Skip to content
Chalk Labs / AI SEO
How-to guide

How to Prepare Your Website for AI Agents

By Rahil Jain, founder of Chalk LabsUpdated How we research

The short answer

Make your site easy for a machine to read and operate: use real buttons and labelled form fields, keep layouts stable, publish accurate product and price data that matches the page, and state your policies in plain text. Google says browser agents read screenshots, page structure and accessibility information, so good accessibility and clean markup are the base layer 12.

  • Browser agents use screenshots, page structure and accessibility information, so accessible HTML helps them 12.
  • Agentic checkout standards exist, but Google's UCP is waitlist-based and OpenAI's ACP is in beta 34.
  • Only 9% of B2B buyers in one survey would let agents execute purchases within approved guardrails 5.
  • Structured data must describe what is visible on the page, or it breaks Google's rules 8.

What does it mean to make a site ready for AI agents?

It means a software agent can find your information, understand your offer and complete a task on your behalf. The task might be to compare prices, book a slot, fill an enquiry form or buy a product. An agent is a program that acts for a person, not just one that answers a question.

Google describes AI agents as autonomous systems that can perform tasks on behalf of people, and says browser agents may visit sites to collect data such as screenshots, page structure and accessibility information 1. That description is the practical guide: if a page is clear to a screen reader and a screenshot, it is clear to an agent.

Two things get called agentic. One is browser agents that operate ordinary pages. The other is commerce protocols that let an agent check out through a merchant's feed and payment system. This guide covers both, in the order most owners should do them. Our agentic search definition gives the background.

Do it step by step

0 of 8 done0%

  1. Choose the actions you most want an agent to complete: request a quote, book a slot, compare prices or buy. Write each as a one-line task.

  2. Complete each task with the keyboard alone. Note every place you get stuck, such as traps, hidden fields or controls that do not respond.

    Tip: If a task needs a mouse hover, redesign it.

  3. Swap clickable div and span tags for button and a elements. Add label for attributes to every form field. Remove invisible overlays 2.

  4. Keep key buttons in the same place across pages, make them clearly visible and avoid layout shifts that move them after load.

  5. Add prices or ranges, inclusions, service areas, delivery, returns and booking steps on stable URLs. Link them from the footer and the relevant pages.

  6. Add Product, Organization or LocalBusiness markup that matches visible content exactly 8. Validate it before release.

  7. If you sell products, make the feed accurate and complete, with the same prices, stock and policies as the site. Join a commerce programme such as UCP only if it fits your business 3.

  8. Check robots.txt, firewall and bot-challenge settings. Then ask an agent-capable assistant to do your three tasks and log the result.

Your ticks are saved in this browser only.

Copy-ready templates

Accessible enquiry form (agent and screen-reader friendly)

html
<form action="/enquiry" method="post">
  <label for="name">Your name</label>
  <input id="name" name="name" type="text" required autocomplete="name">

  <label for="email">Work email</label>
  <input id="email" name="email" type="email" required autocomplete="email">

  <label for="need">What do you need?</label>
  <textarea id="need" name="need" required></textarea>

  <button type="submit">Request a quote</button>
</form>

Native elements, a label tied to each field with for and id, and a button that says what it does .

Product JSON-LD that matches the visible page

json
{
  "@context": "https://schema.org",
  "@type": "Product",
  "name": "Example Standing Desk",
  "sku": "DESK-100",
  "description": "Height-adjustable desk, 140 cm wide, 5-year frame warranty.",
  "offers": {
    "@type": "Offer",
    "url": "https://example.com/products/standing-desk",
    "price": "449.00",
    "priceCurrency": "USD",
    "availability": "https://schema.org/InStock"
  }
}

Placeholder product. Every value must match the visible page . Add review and return-policy details only if you show them.

Agent-readiness checklist

markdown
- [ ] Top 3 tasks completable with keyboard only
- [ ] Real button/a elements; labels linked with for/id
- [ ] No overlays covering controls; key buttons stable
- [ ] Prices or ranges, returns, delivery, hours in plain text
- [ ] Structured data matches visible content
- [ ] Product feed matches site prices and stock
- [ ] Search bots allowed; private pages behind login
- [ ] Bot challenges do not block pricing, booking, contact
- [ ] Agent test logged this month

Re-run monthly, because agent tools change quickly.

What do you need, and how long does it take?

You need access to your site's templates, a developer for a day or two, and your product or service data in one clean place. The base layer takes one to two weeks for a small site. The commerce protocols take longer and are optional for most businesses.

Do not wait for a standard to settle. Most of the work below also helps people who use screen readers, and it makes your normal SEO stronger.

How do you make pages that an agent can operate?

Follow the web.dev guidance for agent-friendly sites. Use native elements such as button and a, not styled div or span tags, because agents recognise native elements as interactive. If you cannot, add the right role and tabindex. Link each label to its field using the for attribute so the agent knows what the field asks for 2.

Avoid transparent overlays that hide interactive elements, because an agent may discard covered nodes. Keep important controls larger than 8 square pixels. Keep layouts stable, because an agent working from screenshots can be confused when an Add to cart button moves between pages 2.

Test with the keyboard only. If you can complete your enquiry form with Tab, Enter and Space, an agent probably can. If a pop-up traps focus or a cookie banner covers the form, fix those first.

What information should an agent be able to read without clicking around?

Put the facts an agent needs in plain text on stable URLs: prices or price ranges, what is included, service areas, opening hours, delivery and return terms, contact details and how to book. An agent that cannot find your return window will skip you or guess.

Pair that text with structured data where it fits. For products, Google recommends providing as much rich product information as available 9. Google's structured data rules say not to mark up content that is not visible to readers, and that markup must be a true representation of the page 8. Never put a price in markup that differs from the price on the page. Use the schema generator for a first draft.

Do you need agentic commerce standards like UCP or ACP?

Only if you sell online and want assistants to complete purchases. Google's Universal Commerce Protocol is described as an open standard that lets merchants enable agentic actions in AI Mode and Gemini, starting with direct buying. Merchants can reuse existing Merchant Center feeds, the page lists a waitlist, and embedded checkout is limited to approved merchants 3.

OpenAI's Agentic Commerce Protocol is an open standard for connecting buyers, their AI agents and businesses. It is maintained by OpenAI and Stripe, licensed under Apache 2.0, and its README says it is in beta, with 2026-04-17 as the latest stable version 4.

Both are young and may change. If you run a shop, first fix your product feed and product pages, because both routes build on that data. Service businesses can skip the protocols and focus on booking and enquiry forms. Our guide to optimizing product pages for AI shopping covers the feed side.

Will buyers really let agents spend their money?

Not yet, on the evidence we have. G2's 2026 Buyer Behavior Report, a survey of 1,038 B2B decision-makers fielded in June 2026, found that over 60% currently use or plan to use AI agents in software buying. Only 9% are comfortable letting agents execute purchases within approved guardrails, and 2% would allow purchases without pre-approval 5.

These are stated intentions, not observed behaviour. Read them as: agents will research and shortlist before they buy. So the first job is to be easy to find and compare, with clear facts. The checkout job comes second.

How do you control which bots and agents can reach your site?

Know that agent fetches behave differently from search crawlers. OpenAI says ChatGPT-User handles actions triggered by a user, is not used to crawl the web automatically, and that robots.txt rules may not apply to it because the user started the action 6. Perplexity says its user-triggered fetcher generally ignores robots.txt for the same reason 7.

That means robots.txt is not a lock. If a page must be private, protect it with a login or server rules. For public pages you want found, allow the search bots and test with the AI crawler checker. Also avoid blocking agents with aggressive bot challenges on pages like pricing, booking and contact, unless you accept losing those visitors.

What mistakes should you avoid?

Do not hide prices behind a form if you can show a range. Do not use a chat widget as your only way to get a quote. Do not rely on images of text for key facts. Do not let markup drift from the page, and do not promise an agent-ready site to buyers before you have tested one.

Also do not chase new files. Google's AI guidance says llms.txt and special markup are not used by Google Search 10. Spend the time on pages, forms and data.

How do you test that an agent can use your site?

Ask a browser-based assistant to do your three most valuable tasks on your site, such as request a quote, book a call or find your return policy. Watch where it stalls. Run the same test monthly, because agent products change quickly.

Record the outcome in a sheet: task, tool, date, success, point of failure. Fix one failure at a time and repeat the test. We found no standard benchmark for agent readiness, so your own task list is the best test.

Frequently asked questions

What is agentic commerce?

It is shopping where an AI agent acts for the buyer, from comparing options to checkout. OpenAI's Agentic Commerce Protocol and Google's Universal Commerce Protocol are two attempts to standardise it. ACP is in beta and UCP access runs through a waitlist, so treat both as early 34.

Do I need an API for AI agents?

Not to start. Agents that use a browser work with ordinary pages, so clean forms and clear text come first. Google's UCP supports a REST API and an MCP binding for merchants who adopt it 3, but most small sites do not need one yet.

Does blocking bots in robots.txt stop AI agents?

Not reliably. OpenAI and Perplexity both say their user-triggered fetchers may not follow robots.txt, because a person started the request 67. Use a login or server-side rules for anything private.

Is this different from normal SEO?

Mostly it overlaps. Clear content, accurate data and accessible pages help search, screen-reader users and agents alike. Google also treats generative AI optimisation as an extension of SEO 10. The new part is testing tasks, not just rankings.

About the author

Rahil Jain

Rahil Jain is an AI SEO (answer engine and generative engine optimisation) consultant and co-founder of Chalk Labs.

LinkedIn · Method and credentials

Sources

  1. Optimizing your website for generative AI features on Google Search (Explore agentic experiences) · Google Search CentralOpened October 20, 2026. Supports: Defines AI agents as autonomous systems that can perform tasks on behalf of people; browser agents may collect screenshots, page structure and accessibility information; points to agent-friendly guidance. Last updated 2026-07-10.
  2. Build agent-friendly websites · web.dev (Google)Opened October 20, 2026. Supports: Use native elements, labels with for, avoid overlays, controls larger than 8 square pixels, stable layouts, WebMCP in development. Last updated 2026-04-01.
  3. Universal Commerce Protocol (UCP) for merchants · Google for DevelopersOpened October 20, 2026. Supports: UCP enables agentic actions in AI Mode and Gemini starting with direct buying; reuse Merchant Center feeds; waitlist; embedded checkout limited to approved merchants; REST API and MCP binding.
  4. Agentic Commerce Protocol · GitHub (agentic-commerce-protocol)Opened October 20, 2026. Supports: Open standard for connecting buyers, AI agents and businesses; maintained by OpenAI and Stripe; beta; Apache 2.0; latest version 2026-04-17.
  5. 2026 Buyer Behavior Report · G2Opened October 20, 2026. Supports: 1,038 B2B decision-makers, fielded June 2026: over 60% use or plan to use AI agents in software buying; 9% comfortable with agents executing purchases within guardrails; 2% without pre-approval. Stated intentions.
  6. Overview of OpenAI crawlers · OpenAIOpened October 20, 2026. Supports: ChatGPT-User handles user-initiated actions, is not used to crawl automatically, and robots.txt rules may not apply.
  7. Perplexity crawlers · PerplexityOpened October 20, 2026. Supports: Perplexity-User supports user actions and generally ignores robots.txt because a user requested the fetch; PerplexityBot surfaces sites in search.
  8. General structured data guidelines · Google Search CentralOpened October 20, 2026. Supports: Do not mark up content that is not visible; structured data must be a true representation of the page. Last updated 2026-07-10.
  9. Product structured data · Google Search CentralOpened October 20, 2026. Supports: Recommends providing as much rich product information as available. Last updated 2025-12-10.
  10. Optimizing your website for generative AI features on Google Search (myths) · Google Search CentralOpened October 20, 2026. Supports: Google Search does not use llms.txt or special AI markup; Google treats optimising for generative AI search as SEO.
How this page was made. Built on Rahil Jain's AI SEO method and drafted with AI research assistance. Every statistic links to the page we read it on (opened October 20, 2026 or nearby). Published October 20, 2026; last updated October 20, 2026. Chalk Labs sells AI SEO services, so we have a commercial interest in this topic. Spotted an error? Tell us and we will fix it.