Free plan

← All guides

Agent Policy for Online Stores: a Practical Pattern

Published November 14, 2026 · Reviewed November 2026 · By the Ingenious AI team

The short answer

An online store's agent policy answers one commercial question: how do agents participate in selling your products? The proven pattern has three rules working together — shopping agents welcomed on the browse-but-not-checkout split, comparison agents granted read and quote with attribution, and a conditional catch-all at walking pace. The published store sample is this pattern's reference implementation, written for a fictional store and ready to adapt.

Generate your policy — free · Validate an existing one

What does a store's policy need to decide?

Four things, in order of money. Whether agents may work the catalogue (the shopping verbs), whether they may spend (checkout — almost always no, separately decided), what your catalogue data is worth (terms and after-life), and at what pace agent traffic may run against infrastructure sized for humans. Every other store-policy question is a refinement of those four. The samples' fictional store, Brightline Goods, answers them in three rules; your store's answers will differ in the numbers and rarely in the structure.

The three-rule store pattern

Rule one — the shopper. A named shopping agent at yes: browse, read, search, shop permitted; checkout, submit-form, api-call disallowed; the sample prices it paid-tier at 60 requests per minute. Rule two — the reader. A comparison agent at yes: read and quote, free, 30 per minute, attribution required in the data-use terms — the pattern the comparison guide specialises. Rule three — everyone else. generic/* at conditional, 10 per minute, same fences. Behind the rules, a conditional default states the house posture: this store talks before it grants. Three rules, roughly fifty lines, and every agent class your store will meet this year has a written answer.

Store-specific judgement calls

Inventory truth: agents quote availability as readily as price. If your stock data lags, say so in the data-use terms ("availability indicative; confirm at checkout") rather than leaving agents to over-promise on your behalf. Promotions: agent-visible prices are your public prices; if trade or member pricing exists behind logins, agents never see it, and the policy need not pretend otherwise. Seasonality: pace caps set for March may pinch in December — the review rhythm in the cadence guide should include a pre-peak pass for retail. Marketplaces within the store: third-party sellers' data may carry its own constraints; where it does, the data-use terms are where the store says so, once, for all agents.

What stores get wrong

Two errors recur. The first is the blanket refusal written from crawler-era instinct — a block that also refuses the customers' own shopping agents, discoverable only when a partner asks why their assistant returns empty-handed from your store. The second is the unfenced welcome: shop granted with checkout left unlisted and data-use terms empty, a policy that reads as permission for everything it forgot to refuse. Both errors are one review away from the pattern above, which is neither maximally open nor maximally closed — it is priced, paced, and fenced, the three adjectives of every store policy that survives its first peak season.

Stores ask, finally, how they will know the policy is working. The honest metrics are unglamorous and available: agent-shaped sessions your analytics can see arriving inside the stated pace; carts built and handed back, where your platform can attribute them; operator enquiries that begin with the file's own vocabulary — a rule quoted, a term referenced — because those enquiries are the policy doing commercial work; and the refusal side's quiet zero, agent-completed transactions requiring unwinding. Review those four at the quarterly pass and the file stops being a compliance artefact. It becomes what the best store policies already are: a small, public, self-describing piece of the store's trading machinery.

Worked example: a homeware store adapts the sample in an afternoon

Fictional example. Kilnhouse, a fictional homeware store with 2,750 products, started from the published store sample and changed the decisions, not the format. Its shopping rule went to a different named agent at 45 requests per minute on free terms — Kilnhouse treats agent shoppers as pure demand. Its reader rule added search and a 24-hour price-cache allowance in the data-use terms. Its catch-all matched the sample. One adaptation was the store's own: a fourth rule for a local same-day-delivery agent, conditional, opened as a placeholder for a negotiation the owner intends to have. First-quarter tally under the policy: 540 agent-built carts, 131 completed orders, 0 agent checkouts, and 1 delivery negotiation — the placeholder rule having done its quiet work of advertising the conversation.

The store policy checklist

  1. Start from the sample. Adapt the published store sample's three-rule pattern; change decisions, not structure.
  2. Split shop from checkout. Grant the shopping verbs; fence the money verbs in every rule including the catch-all.
  3. Give readers their own rule. Comparison agents get read/quote with attribution and a cache window in the data-use terms.
  4. Pace for your peak. Set caps your infrastructure survives in your busiest week, and calendar a pre-peak review.
  5. Write the after-life. No resale, no image training, availability caveats where your data lags — in the terms, in the same pass.

Keep reading

Agent Policy for Banks and Finance Sites · Agent Policy for News Sites and Blogs · Agent Policy vs robots.txt, Compared · Spec samples · Conformance · Pricing · All guides

Frequently asked questions

Should my store allow agents at all?

The decision guide works the question in full, but the short store answer is: agents are a channel your customers choose, and the policy is how you appear in it on your terms. Stores with public catalogues generally gain more from a priced, fenced welcome than from a refusal their customers' agents experience as absence.

What does the store sample charge shopping agents?

Its shopping rule carries paid-tier terms at 60 requests per minute — the sample store prices working catalogue access. Free terms are equally expressible and suit stores treating agent traffic purely as demand. The pattern is the three rules and the fences; the price signal is your commercial call.

How do I handle prices that change hourly?

Through the data-use terms: grant comparison agents a short cache window (or none), state that prices are confirmed at checkout, and size your pace cap so a diligent agent can refresh the catalogue inside the window. Freshness is a terms-and-pace problem, and the policy has exactly the fields for it.

One of my suppliers restricts their product data. Where does that go?

In data_use_terms on the affected rules — the plain-language field exists for constraints exactly like supplier terms: what may be shown, retained, or passed on. Write the restriction once in the policy rather than discovering, per agent conversation, that it was never stated.

Can I run different policies for different storefronts?

Yes — each origin publishes its own file describing its own terms, which is the multi-site pattern. Shared ownership argues for a shared template with per-store decisions, reviewed as a fleet; the multi-site guide covers the register and review habits that keep several store policies coherent.

Generate a policy — free · Validate a policy · What an agent access policy is

← Back