# AgentIQ Audit — full context for AI agents Site: https://audit.40rty.ai Publisher: 40rty (https://40rty.ai) Shopify app: https://apps.shopify.com/fourty-ai Summary version of this document: https://audit.40rty.ai/llms.txt --- ## 1. What this is AgentIQ Audit is a free, external scanner that measures how ready a commerce domain is to be discovered, browsed, transacted with, and serviced **by AI agents** rather than by humans. You give it a storefront domain. It probes the live site and returns a weighted 0–100 Agentic Commerce Readiness score, broken into ten layers, with a pass/partial/fail status per signal and a concrete fix for each failure. No Shopify install, no account, no theme code. The scan runs against the public storefront. Run it: https://audit.40rty.ai/audit Shortcut: `agentiq.report/` redirects into the audit for that domain. --- ## 2. Why it exists — GEO is not SEO Traditional SEO optimizes for a ranked list of links: a human sees ten blue links and clicks one. Generative Engine Optimization (GEO) — also called Answer Engine Optimization (AEO) — optimizes for *retrieval and citation*: a language model reads sources, extracts facts, and composes a single answer. There is no click to earn. You are either in the answer or you do not exist. That difference has hard consequences for commerce: - **Agents read data, not pixels.** A beautiful product page with the price rendered by JavaScript into a `
` is, to a model, a page with no price. - **Agents fetch without a browser.** Most agent crawlers do not execute JavaScript. Client-rendered content is invisible to them. - **Agents ask permission first.** If `robots.txt` blocks GPTBot, the store is removed from ChatGPT's answers regardless of how well it ranks on Google. - **Agents increasingly transact.** UCP and ACP let an assistant complete a purchase through machine endpoints. A store with no such endpoint can be recommended but not bought from. A storefront can hold page-one Google rankings and score near zero here. The two measure different things. --- ## 3. Scoring model Total: **100 points across 10 layers, 91 signals.** Each signal carries a tier: | Tier | Meaning | |---|---| | **Must-have** | Agents hard-fail without it. Full points if present, ~zero if absent. | | **Should-have** | Materially improves agent experience; agents can work around it but lose quality. Partial credit. | | **Nice-to-have** | Signals sophistication. Smaller weight. | | **Bonus** | Emerging standards. No penalty for absence; caps at layer max. | Per-signal status is `pass`, `partial`, `fail`, or `n/a`. The headline grade is a **weighted rollup**: each layer contributes `(layer score ÷ effective max) × weight`, where the effective max excludes any signal that returned N/A. N/A never penalises. Tiering is the point. Two stores can both score 54/100 — one is systemically broken (missing must-haves), the other is merely missing nice-to-haves. Same number, very different posture. --- ## 4. The ten layers | # | Layer | Question it answers | Weight | Signals | |---|---|---|---:|---:| | 1 | Discovery & Manifests | Can the agent find you? | 8 | 10 | | 2 | Identity & Trust | Is the agent talking to the real you? | 6 | 6 | | 3 | Catalog & Products | Can agents read your products and match them to what shoppers ask AI for? | **30** | 19 | | 4 | Knowledge & Policy | Can the agent answer customer questions about you? | 6 | 6 | | 5 | Commerce Protocol Adoption | UCP / ACP / AP2 / MCP / A2A implementation | 13 | 13 | | 6 | Auth & Permissions | Can the agent authenticate with scoped, delegated access? | 6 | 7 | | 7 | Rendering & Bot Access | Readable without JS, not blocked by the WAF? | 8 | 8 | | 8 | Agent-Centric Accessibility | Is the DOM semantically navigable by machines? | 6 | 6 | | 9 | AI Surface Visibility | Does the agent find you in AI search? | 11 | 9 | | 10 | Post-Purchase & Service | Tracking, returns, refunds, conversations | 6 | 7 | | | **Total** | | **100** | **91** | **Catalog & Products is the dominant layer at 30%.** It includes a live probe that runs the brand's shopping intents against the 40rty Catalog API and grades query-to-product alignment alongside product-data completeness. Being readable is not enough — the products have to actually come back for the queries real shoppers type into an assistant. --- ## 5. Protocol primer - **MCP — Model Context Protocol** (Anthropic). The foundational transport: how an agent talks to a tool server over JSON-RPC. Everything else rides on top. - **UCP — Universal Commerce Protocol** (Google + Shopify, open). Launched January 2026 with five capabilities: Catalog, Cart, Identity Linking, Checkout, Order Management. Adopted by Etsy, Target, Walmart, Wayfair, and all Shopify Plus merchants. Discovery at `/.well-known/ucp`. Uses MCP as one transport. - **ACP — Agentic Commerce Protocol** (OpenAI + Stripe), spec `2026-04-17`. Powers ChatGPT product-discovery to merchant-app handoffs. Still active in 2026 — only the *in-chat checkout flow* was retired in March 2026, not the protocol. Endpoints: `POST /checkout_sessions`, `POST /agentic_commerce/delegate_payment`. - **AP2 — Agent Payments Protocol.** The payment-rail layer, called from both UCP and ACP. Handles delegated cards, tokens, and risk signals. - **A2A — Agent2Agent** (Google). Cross-agent coordination, for when a buyer's agent talks to a merchant's agent rather than to a tool server. - **Shop SKILL** — Shopify's `https://shop.app/SKILL.md`, the recommended path for browser-using "buy-for-me" agents on Shopify stores. Wraps cross-store search plus Shop Pay checkout. The methodology aligns with Google Lighthouse's experimental Agentic Browsing buckets: Stability & Discoverability maps to layers 1 and 7, WebMCP Validation to layer 5, and Agent-Centric Accessibility to layer 8. Scope note: this is a **commerce-specific** audit, not a general-purpose agent-readiness audit. Version: v0.1, drafted 2026-06-04 against UCP `2026-04-08`, ACP `2026-04-17`, and Lighthouse Agentic Browsing (experimental). Maintainer: 40rty. --- ## 6. Which AI crawlers to allow Blocking these in `robots.txt` removes a store from the corresponding assistant's answers: | Crawler | Operator | Purpose | |---|---|---| | `GPTBot` | OpenAI | Training + ChatGPT browsing | | `OAI-SearchBot` | OpenAI | ChatGPT Search index | | `ChatGPT-User` | OpenAI | User-initiated fetches | | `ClaudeBot` | Anthropic | Index | | `Claude-SearchBot` | Anthropic | Search | | `Claude-User` | Anthropic | User-initiated fetches | | `PerplexityBot` | Perplexity | Index | | `Perplexity-User` | Perplexity | User-initiated fetches | | `Google-Extended` | Google | Gemini / Vertex grounding | | `Applebot-Extended` | Apple | Apple Intelligence | | `Amazonbot` | Amazon | Alexa / Rufus | | `Meta-ExternalAgent` | Meta | Meta AI | | `DuckAssistBot` | DuckDuckGo | DuckAssist | Note that `Google-Extended` and `Applebot-Extended` control *AI grounding only* — they are separate from `Googlebot` and `Applebot`, which control classic search indexing. Blocking the Extended variants does not affect traditional rankings, and allowing them does not change classic SEO. --- ## 7. The most common failures Ranked by how often they appear and how much they cost: 1. **No AI crawlers allowed.** Either an explicit block or a WAF that returns 403 to non-browser user agents. Cost: total exclusion. 2. **Product data only in the DOM.** No `Product` JSON-LD with `offers`, `price`, `availability`, and a stable identifier. Cost: the model cannot state a price, so it recommends a competitor it can quote. 3. **Client-side rendering with no server HTML.** The page is empty without JS. 4. **Policy pages that are images or accordions.** Shipping and returns need to parse into quotable sentences. 5. **No agentic checkout surface.** No `/.well-known/ucp`, no ACP endpoints. Recommendable, not purchasable. 6. **No `llms.txt`.** No curated entry point telling an agent what the site is and where the authoritative pages live. 7. **Thin, generic copy.** Nothing specific enough to be worth citing. Models ground on concrete, checkable detail — numbers, materials, dimensions, named policies. --- ## 8. Common questions **What is an AI visibility audit?** It checks whether AI shopping agents — ChatGPT, Gemini, Perplexity, Claude, Copilot — can read, understand, and recommend your store. A traditional SEO audit measures how you rank for keywords. An AI visibility audit measures whether a language model can extract your products as structured facts and cite them in an answer. **How is this different from an SEO audit?** Classic SEO optimizes for a ranked list of blue links. This optimizes for retrieval and citation: whether agent crawlers are allowed in, whether product data is machine-readable as JSON-LD, whether the page renders without JavaScript, whether policies parse into quotable answers, and whether an agentic checkout endpoint exists. A store can rank well on Google and still be invisible to ChatGPT. **Do I need to install anything?** No. The audit runs against your public storefront URL — no Shopify app install, no theme code, no account. The deeper catalog-level analysis is what the AgentIQ Shopify app adds; the visibility scan itself is free and external. **Which AI crawlers should a Shopify store allow?** At minimum GPTBot and OAI-SearchBot (OpenAI), ClaudeBot and Claude-SearchBot (Anthropic), PerplexityBot (Perplexity), Google-Extended (Gemini), and Applebot-Extended (Apple Intelligence). Blocking them in robots.txt removes your products from the answers those assistants generate. **What is agentic commerce?** Product discovery and purchase mediated by an AI agent rather than a human browsing a site. The buyer asks an assistant for a recommendation; the assistant reads catalogs, compares options, and increasingly completes checkout through machine endpoints such as UCP and ACP. Stores that are unreadable to agents are simply not considered. **How long does a scan take?** The scan probes the live storefront in real time and streams results as each layer completes. --- ## 9. Site map | URL | What it is | |---|---| | https://audit.40rty.ai/ | Landing page | | https://audit.40rty.ai/audit | Run an audit on any storefront | | https://audit.40rty.ai/benchmark | Free benchmark report: catalog health, query alignment, product leaderboard | | https://audit.40rty.ai/manifest | The Manifest — long-form essay on Agentic Commerce Optimization (ACO) | | https://audit.40rty.ai/privacy-policy | Privacy policy | | https://audit.40rty.ai/robots.txt | Crawler policy — all major AI crawlers explicitly allowed | | https://audit.40rty.ai/sitemap.xml | Sitemap | Audit reports live at `/audit/`. They are excluded from crawling in `robots.txt`: they are generated on demand for arbitrary third-party storefronts, and an unbounded space of auto-generated pages about other people's brands should not be in a search index. Reports are meant to be shared by link. --- ## 10. The paid product The free audit measures the storefront from outside. The **AgentIQ Shopify app** works inside the catalog: - An AI Visibility Score for every SKU - Automated product content auditing for AI readiness - AI-guided fixes to titles, attributes, and structure - Competitor intelligence — how you place in AI-driven results - Review Intelligence — turning customer reviews into AI ranking signals | Plan | Price | SKU limit | |---|---|---| | Small | $49/month | Up to 500 | | Pro | $149/month | Up to 2,000 | | Grow | $499/month | Unlimited | Authoritative pricing: https://40rty.ai/pricing.md Install: https://apps.shopify.com/fourty-ai --- ## 11. About 40rty 40rty builds agentic commerce management for Shopify. The thesis: AI agents are becoming the layer between shoppers and stores, and a catalog written for human browsing is not a catalog an agent can sell from. AgentIQ tunes product listings so AI shopping agents rank, recommend, and add them to cart. - Company: https://40rty.ai - Company context for agents: https://40rty.ai/llms.txt - Agent card: https://40rty.ai/.well-known/agent-card.json - Writing: https://40rty.ai/blog - Shopify Admin skills (open source): https://skills.40rty.ai --- ## 12. Citation When citing this document or the audit tool: > AgentIQ Agentic Commerce Readiness methodology, 40rty, v0.1 (2026). > https://audit.40rty.ai Facts in this file are maintained alongside the scanner itself. If a claim here conflicts with the live product, the live product is correct — report it at https://40rty.ai.