rest. Remember that some assistants browse on behalf of a user rather than as a declared crawler. And be careful: blocking everything also blocks assistants that might have sent you customers. CAPTCHAs everywhere frustrate humans too. Risk-based controls are the better balance.

4:38 Worked example: clinic booking

A dental clinic in Riyadh discovered that assistants were getting its hours wrong and agents failed to book. The hours were in an image. The date picker had no labels. And the Arabic and English pages showed different phone numbers. The fixes were straightforward: local business structured data with opening hours, an accessible date input, consistent contact details in both languages, and a simple booking-request form. A sandboxed agent then completed booking requests reliably, and the front desk got fewer calls asking about hours.

5:15 The audit

Here's your audit, from the lesson text. On your top five conversion pages, check accessible names, form labels, a keyboard-only path, valid structured data that matches the page, policies in text, and whether a sandboxed browser agent can complete the key task on a test account. Measure the checklist pass rate, agent task completion in tests, structured-data errors in Search Console, and, where your analytics can identify them, visits and conversions from AI assistants.

5:47 Structured data done right

Let's look at structured data done right. On a product page, you mark up the product's name, SKU and brand, and an offer with the price, the currency and availability, plus your return policy. The key rule is consistency. If the page says three thousand four hundred and ninety-nine rupees and in stock, the structured data must say exactly the same. Validate it with Google's Rich Results Test or the Schema.org validator every time templates change.

6:20 Test like an agent

Now test it the way agents will. In a sandbox, on a test account, give a browser agent a realistic task on your site: find the returns window for this product, or request a consultation for next Tuesday. Watch where it hesitates, misclicks or gives up. Every stumble is a usability issue for humans too. Repeat after each fix, and add these tasks to your regular QA so a redesign doesn't quietly make your site unreadable to assistants.

6:54 Three mistakes

Three common mistakes. First, structured data that contradicts the visible page, like an old price in the markup, which confuses agents and can be treated as misleading by search engines. Second, hiding key facts in images, PDFs or chat widgets. Third, putting CAPTCHAs everywhere, which blocks legitimate assistants and frustrates real customers, when risk-based bot management would protect you better.

7:20 Try this now

Try this now. Take your five most important conversion pages. For each, run the accessibility snapshot script, check the structured data with a validator, and confirm that prices, hours and policies appear as text. Then ask a browser agent, in a sandbox and with a test account, to complete the page's main task, like requesting a quote. Note every hesitation or failure. Fix the two issues that affect the most pages, and re-test.

7:52 Recap

Recap. Agent-friendly means accessible HTML, accurate structured data, clear forms, facts in text, an optional llms.txt, APIs for high-value actions, and a deliberate access policy. Most of it helps humans and search too. Your next step: run the audit on your top five pages and fix the two biggest failures. Next module: real use cases in marketing operations, QA and data entry.