LLMRank guide
How AI assistants choose which products to recommend
A practical guide to AI search visibility (GEO): what product data language models can actually use, how to structure it, and how to measure whether AI assistants can find and cite your catalog.
What an AI assistant can actually use
When a shopper asks an AI assistant for a product recommendation, the assistant can only recommend what it can retrieve and verify. In practice that means three things: a machine-readable catalog (an API, a feed, or a tool the model can call), complete and current fields on every item, and verifiable trust signals such as a real domain, clear policies and genuine reviews.
A beautiful storefront that only renders prices in JavaScript, or a catalog locked behind a login, is invisible to most AI shopping agents. The first step of generative engine optimisation (GEO) is not wording — it is exposure.
The fields that decide whether you get cited
Across AI shopping queries, the same fields decide eligibility again and again: a specific title, a plain-text description of at least a paragraph, a current selling price, a direct product URL, at least one image, brand, category, and live availability. Missing any one of these typically removes the product from consideration entirely, not just from the top spot.
Recommended fields — tags, attributes such as materials or ingredients, GTIN/SKU identifiers, shipping and returns policies, and review summaries — are what separate rank 1 from rank 5 once eligibility is met. LLMRank's completeness component scores exactly these eight required and eight recommended fields, so you can see precisely what is missing.
Freshness and availability are ranking signals, not housekeeping
AI assistants are penalised by their users when they recommend products that are out of stock or priced differently than quoted. Models and the tools behind them therefore prefer listings with recent updates and explicit availability. A feed that updates weekly will systematically lose to one that updates hourly, all else equal.
Keep price and stock synchronised automatically — a scheduled feed import or a push webhook — rather than relying on occasional manual edits.
Trust signals an AI can verify
Models weigh signals they can check: a verified domain, published shipping and returns policies, genuine review counts and ratings, and a consistent merchant identity across surfaces. Claims that cannot be checked — 'best quality', 'number one' — carry no weight and can hurt credibility.
Write descriptions as verifiable facts: quantities, certifications with the certifying body named, materials, test methods. If a fact cannot be traced to something checkable, leave it out.
How to measure AI visibility
You cannot improve what you cannot see. Run the queries your customers actually ask — 'waterproof jacket under 250', 'third-party tested cinnamon supplement' — against an AI search surface and check three things: does your product appear at all, where does it rank, and can you get an explanation for that rank?
LLMRank exists to make that measurement exact: every search returns a per-component score (relevance, completeness, freshness, availability, trust, performance) with the specific fix that would raise it, and the AI Query Simulator lets you test queries before your customers do.
Check your own catalog
Import a feed and see exactly which fields are blocking each product, with a per-component score and a drafted fix for every gap.