Is your website ready for AI agents? A practical checklist
Use this checklist to inspect an existing website. Pick a service and a realistic customer task, then follow the path from finding information to sending an inquiry. Record what happened and what needs correction.
For background on the stages involved, read what websites need to do for agents.
1. Check the business facts
Find the business name, contact details, services, and service area. Compare the home page, relevant service page, footer, and structured data.
A good result is consistent information with enough detail to answer the task. Fix conflicting phone numbers, old service descriptions, and vague geographic claims. Check exclusions and pricing conditions that affect the customer’s request.
2. Reach the page through normal navigation
Start at the home page and find the service. Follow links using their labels. Check the page on a narrow screen and with keyboard navigation.
A good result is a public URL with readable content and a clear next step. Fix broken links, hidden essential information, and controls whose purpose is unclear without a screenshot.
3. Inspect crawlability and page structure
Check that the public page returns successfully and that intentional indexing settings match your plan. Review robots.txt, the sitemap, page titles, canonical URLs, and headings.
A good result is a consistent description of the published pages and their preferred addresses. Fix accidental noindex settings, stale sitemap entries, and unexpected redirects. Investigate access blocks that prevent the clients you intend to serve from reaching the content.
4. Compare structured data with visible content
Inspect the business and page metadata. Check names, addresses, service details, and page identity against what the visitor sees.
A good result is an accurate structured representation of the actual content. Fix invented or outdated values and unrelated markup. A validator can help identify syntax errors, but it cannot decide whether a business claim is true.
5. Read the alternative formats
If the site offers Markdown, request it and compare its answers with the HTML page. Check any llms.txt directory or published tool description for current URLs.
A good result preserves important details and links. Fix missing qualifications, outdated content, and representations that disagree about the service.
6. Test the agent connection
If the site exposes MCP or WebMCP, use the intended client to inspect available tools. Test the documented connection path instead of assuming the agent will discover everything automatically.
A good result lets the agent identify the intended action and understand its inputs. Fix unclear descriptions, missing required-field explanations, and tools that advertise actions the site cannot complete.
7. Submit a test inquiry
Use clearly labeled test data and authorize the submission. Check the result reported to the sender, the saved entry, and delivery to the business.
A good result distinguishes an accepted inquiry from a confirmed booking or completed purchase. Test a missing field and a repeated request. Fix misleading confirmations, duplicate entries, and failures that leave the sender unsure what happened.
8. Keep a record of what you tested
Record the date, URLs, client, browser, and result. Prioritize failures that prevent a useful answer or a successful inquiry. Repeat the relevant checks after changing forms, domains, languages, or important business information.
Wirechunk includes a Readiness view and public agent interfaces. Learn how to use them.