How to brief your agency on agent readiness

"Make the site work for AI agents" isn't a brief an agency can price or deliver. This guide shows how to turn it into one: what's in scope, which checks to name, how you'll accept the work, what to ask for in reporting, and a template you can copy.

Why agent readiness needs a proper brief

Agent readiness touches several kinds of work at once: templates and components, structured data, robots.txt, bot protection settings and checkout. Those often sit with different people, and some sit with vendors your agency can't change. Our post on who owns agent readiness covers how it falls between teams.

A good brief does three things. It says exactly which pages and settings are in scope. It names the outcomes in terms anyone can check, rather than "best practice". And it says how progress will be reported, so you can tell a finished fix from a promised one.

Hard to deliver

• "Make the site AI-ready"
• "Improve our AgentScore"
• "Follow best practice for agents"

Easy to accept

• "Product pages pass the product data and prices checks"
• "No new failures on any other check"
• "A Ghost Agent reaches checkout from a product page"

Set the scope

List the pages and journeys that make you money, and the parts of the site the agency is allowed to change. A typical online store's list:

• Templates: home page, category or collection pages, product pages, cart and checkout (where your platform lets you change it).
• Site files: robots.txt, the sitemap and llms.txt.
• Shared components: header, navigation, search, pop-ups and cookie banners, variant pickers and add-to-cart buttons.
• Settings: guest checkout, CAPTCHA settings and anything in your platform's admin.

Then say what's out of scope, and who owns it instead. Bot protection and CDN rules often belong to an IT or security team, or to the CDN vendor itself. Name them in the brief, so the agency knows who to raise a finding with rather than parking it. Our questions for your CDN or bot protection vendor help with that side.

• Agency Templates, components, structured data
• Content or SEO team Titles, descriptions, alt text, llms.txt
• Security or CDN admin Bot protection, challenges, rate limits
• Platform Defaults you can only ask about

Name the checks

Run AgentScore before you write the brief and attach the report. It scores the site out of 100 across three categories, with a written fix for each problem, and every check can be named in the brief by its title. Reading your AgentScore report explains what to fix first.

Area	Checks to name (examples)	Usually fixed by
Access	"robots.txt lets AI assistants and search agents in", "Shoppers can check out without an account"	SEO team, platform admin
Access (protection)	"Bot protection lets AI agents through", "AI agents aren't blocked or shown a CAPTCHA at the cart and checkout"	CDN or security admin
Readability	"Product pages give price and stock in a form agents can read", "Prices are in the page HTML", "Content loads without JavaScript"	Developers
Navigability	"Buttons and links have names", "Pop-ups and banners can be dismissed by agents", "Agents can find and press your add-to-cart or sign-up button", "Cart and checkout fields are labelled for agents"	Developers

Many of these overlap with accessibility work, so if the agency already does accessibility audits, ask them to include them. See accessibility work is agent readiness work.

The free scan tests task completion with a single task at most, like adding a product to the cart, and only when its AI agent runs. It doesn't show whether an agent can finish a purchase. For that, name the journeys you care about and test them with Ghost Agents, which send real AI agents through a journey, stop before payment and record every step. See how to test your key journeys with AI agents.

Write acceptance criteria

Acceptance criteria are what you'll check before you sign off. Keep them few, specific and checkable by someone who isn't a developer.

• Named checks pass Each check in scope passes in a scan of the live site.
• No regressions No check that passed in the baseline scan now warns or fails.
• Journeys pass Named Ghost Agent journeys pass in every run, not just some.
• "Not tested" explained Any skipped check comes with the scan's reason and a manual test.

Some things to know when you set them:

• Prefer checks to a target score. A score blends many checks, and checks that don't apply to your site are left out. "These six checks pass" is clearer than "score of 80".
• Expect some "not tested" results. AgentScore skips a check rather than guess, for example when your bot protection verifies where crawlers really come from (the scanner's requests come from our servers, not the AI companies'), or when it can't add an item to the cart. A skip doesn't count against you, but it does mean someone should test that part another way.
• Ask for consistent journeys. Ghost Agents run each check several times, because AI agents vary. "Flaky" means some runs passed and some failed. Accept a journey when it passes every run, not once.

Ask for useful reporting

Ask for reporting that shows evidence, not activity:

• A baseline and a final scan report, with the date of each, so changes can be compared check by check.
• A change list: each fix, the pages and components it touched, and the check or journey it was for.
• What couldn't be fixed and why, with the owner it was handed to (platform, CDN vendor, security team).
• Journey replays for any Ghost Agent test that failed, showing the step where the agent got stuck.
• A note on what to re-check after releases, such as theme updates, new apps or new pop-ups.

If you use Ghost Agent Lab, you can give an agency the Viewer role so they see results without changing anything (or Member, if they'll update fixes themselves), export the readiness findings as CSV for their ticketing tool, and track each fix on the Fixes page, where a fix marked shipped is re-tested straight away. For the monthly view, see agent readiness KPIs.

1. Week 1 Baseline Scan, journeys, brief agreed
2. Weeks 2–4 Fixes Change list updated as work ships
3. Week 5 Acceptance Final scan and journeys against the criteria
4. After Upkeep Re-check after each release

A template brief

Copy this, fill in the brackets and attach your baseline scan report. For a longer plan around it, see a 90-day agent readiness plan.

Project: Agent readiness for [northwind.example]

Goal
AI agents (assistants such as ChatGPT, Claude and Perplexity, and
browser agents) can reach, read and use our product pages, cart
and checkout, so they can recommend us and help shoppers buy.

In scope
- Templates: home, collection, product, cart, checkout (where editable)
- Files: robots.txt, sitemap, llms.txt
- Components: header, search, pop-ups, cookie banner,
  variant picker, add-to-cart button
- Settings: guest checkout, CAPTCHA settings

Out of scope (owner)
- CDN and bot protection rules ([security team / vendor])
- Payment provider settings ([finance])

Baseline
- AgentScore report dated [date], attached
- Ghost Agent journeys: [find a product and reach checkout],
  [use search to find a size 10 running shoe]

Acceptance criteria
1. These checks pass on the live site: [list check titles]
2. No check that passed in the baseline warns or fails
3. Each named journey passes in every run
4. Any "not tested" check is explained and tested by hand

Reporting
- Change list: fix, pages touched, check or journey it serves
- Blockers handed to other owners, with dates
- Final scan report and journey results
- What to re-check after future releases

Contacts
- Our owner: [name]  Security/CDN: [name]  Platform: [name]