Newsletter

All Pages

Every published page on this site (72 pages), grouped by section. This list is generated from the content collections on every build, so it is always complete. The machine-readable version is the XML sitemap at /sitemap-index.xml; AI agents can also start from /llms.txt.

Start Here

Handbook

Platforms

  • AI Shopping Platforms: How to Get Selected Across Engines One catalog, many agents: the shared signals that surface you across ChatGPT, Gemini, and Perplexity, plus the per-engine tie-breakers. Updated 2026-07-07.
  • ChatGPT Shopping Optimization Get your products surfaced in ChatGPT shopping: fix feed completeness, title-query match, and crawler access, not an 'enroll' button. Updated 2026-07-08.
  • Why You're Not in ChatGPT Shopping Products missing from ChatGPT shopping? It's almost always data gaps: vague titles, missing availability, blocked crawlers. Diagnose in order. Updated 2026-07-08.
  • Gemini & Google AI Mode Shopping Ranking Google AI Mode likely ranks via UCP and your Merchant Center feed: the lever is feed quality and Shopping Graph coverage, not ad spend. Updated 2026-07-07.
  • Perplexity Shopping Optimization Perplexity recommends what it can crawl and cite: win visibility with citable, structured product pages and clear crawler access. Updated 2026-07-08.

Tools

  • Merchant Tools Interactive tools for merchants: two evidence-typed checklists and the client-side Agent Legibility Analyzer.
  • Agentic Readiness Checklist A self-assessment of whether AI shopping agents can discover, read, and transact with your store: the eligibility layer, linking the validator that machine-checks each item.
  • Agent Selection Checklist A self-assessment of what wins the agent's choice once you're eligible: the legibility, completeness, and measurement signals no validator checks.
  • Agent Legibility Analyzer Paste product HTML, JSON-LD, or text and see token counts, detected structured fields, and a heuristic legibility self-assessment. Runs entirely in your browser.

Research

  • The AgentMint.net Research Lab Independent, reproducible tests of what moves agent selection: methodology public, results labeled 'Data collection in progress' until real. Updated 2026-07-07.
  • How We Test Agent Selection A repeatable protocol to measure your agent win rate across ChatGPT, Gemini, and Perplexity: prompts, sampling, and honest error bars. Updated 2026-07-07.
  • Do Description Rewrites Shift AI Recommendations? A field test replicating the ACES finding that description rewrites shift agent recommendations: methodology now, data as it lands. Updated 2026-07-08.
  • The Agent Selection Index A quarterly, reproducible benchmark of how AI shopping agents choose products and offers; methodology published now, numbers gated until author-verified. Updated 2026-07-08.

Reference

  • Reference Look-up surfaces: the glossary, the AI shopping crawler reference, protocol comparisons, and copy-paste blueprints.
  • Agentic Commerce Glossary Plain-English, sourced definitions for the agentic-commerce stack (agent selection, win rate, AEO, UCP, ACP, product feed), each typed by evidence. Updated 2026-07-07.
  • Agent Win Rate: Definition & How to Measure Agent win rate = the share of agent shopping sessions where your product is recommended or bought: here's how to define and measure it honestly. Updated 2026-07-07.
  • AI Shopping Crawler Reference: Every User-Agent Token The dated reference for every AI shopping-relevant crawler token: OpenAI, Google, Anthropic, Perplexity, Amazon, and Meta, plus what's confirmed absent. Updated 2026-07-08.
  • UCP vs ACP: Which Gets You Selected UCP (Google) vs ACP (OpenAI/Stripe): scope, agent reach, and which one gets your store selected on Gemini vs ChatGPT. Implement both. Updated 2026-07-07.
  • Agentic Payment Protocols, Explained AP2, Visa Trusted Agent, and Mastercard Agent Pay decide whether an agent can complete a purchase with you: table stakes to be selectable. Updated 2026-07-08.
  • Blueprints Versioned, copy-paste-ready, claim-typed templates for the machine surfaces this handbook teaches.
  • llms.txt for e-commerce catalogs A production-shaped llms.txt template for a store catalog, ordered by decision weight so agents read your product feed, offers, and policies first. Updated 2026-07-08.
  • Product markdown-mirror template A copy-paste per-product Markdown mirror template, ordered by decision weight so AI shopping agents read the facts that drive selection first. Updated 2026-07-08.
  • Token-efficient product JSON-LD patterns Copy-paste product JSON-LD skeletons (single product and multi-variant ProductGroup) for AI shopping agents, with the fields that carry selection weight. Updated 2026-07-08.
  • Edge worker for machine surfaces Two Cloudflare Worker variants for caching llms.txt and .md mirrors: the CDN path honors stale-while-revalidate; the Cache API refreshes manually. Updated 2026-07-08.

Signals

  • The Signals Database The structured catalog of agent-selection signals: what makes AI shopping agents pick one store, product, or offer over another, each evidence-typed with honest per-platform status.
  • ACP feed and checkout readiness Whether a store targeting ChatGPT checkout ships the ACP feed with correct eligibility flags and the return fields checkout-eligible items require. Updated 2026-07-08.
  • Agent infrastructure reachability Whether a crawler you allow in robots.txt actually receives your full page - a 200 over HTTPS, not a WAF challenge, rate-limit, or geoblock - rather than being silently turned away. Updated 2026-07-08.
  • Agent-readiness monitoring and ownership Whether you actually watch that agents reach, read, and buy, by confirming agent user-agents in logs, segmenting AI-referred traffic, and giving readiness a named owner who re-checks it. Updated 2026-07-08.
  • Defined agent win-rate measurement You define agent win rate for your catalog and measure it with a fixed prompt set across ChatGPT, Gemini and Perplexity, reporting a range with stated sampling rather than a single confident number. Updated 2026-07-08.
  • Track AI-answer presence, not just referrals Beyond referral traffic, you track whether each engine names or links your store for your top category queries, since answer-presence is the leading indicator of whether you're even in the consideration set. Updated 2026-07-08.
  • Descriptions answer relayed buyer questions Product copy answers, in buyer language, the concrete questions agents relay from users (fit, compatibility, use-case) rather than marketing fluff that dodges them. Updated 2026-07-08.
  • Content equivalence (no cloaking) Serving equivalent facts in different formats at one URL is legitimate content negotiation, but serving materially different content by user-agent is cloaking, a spam violation that reduces ranking eligibility. Updated 2026-07-08.
  • Crawl directives (robots.txt + sitemap) Whether your robots.txt and XML sitemap let AI search and citation crawlers find and reach your product URLs at all. Updated 2026-07-08.
  • Server-rendered, crawlable content Commerce-critical fields and full policy text appear in the raw server HTML, not behind JavaScript, a PDF, or a click-to-fetch accordion, so a non-executing crawler can read them. Updated 2026-07-08.
  • Delivery ETA is computable per destination Concrete handling time, transit time, order cutoff and per-destination shipping data let an agent answer 'when will it arrive?' and 'can it arrive by [date]?' rather than defer to a competitor that can. Updated 2026-07-08.
  • Query-conditional rewrites are a targeted experiment Rewriting descriptions to match likely queries produces concentrated, not universal, gains, so treat it as a measured experiment on candidate SKUs rather than a blanket promise. Updated 2026-07-08.
  • Earn endorsement, never self-label sponsored Agents reportedly penalize a self-applied 'Sponsored' tag and reward legitimate platform endorsements, so the durable play is earning genuine editor's-pick or verified-seller status, not faking badge markup. Updated 2026-07-08.
  • Facts per token, complete against peers Your listing packs specific, verifiable facts per token (specs, price, GTIN, availability, policy) and fills the comparable attributes the top listings in your category expose, so an agent has real substance to decide on rather than superlatives. Updated 2026-07-08.
  • Guest checkout completability Whether an agent with no account credentials can complete a purchase, because no forced signup, password, or email verification blocks the path. Updated 2026-07-08.
  • llms.txt priority ordering The llms.txt spec exposes only two prioritization levers: the order of the link lists and reserving the 'Optional' H2 for URLs an agent may skip. Updated 2026-07-08.
  • Machine-readable page representation Serving a token-light Markdown copy of a page, either via same-URL Accept negotiation or a distinct .md mirror, and advertising it with rel=alternate, hands agents a cheaper-to-read representation of the same facts. Updated 2026-07-08.
  • Price sensitivity varies by model The same discount moves engines differently, so a markdown that shifts share on one engine may barely move another and price strategy should be tested per engine, not blended. Updated 2026-07-08.
  • Offer consistency and freshness across surfaces Price, availability, shipping, and return values show identical, current numbers across the product page, structured data, feed, and checkout, so an agent never arbitrates a conflict or quotes an offer you no longer honor. Updated 2026-07-08.
  • Offer cost legibility An agent can read your full cost on a cold product page - the current price and any sale in parseable text with currency and tax, plus shipping, fees, unit price, and any free-shipping threshold - and every figure matches checkout. Updated 2026-07-08.
  • Offer markup completeness and truthfulness Whether the Product/Offer JSON-LD carries a valid price and currency, truthful availability and condition, a resolvable image, and present return/shipping objects that match what the shopper and checkout see. Updated 2026-07-08.
  • Origin serve-ability for agents Keeping your origin open and cheap to serve, by edge-caching the near-static machine surfaces and allowlisting verified bots while rate-limiting only unverified spoofers, so a verified agent always gets a fast 200 instead of a block. Updated 2026-07-08.
  • Agent-payment rail readiness Whether your payment stack can complete an agent-initiated purchase over rails like AP2, Visa Trusted Agent Protocol, or Mastercard Agent Pay. Updated 2026-07-08.
  • Audit grid position bias per provider Grid position is a reported, provider-specific selection lever, so you record your rendered on-screen position per engine to avoid mistaking a position swing for a content win or loss. Updated 2026-07-08.
  • Competitive price awareness with parity Price is a real selection lever whose weight varies by model (ACES measures per-model price elasticity), so you track where your total landed cost ranks for target queries and hold price parity between agents and humans. Updated 2026-07-08.
  • Product identity and canonical URL Whether each product has one stable canonical URL matching the feed link plus real manufacturer identifiers, so an agent can match your item to the same product across competing stores. Updated 2026-07-08.
  • Re-test after every major model update Model updates can drastically reshuffle market shares, so a win measured last quarter can silently evaporate and the win-rate protocol should re-run within a set window on each major release. Updated 2026-07-08.
  • Return terms are fully legible The return window, who pays return cost, the return method, and the refund type are each stated in plain prose that matches the structured-data fields, leaving nothing for the agent to infer. Updated 2026-07-08.
  • Review depth: count, recency, distribution Rating is a reported selection lever, so complete review data (a truthful count, recent reviews, and a full star distribution shown on the PDP) gives a rating-sensitive agent more credible depth to weigh than a bare average. Updated 2026-07-08.
  • Seller identity reachable in two hops A crawlable path to seller identity, address and contact within two hops of a product page signals a real, accountable merchant that trust-weighting agents won't discount. Updated 2026-07-08.
  • UCP profile and order-management readiness Whether a store targeting Google/UCP surfaces publishes a public /.well-known/ucp profile and can return order status for the capabilities it declares. Updated 2026-07-08.
  • Validation hygiene (use existing validators) Whether structured data, feeds, and protocol manifests are run through existing maintained validators rather than a homegrown checker. Updated 2026-07-08.
  • Variants and titles use the words shoppers type Variant names and product titles lead with explicit, literal attributes (category, size, color, material, use-case) an agent can map to a user's query, not internal codes or brand-only names. Updated 2026-07-08.

Site

  • Agent Selection Updates A dated log of changes to how AI shopping agents select products and stores, and to the protocols merchants track. Updated 2026-07-09.
  • Privacy Policy How AgentMint.net handles your data: no analytics and no cross-site ad tracking, plus an optional email newsletter. We never sell your data. Updated 2026-07-08.
  • All Pages Every published page on AgentMint.net, grouped by section.

[ newsletter ]