A shopping agent sent to find “a waterproof hiking boot under $150 in stock near me” will not open your homepage. It will query a feed, check a schema block, and look for a manifest telling it what it is allowed to do next. An agentic commerce readiness checklist is built for that query, not for a person scrolling a category page.
Key Takeaways
- Agentic commerce readiness depends on structured, current data an agent can parse without rendering a page, not on persuasive copy or page layout.
- Product schema and catalog feeds establish what exists; an agents.md or llms.txt manifest establishes what an agent is permitted to do with it.
- Machine-readable offer state (price, stock, promotion validity) has to update in near real time, because an agent that acts on stale data damages trust it cannot easily repair.
Catalog Data Is the First Agentic Checkpoint
Before an agent can recommend a product, it has to find it in a form it can parse without loading a browser. That means a catalog feed with consistent identifiers (GTIN, MPN, SKU), accurate category mapping, and attributes that match how people actually describe the product, not how your merchandising team labels it internally.
Most retailers already export a feed like this for Google Shopping or Meta. The gap is coverage and freshness. A feed updated nightly is fine for a search results page a person scans and judges. It is a liability for an agent that transacts on the data in front of it the moment it reads it. If the boot sold out four hours ago, the feed needs to know.
Product Schema Is the Floor, Not the Proof
Schema markup (Product, Offer, AggregateRating, Review) tells an agent what a page claims about itself. It does not tell the agent whether to believe the claim. Trust is a graph of agreement across independent sources, so a retailer’s own schema carries less weight than the same facts corroborated elsewhere: a retailer review site, a comparison publisher, a manufacturer spec sheet.
This is where retailers tend to overinvest in one layer and skip the other. Full schema coverage with no independent corroboration gives an agent a complete but unverified claim. Strong third-party coverage with thin or missing schema gives an agent verified facts it cannot efficiently retrieve. Both have to be present.
An agent does not see your homepage, it sees whatever your feed, your schema and your manifest say is true, and it treats all three as equally authoritative until proven otherwise.
What an agents.md or llms.txt Manifest Actually Does
A manifest is new ground for most retail SEO teams, and it does something schema does not: it tells an automated agent what it is permitted to crawl, query, and act on, and where to find the machine-readable endpoints that matter (product feed location, checkout API if one is exposed, rate limits, contact for programmatic access). Think of it as a robots.txt written for an agent that wants to do something, not just index something.
Without one, an agent falls back to whatever it can infer from your robots.txt and sitemap, which were written for crawlers that read and rank, not for agents that read and act. A retailer with no manifest is not necessarily blocked. It is just invisible as a preferred source when an agent is choosing between a retailer that has stated its terms clearly and one that has not.
This is retrieval presence in its most literal form: structured, current, and built for a machine to find on the first request.
Machine-Readable Offer State Means Price, Stock and Promotions an Agent Can Trust
Offer state is the part of this checklist that expires fastest. Price, inventory count, and promotion validity window all need a timestamp an agent can check, and an endpoint that reflects reality at the moment of the query, not at the moment of the last batch export.
A retailer that gets this wrong does not lose a sale politely. An agent that recommends a product, quotes a price, and then hits a checkout flow that shows a different number has a trust problem that compounds the wrong way. Citations compound when they hold up; they also stop compounding the moment an agent’s users report that the data was wrong. Offer state is the one part of the checklist where “mostly accurate” behaves like “unreliable.”
This Is Not the Same Prep as Ecommerce SEO
Standard ecommerce SEO prep optimizes for a ranking algorithm and a human who clicks, scans, and decides. It rewards keyword-rich titles, internal linking, page speed, and conversion-focused copy. None of that is wasted work, but none of it is what an agent is evaluating.
Agentic readiness optimizes for a different consumer: software that parses a feed, checks a manifest, verifies a claim against outside sources, and executes an action. Share of Voice, how often a brand is mentioned across the web, can be high for a retailer that still has near-zero Share of Citation, the rate at which agents actually pull that retailer in and act on its data. A retailer can rank well in Google and still be functionally absent from an agent’s decision, because the two systems are reading for different things. The SEO team’s work extends into this layer; it does not get replaced by it.
Where to Start This Quarter
Order of operations matters more than completeness. Start with offer state accuracy on your top 20% of SKUs by revenue, since stale price or stock data there does the most damage. Add or audit Product and Offer schema on the same set, checking it against what your actual feed says rather than assuming they match. Publish a basic agents.md or llms.txt pointing to your feed location and stating what automated access is allowed. Then widen catalog coverage and schema depth outward from that core set, rather than trying to cover the full catalog at once.
Treat this as a standing audit, not a one-time project. Feeds drift, promotions expire, and a manifest that was accurate at launch goes stale the first time the team changes a URL structure without updating it.
FAQs
Do I Need Both Product Schema and a Manifest File?
Yes. Schema describes the product and its offer. A manifest describes what an agent is allowed to do with that information and where to find the machine-readable endpoints behind it. One without the other leaves an agent with either unverified claims or no clear path to act on verified ones.
How Often Does Offer State Need to Update for an Agent to Trust It?
As close to real time as your systems allow. A nightly batch feed is adequate for a page a person reads and judges for themselves. An agent that quotes a price or confirms stock from that same feed is making a claim on your behalf at query time, so the gap between reality and the feed is the gap where trust breaks.
Does This Replace Our Existing Ecommerce SEO Work?
No. Ecommerce SEO still earns rankings and clicks from people searching directly. Agentic readiness is a separate layer built on top of it, using the same catalog and product data but structured for a different consumer: software that parses and acts rather than a person who scans and clicks.
What Is the Fastest Way to Check if Agents Can Already Read Our Catalog?
Query a few of your top products through a model that cites sources inline, such as Perplexity, and see whether your retailer appears as a cited source with accurate price and stock. If it does not appear, or appears with outdated details, that gap is your starting point.
