Part of Commerce Readiness
Agent Merchant
Verify whether a shopping agent can buy from your store: UCP, Universal Cart and a live checkout
What are LLMs and why do they matter for your business?
LLM stands for Large Language Model. These are the artificial intelligence systems behind tools like ChatGPT, Google Gemini, Claude, and Perplexity. They work by processing enormous amounts of text to learn human language patterns, which enables them to answer questions, generate content, and hold natural conversations.
When a user asks ChatGPT "what's the best project management tool?" or asks Perplexity "which laptop should I buy for graphic design?", the LLM analyzes everything it learned from the web and selects the sources it considers most trustworthy to build its response. If your brand, product, or website doesn't have the right signals, the model simply won't include you in that answer, and your competition will take that spot.
Google AI Overviews is Google's version of this technology: instead of showing just a list of links, it now generates a direct AI-powered answer at the top of search results. This means the first result many users see is no longer a link to your site, but an AI-generated response that may or may not mention you.
In short: LLMs are the new intermediaries between your business and your potential customers. More and more people use these tools to research, compare, and decide what to buy. AuraMetrics helps you ensure that when AI answers questions about your industry, your brand is part of the answer.
What it does
Agent Merchant audits whether your store is ready for AI shopping agents: autonomous systems that discover products, compare, build a cart and reach payment on someone's behalf. Google (Universal Cart and the UCP protocol), OpenAI and the major assistants already run or are preparing this kind of purchase. The module scores four pillars: access (can an agent read your site, or do robots.txt, mandatory JavaScript and anti-bot systems stop it?), structured data (do your product pages expose price, stock, variants and policies in a machine-readable format?), UCP checkout (do you publish a valid /.well-known/ucp manifest with the checkout capability declared?) and AP2 / Universal Cart payments (can your payment flow accept a purchase mandate from an agent?). Unlike a feed audit, Agent Merchant does not stop at the catalog: it verifies the protocol and tests the full flow.
Why it matters
When an assistant buys on the user's behalf, a store the agent cannot navigate simply does not exist for that sale. Most sites fail today for trivial reasons: blocking popups, mandatory login, captchas in the cart or a checkout that only works with heavy JavaScript. Publishing a valid UCP manifest and having a checkout an agent can complete positions you as an early adopter in a channel that is just starting. Agent Merchant tells you exactly which step stops you and what to fix first.
How it works
It first crawls the site with a focus on product pages, cart and checkout, and reads robots.txt and the /.well-known endpoints. It then verifies the UCP manifest deterministically: it fetches /.well-known/ucp, validates version, services, capabilities and payment handlers, and emits PASS/FAIL findings cited to the matching section of the spec. This part uses no AI: it is code, and the result is reproducible. From that it computes a per-pillar score from the detectors and generates a single narrative explaining findings and priorities. Optionally it runs the ghost agent, a live test: a real browser-driven agent walks your store the way a shopping assistant would. It finds a product, adds it to the cart, proceeds to checkout, fills shipping with synthetic data and stops at the payment step. It never enters a card or completes a purchase: the moment a payment field appears, the test ends and records how far it got.
- 1Enter the store's domain
- 2Crawl of product pages, cart and checkout, plus robots.txt and /.well-known
- 3Deterministic validation of the UCP manifest and payment handlers
- 4Per-pillar score and narrative with priorities
- 5Optional: live ghost agent up to the payment step
What each status means
Metrics you get
Recommended use cases
Frequently asked questions
What is UCP?
UCP (Universal Commerce Protocol) is the open agentic-commerce protocol driven by Google. A store adopts it by publishing a manifest at /.well-known/ucp declaring its services, capabilities (checkout is required) and payment handlers. Agent Merchant fetches and validates that manifest.
What is Universal Cart and how does AP2 relate?
Universal Cart is Google's cart experience spanning Search, Gemini, YouTube and Gmail. To take part, a store needs UCP and AP2, the payments protocol with signed mandates that lets an agent pay with verifiable authorization. The payments pillar measures how close you are.
Does the ghost agent buy anything?
No. It has two guards: the prompt forbids completing purchases or entering cards, and the code ends the walk the instant a payment field appears. It uses synthetic, clearly fictitious shipping data.
What if my store has no UCP?
That is the most common case today. The report marks the manifest as missing, still evaluates access, structured data and checkout, and shows the shortest path to publishing one.
Does it work on any e-commerce platform?
Yes. The analysis works on the published site with no integration. Shopify, WooCommerce, VTEX, Magento or a custom build are audited the same way.
Can I audit competitors?
Yes. Each domain includes 3 free competitor analyses per month. Turn on competitor mode, pick your own domain and analyze the competitor URL to compare pillar by pillar.
How often should I run it?
After every change to checkout, payment gateway or platform, and at least once a quarter: protocols and agent capabilities move fast.