Free plan

← All guides

Agent Policy for Price-Comparison Agents

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

The short answer

Price-comparison agents are the readers your catalogue most often meets: they read and quote product data so their users can compare before buying. The policy pattern that serves them — and protects you — grants reading and quoting on stated terms, caps the pace, and refuses the after-life you do not want in data_use_terms: no resale of the catalogue, no training on your images. The store sample's reading rule, written for perplexity/agent, is the template.

Generate your policy — free · Validate an existing one

What is a price-comparison agent, in policy terms?

It is an agent whose work is reference, not transaction. Its verbs are browse, read, search, and quote; its product is an answer — your price, beside a competitor's, in front of a shopper. That makes its policy needs distinct from a shopping agent's: it never needs shop or checkout, and granting them adds risk without adding function. The comparison agent's real footprint is its pace (it may re-read prices often, because prices rot) and its retention (a price quoted today is data kept somewhere). A good comparison rule therefore spends most of its care on rate_limit and data_use_terms — the two fields the samples' reading rule is mostly made of.

The reading rule, dissected

The store sample's version grants perplexity/agent reading and quoting on free terms at 30 requests per minute, with the requirement that quotes link back. Every element is a decision. Free, because comparison traffic is demand arriving. 30 per minute, because price-checking is fetch-light compared to cart work. Link back, because the quote is the shop window: a comparison that names your price without naming your store is value leaking. Adapt the numbers, keep the structure. Stores whose prices are the product — dynamic, negotiated, or deliberately unpublished — read this pattern in reverse: the same fields, set to conditional or no, state the refusal with equal clarity. The pattern is the fields; the posture is yours.

Freshness, pace, and the honest tension

Comparison lives on freshness, and freshness costs requests. A price compared from a week-old cache misleads the shopper and embarrasses both the agent and you. Owners setting a comparison rule's cap should reckon with the refresh arithmetic of their own catalogue: a 5,000-product catalogue re-read at 30 per minute takes hours; at 60, half that — which is why working comparison arrangements often pair a moderate continuous cap with generous data_use_terms about caching ("prices may be cached for up to 24 hours" turns a thousand fetches into one). Writing the cache window into the terms is the single highest-leverage sentence in a comparison policy: it aligns the agent's freshness duty with your server's comfort, in the open. The rate limits guide covers the ladder logic in full.

What to refuse, even while welcoming comparison

Two refusals survive almost every negotiation. Resale: your catalogue re-sold as a data feed arms competitors with your own work; the samples' "no resale of catalog data" clause exists for exactly this. Training on images and creative: your photography and copy are brand assets with a life beyond price reference; refuse their training use in the same sentence block. Neither refusal touches the comparison function — an agent can quote today's price and link back without keeping your catalogue forever or learning to imitate your photography. If an operator argues a refusal blocks their model, the disagreement is about the after-life, and the data-use guide is the place to have it precisely.

Worked example: a tool retailer's comparison terms, one year on

Fictional example. Bridge & Bit, a fictional tool retailer with 7,400 products, grants 3 named comparison agents reading and quoting free at 40 requests per minute, data-use terms allowing 24-hour price caching, requiring store attribution on quotes, and refusing resale and image training. Its catch-all holds unknown comparators at conditional, 10 per minute. The measurable year: agent-attributed sessions became the retailer's second-largest referral category; price-accuracy complaints mentioning the store on comparison surfaces numbered 4, each traced to a comparator working from a competitor's stale mirror rather than Bridge & Bit's own permitted feed of itself. The retailer's conclusion is the pattern's whole argument: the cheapest way to be quoted correctly is to be readable on terms.

The price-comparison checklist

  1. Grant the reference verbs. browse, read, search, quote — and leave transaction verbs out; comparison never needs them.
  2. Set a freshness-aware pace. Size the cap against your catalogue's size and price-change rhythm.
  3. Write the cache window. State how long prices may be cached in data-use terms — the sentence that buys both freshness and quiet servers.
  4. Require attribution on quotes. Your price beside your name; the link back is the comparison's rent.
  5. Refuse resale and image training. The after-life fences that survive every negotiation, stated in the same rule.

Keep reading

How Often Should You Update Your Agent Policy? · Agent Policy vs Terms of Service: Which Does What? · Agent Policy vs robots.txt, Compared · Spec samples · Conformance · Pricing · All guides

Frequently asked questions

Do comparison agents need shop or checkout?

No. Comparison is reference work: read prices, quote them, send the shopper onward. Granting transaction verbs to a comparison rule adds your checkout risk to an agent whose function never uses it. Keep each rule fitted to the job it names — comparison rules are the cleanest example of the principle.

Free or paid terms for comparison access?

Most stores choose free, treating comparison as demand generation — the samples do. Paid-tier fits owners whose catalogue data itself has a market (specialist, hard-won, or licensed data). Either way the number that matters operationally is the pace cap paired with the cache window, not the term label.

How do I stop my prices being quoted without my store's name?

Put attribution in the rule's data_use_terms, as the store sample's reading rule does: quotes must link back. Agents built to honour declared terms honour that one like any other, and a comparator that strips attribution anyway has identified itself — the record of your published requirement is what you act on.

What pace do comparison agents actually need?

It scales with your catalogue and how often your prices move. A few thousand stable products refresh comfortably at 30 to 60 requests per minute; fast-moving prices argue for a cache window in the terms instead of raw request volume. Reckon the refresh time for your whole catalogue at the cap you are considering — that arithmetic settles most arguments.

Can I welcome comparison but refuse scraping?

That distinction is precisely what the vocabulary encodes. Comparison uses read and quote; scraping at scale is crawl-full-site. Permit the first pair, disallow the third, and the policy says — in machine-readable form — the sentence owners have been trying to say to bots for a decade: answer questions from us, do not mirror us.

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

← Back