Something quietly significant happened this month: Cloudflare put an "Agent Readiness" score inside every website dashboard it serves. If your site runs behind Cloudflare, there is now a screen telling you, pass by fail, whether an AI agent can discover you, parse you, and transact with you. Roughly a fifth of the web just got a report card for a shopping channel most retailers haven't started thinking about.
I went through the whole checklist against how we score retailers at BlynkAudit. Two things stood out. First, most of what it checks is open standards, not Cloudflare products. You can ship the majority of it in a week, on any stack. Second, for retail specifically, the checklist stops exactly where the money starts. More on that below.
Here are the quick wins, in the order I would do them.
1. Write your robots.txt for agents, not just Google
Cloudflare's own data says 78% of top sites have a robots.txt, and almost all of them are tuned for search engine crawlers only. AI agents arrive as GPTBot, ClaudeBot, PerplexityBot and friends, and when your file says nothing about them, you are leaving the decision of whether you're open for agent business to defaults you don't control. Name the AI crawlers explicitly and state your policy. Whatever your policy is, make it a decision rather than an accident.
2. Make sure the map matches the shop
An XML sitemap is twenty-year-old technology and agents lean on it harder than search engines ever did, because an agent on a task budget doesn't crawl speculatively. If your sitemap is stale, missing product URLs, or unreferenced from robots.txt, an agent enumerates a shop that doesn't match the one your customers see.
3. Offer a version without the wrapping paper
The newest quick win, and the one with the biggest measured payoff: serve clean markdown when a client asks for it. Cloudflare reports up to an 80% token reduction when agents read markdown instead of parsing your HTML, which translates directly into how much of your catalogue an agent can afford to read before its budget runs out. An llms.txt file pointing agents at your important pages does a lighter version of the same job. Every byte of template chrome an agent has to parse is money it doesn't spend understanding your products.
4. Structured data: the retail win the checklists skip
Here is the gap that surprised me: agent readiness checklists, Cloudflare's included, don't score structured data at all. For retail this is the whole game. GTINs, prices, availability, returns policy, shipping details, all marked up in schema.org, are the difference between a site an agent can read and a site an agent can buy from with confidence. In our weekly index data, structured data depth separates leaders from the pack more reliably than any protocol adoption. If you do one thing on this list, do this one.
5. Don't slam the door you're trying to open
The most expensive own goal we measure: aggressive bot protection that turns away every automated visitor, including the ones carrying a customer's mandate. One of Australia's largest furniture retailers currently blocks our shopping agent at the front door. Its agent-readiness score isn't low, it is unmeasurable, and an agent shortlisting on a customer's behalf treats an unenterable shop as a shop that doesn't exist. Review your bot rules and decide, deliberately, which agents get in. Blocking scrapers is legitimate. Blocking your next sales channel by default is not a security posture, it is a closed sign.
The groundwork tier
Past the quick wins sits real engineering: machine-readable API catalogues, documented authentication paths for agents, OAuth discovery, MCP endpoints that let an agent transact rather than browse. This is where agentic commerce is heading, and it is worth a line on your 2027 roadmap. But be honest about maturity: much of the "advanced" tier across the industry is unratified drafts and vendor proposals. Adopt what has settled, watch the rest.
The thing no checklist can prove
A checklist tells you the door has the right signs on it. It cannot tell you what happens when a shopper walks through: whether an agent can actually find the product, read the price, fill the cart, and reach checkout without a human rescuing it. Every week we send a real AI shopping agent through real retail funnels, and the pattern is consistent: sites that pass the plumbing checks still fail the shopping trip, and they have no dashboard that would ever tell them.
The quick wins above are worth a week of anyone's time. Just don't mistake an open door for a sale.