How ready is your store for AI understanding? Get your free AI Commerce Score™ in 10 seconds and see the 8 factors behind it.
AI Commerce Score measures AI readiness and understanding signals. It does not predict recommendation frequency. Recommendation behavior is measured separately through Recommendation Intelligence.
Every store lands in one of four bands. The band shows how confidently AI systems can read, parse, and trust the store's data, not whether AI will recommend it in practice. See Recommendation Intelligence for that.
AI vision systems now read product images, not just text. GPT-4o Vision and Apple Intelligence can see what is in a photo and compare it against the surrounding text. A store that tells AI its product is a black leather boot but has a generic alt tag reading image_322.jpg is failing this factor silently. The AI eye sees the product. The AI brain cannot match it to anything.
We send your product images to GPT-4o Vision and compare its semantic description against your actual alt text. If the image shows a black leather Chelsea boot and your alt attribute says product-image-4, that gap costs you points. AI cannot confidently surface a product it cannot semantically verify.
We check whether critical information is baked into images as graphics rather than HTML text. If your 20 percent sale badge is a PNG overlay, AI crawlers without Vision API cannot read it. That information is invisible to most AI systems.
We verify that what appears in the product photo matches what the HTML description says. Color mismatches, size discrepancies, or model photos that do not reflect the actual product all reduce AI confidence in the accuracy of your data.
Structured data is the primary language AI agents use to understand what a store sells, at what price, and how trustworthy it is. JSON-LD schema is not optional in 2026. It is the difference between being in the AI Commerce Graph and being outside it. Most stores we have scanned are missing critical schema markup.
We validate the presence and correctness of Product, Offer, and AggregateRating JSON-LD markup. Missing or invalid schema means AI agents cannot reliably extract product details, pricing, or social proof data.
We check for Organization or Store schema that connects your website to your Persistent Merchant Identity. This is how AI systems build a consistent understanding of your brand across platforms and over time.
We check for a valid llms.txt file in your domain root. This is the file that tells AI agents what your store does, what it sells, and how to navigate your product catalog. Every Shopify store now has this file via Agentic Dashboard, but quality varies enormously.
A technically broken store is an invisible store. AI crawlers face the same blockers as Googlebot, plus a few new ones specific to LLM agents. Cloudflare WAF blocking GPTBot, LCP over four seconds, a DOM with 400 nested divs and no semantic hierarchy. All of these are scoring against you right now.
We check whether your robots.txt and any WAF configuration block or restrict known AI crawlers including GPTBot (ChatGPT), ClaudeBot (Claude), and AppleBot (Apple Intelligence). A blocked bot cannot read the store at all.
We pull LCP (Largest Contentful Paint) and CLS (Cumulative Layout Shift) via the Google PageSpeed API. Target is LCP under 2.5 seconds and CLS under 0.1. Slow or unstable pages break AI agent navigation flows and reduce crawl depth.
We analyze the heading hierarchy (H1 through H3), the ratio of semantic HTML to div soup, and the overall cleanliness of the rendered DOM. AI agents use heading structure to build a map of page content. A missing H1 or a broken hierarchy means the map is wrong.
When an AI agent is about to route a buyer to your store or complete a purchase autonomously, it runs a trust check. Not on design, not on visual badges. On machine-readable evidence that the transaction will go smoothly and the buyer will not be left with a problem. A large share of stores fail this check.
We check whether your return policy and terms of service are in crawlable HTML text, not locked inside a JavaScript modal or a PDF. AI agents need to read these directly. If they cannot, they treat the transaction as higher risk and reduce recommendation confidence.
We test whether the cart and checkout flow is accessible to autonomous agents. Agentic Storefronts require a clean, machine-navigable checkout path. Popup interruptions, mandatory account creation, and broken cart APIs all reduce this score.
We check whether the company behind the store is clearly identified in the footer or contact pages, with a readable business name, address, and contact method. AI systems cross-reference this against external databases to verify the merchant is legitimate.
AI agents are starting to complete purchases. When an agent tells a buyer it found the perfect product at 89 dollars with free shipping and the checkout shows 109 dollars with a delivery fee, that agent has failed the buyer. AI systems are learning to avoid stores that create this kind of gap. Many stores fail commerce accuracy checks.
We compare the price visible in your HTML against the price in your Google Shopping XML feed or Shopify API. Discrepancies of any size are flagged. AI agents that surface a wrong price to a buyer lose trust immediately and that trust loss is remembered.
We check whether your availability signals are accurate across your feed, your schema markup, and your live page. If your feed says in stock and your page says available in 14 days, AI gets contradictory data and reduces its confidence score for your store.
AI agents do not match stores to keywords. They match stores to buyer intent. When someone tells Apple Intelligence to find a gift for their dad who loves hiking under 60 dollars, the AI is looking for a store whose content maps semantically to that specific, contextual request. Generic copy cannot match specific intent. This is where most stores lose AI visibility.
We use semantic embedding comparison to measure how well your product descriptions and category copy cover the actual language buyers use in LLM search prompts. Use-case descriptions, context phrases (good for cold weather, perfect for marathon training), and gift context signals all matter here.
There is a difference between appearing in an AI response and being confidently recommended. A store that AI hedges on gets a mention. A store that AI trusts is easier for AI to surface with confidence. Recommendation Confidence measures the social proof signals that AI systems can actually parse and analyze, not just see.
We check for AggregateRating schema with a real ratingValue and reviewCount from a verified review platform. We also run sentiment analysis on accessible review content. AI systems analyze sentiment to predict post-purchase satisfaction, which feeds directly into how confidently they recommend a store.
AI systems are trained on the entire internet. They know what people say about brands outside of brand-owned channels. A store that has no Reddit mentions, no press coverage, and no citations in AI indexes is treated as unverified. Currently this factor carries 5 percent weight. Based on how quickly LLM training data is expanding, we expect this to be among the most important factors by 2027.
We check for brand mentions and citations in external LLM indexes, Reddit discussions, Perplexity citations, and AI-readable databases. A brand that appears consistently across external sources has a stronger Persistent Merchant Identity in the AI Commerce Graph.
Free, no signup, results in 10 seconds.
Illustrative example · single-site signal for atomfoundry.dev.
View full signals →