Most brands don't have Amazon's problem. While a few large marketplaces deliberately wall off AI shopping agents, the typical brand has the opposite problem, it blocks agents by accident and never knows. To stay visible, do the reverse of a walled garden: keep the shopping agents reachable, keep your product facts legible, and keep checkout completable.
This is the action companion to the explainer on why retailers block AI shopping agents. Below is a 6-step checklist, ordered from highest to lowest return.
Step 1: Don't disallow the shopping crawlers in robots.txt
Check your robots.txt and confirm you are not disallowing the agents that power AI shopping. The live-answer and search fetchers are the ones that matter for being cited and recommended right now:
| Agent | Company | Blocking it costs you |
|---|---|---|
OAI-SearchBot |
OpenAI | ChatGPT search and shopping placements |
ChatGPT-User |
OpenAI | "Find me the best X" live fetches |
PerplexityBot |
Perplexity | Perplexity shopping answers |
ClaudeBot |
Anthropic | Claude citations |
Google-Extended |
Google's AI surfaces |
Blocking the training crawler (GPTBot) is a separate, defensible choice; blocking these live fetchers directly removes you from AI shopping answers. The full list is in the AI crawler user-agent directory.
Step 2: Make sure your CDN or WAF isn't blocking agents behind your back
This is the most common own-goal, and it is invisible in normal analytics. A robots.txt that says "come in" means nothing if your CDN or bot-management layer returns a 403 to the agent first. Fetch three or four key product pages as GPTBot, OAI-SearchBot, PerplexityBot, and ClaudeBot and confirm you get HTTP 200, not a 403 or a challenge page. If you see blocks, work through whether your CDN is blocking AI crawlers and allow the user-agents explicitly. Do this check first; nothing downstream matters if the agent can't reach the page.
Step 3: Expose every buying fact as server-rendered text
An agent lifts what it can parse. Put price, availability, variants, shipping, and return terms in the server-rendered HTML as plain text, not only inside JavaScript widgets, images, or PDFs. A shopper agent comparing options will use whatever it can read cleanly and skip whatever it can't, routing instead to a marketplace or review site that states the fact plainly. This is the same discipline behind AI-search citation: if the fact isn't readable as text on the live page, it doesn't exist for the agent.
Step 4: Add product structured data
Mark your buying facts up with structured data so the agent reads them without guessing. Use Product and Offer for price, availability, and condition, FAQPage for question-and-answer blocks, and Organization for your entity. Schema doesn't replace on-page text, agents read the visible content too, but it labels what each fact is, which reduces misreads when the agent weighs you against alternatives. Getting the feed and the markup right is the whole game in agentic-commerce readiness and in preparing your product catalog for AI agents.
Step 5: Keep discovery and checkout completable on your own site
The surface a closed marketplace can never lock an agent out of is your own storefront, so make it work end to end. Search and filtering should function without a bespoke, JavaScript-only interaction an agent can't drive; cart and checkout steps should be reachable with real, semantic <a> and <button> elements. A page an agent can read but not act on wins the comparison and then loses the sale. This is the mechanics behind discovering in AI but buying on your site: own the transaction the marketplace is trying to keep.
Step 6: Monitor across more than one engine
Presence is not uniform. A product visible in ChatGPT can be absent from Perplexity or Google's AI surfaces, because each routes through a different model with different recall, and because each shops a different set of open and closed sources. Track how you are surfaced and compared for the real tasks buyers hand an agent ("best X under $Y for Z"), across engines, not just whether you rank in classic search. The same fragmentation shows up in how product discovery differs across ChatGPT, Google, and Perplexity.
What about agent checkout protocols like UCP and ACP?
Track and pilot them, but don't treat them as the entry ticket. Being discovered and recommended comes first and depends on the six steps above; open checkout standards like UCP and the Agentic Commerce Protocol matter once an authorized agent is ready to transact. They are moving fast and still settling in 2026, so experiment lightly and fix the reachability and legibility fundamentals first, as covered in WebMCP vs UCP.
In a market where big platforms block on purpose, the brands that win AI shopping are simply the ones that never blocked by accident, and made their product facts easy to read.
Do this well and your products stay selectable by the agents already shopping on your customers' behalf. Knowing whether ChatGPT, Perplexity, Google, and the agents built on them can actually reach and recommend your products, across engines and over time, is exactly what Buffy Intel measures. Stay open and legible; verify the agent can find you.