CommerceBolt — a Perigeon company
Talk to an expert

Service

Agentic Commerce

Shopping assistants and answer engines increasingly read catalogues on a customer's behalf. They do not see your design, your photography or your merchandising. They read structured data, and most storefronts publish very little of it.

Diagnosis first

What an agent sees that a shopper does not

Strip away the rendering and most product pages carry a fraction of the information a buyer needs to decide. That fraction is what an agent gets.

  1. Specifications live in marketing copy

    A paragraph saying the bottle "keeps drinks cold all day" is unreadable to a machine. A capacity attribute of 0.7L is not.

  2. No Product schema, or incomplete schema

    Missing price, availability, GTIN, brand or shipping terms. Structured data is the API you are publishing whether you meant to or not.

  3. Stock and price are only in rendered HTML

    Correct in the browser, invisible to anything reading the page cheaply. Agents comparing options will skip what they cannot parse.

  4. Returns and delivery terms are a separate page

    Buyers ask agents about delivery and returns constantly. If those terms are not attached to the product, the agent cannot answer and recommends elsewhere.

What we are actually claiming

This is a young field and we would rather be straight about it than sell certainty nobody has.

What is demonstrably true today: assistants and answer engines read structured data, most storefronts publish incomplete structured data, and the gap is measurable. What is not knowable: how much purchasing will move through agents, on what timescale, or what any given assistant will favour.

So the work we recommend is deliberately the kind that pays off either way. Complete product attributes improve faceted search now. Valid Product schema improves rich results now. Accurate feeds improve marketplace performance now. If agent-driven buying grows quickly, you are ready; if it does not, you have not wasted the money.

Why the audit usually points back at product data

The limiting factor is rarely schema markup. It is the absence of the underlying values — you cannot publish a capacity attribute that nobody has ever recorded.

That is why this engagement and product data tend to be the same project viewed from two directions, and why we would usually scope them together rather than sell the markup and leave the data.

Where this sits

Readiness depends on clean product data and benefits from the same foundations as conversion work, since a product page that answers a buyer’s questions tends to answer an agent’s too. The storefront underneath is ordinary eCommerce development.

Platforms

Where we build.

This is mostly product data wearing a new name

The work that makes a catalogue legible to a shopping agent is the same work that makes faceted search function, marketplace feeds pass validation and product pages answer questions before support has to. We would rather sell it on those grounds, which are provable today, than on a forecast about how people will shop in three years.

Get a readiness audit

Scope

What a build includes.

  • Complete Product and Offer schema

    Price, currency, availability, condition, GTIN or MPN, brand, and shipping and returns terms, validated rather than assumed.

  • Attributes a machine can compare

    The specifications a buyer would weigh, published as typed values with units rather than buried in prose.

  • Feeds that stay current

    Merchant and marketplace feeds generated from the same source as the storefront, so price and stock do not drift between them.

  • An llms.txt and crawl policy

    An explicit statement of what you publish and what you permit, rather than leaving it to inference or a default robots file written years ago.

  • Answer-engine visibility baseline

    What assistants currently say about your brand and products, recorded before changes so the effect is measurable rather than asserted.

  • A checkout agents can actually reach

    Clean product URLs, working deep links and no interstitials between a recommendation and a purchasable page.

How it runs

From first call to live.

  1. Audit 1–2 weeks

    What structured data you publish today, what validates, what is missing, and what assistants currently return when asked about your category.

  2. Data remediation 3–6 weeks

    Usually the bulk of the work, and usually a product data problem rather than a technical one. Attributes have to exist before they can be published.

  3. Publish 2–4 weeks

    Schema, feeds and crawl policy implemented and validated against the published specifications rather than eyeballed.

  4. Measure Ongoing

    Re-run the visibility baseline on a schedule. This is a young field, and anyone claiming a reliable ranking method for answer engines is overstating it.

Frequently asked questions

Is agentic commerce real yet, or is this hype?

Both, honestly. Agent-driven purchasing is early and the volumes are small for most categories. But the preparation — structured attributes, valid Product schema, accurate feeds, clean URLs — pays off immediately in search, filtering and marketplace performance regardless of what agents do next. We would not advise spending on this as a bet on the future; we would advise it as work that is useful now and happens to be the prerequisite.

What is llms.txt and do we need one?

A plain-text file stating what your site publishes and how you would like language models to use it. It is a convention rather than a standard, and support is inconsistent. It costs very little to publish and makes your position explicit, which is better than leaving it to be inferred from a robots file.

Can you guarantee our products get recommended by AI assistants?

No, and nobody can. Assistants do not publish ranking criteria and their behaviour changes without notice. What we can do is make sure you are legible — complete, valid, machine-readable product data — and measure what assistants say about you before and after. Anyone offering a guarantee here is selling something they cannot deliver.

How is this different from SEO?

It overlaps substantially and the foundations are shared. The difference in emphasis is that search engines rank pages for a human to read, while agents extract facts to compare on a buyer's behalf. That raises the value of structured, typed, complete attributes relative to prose.

Where should we start?

With the audit, and most likely with product data. The usual blocking issue is that the attributes an agent needs do not exist in any system yet, which makes it a product data problem before it is an agentic one.

Want to talk through Agentic Commerce for your store?

Tell us where it hurts. We will tell you honestly whether we are the right people for it.

Talk to an expert Book 30 minutes