How ranking works

Every listing carries an LLMRank-Score between 0 and 100. It is a weighted sum of six components. Nothing else influences ranking: there is no paid placement, no hidden multiplier and no per-merchant adjustment. The weights below are the live weights, served from the same table the scoring engine reads.

LLMRank-Score =
  0.30 x Relevance
+ 0.20 x Completeness
+ 0.15 x Freshness
+ 0.10 x Availability
+ 0.10 x Trust
+ 0.15 x Performance

Relevance

weight 0.30

What it measures: Token and phrase overlap between the query intent and the listing title, description, tags, brand, category and attributes, plus filter fit (price band, category, brand).
How to raise it: Write the words buyers use in the title and first sentence of the description; fill in brand, category and structured attributes.

Completeness

weight 0.20

What it measures: Presence and quality of the fields an AI needs to answer confidently: description length, image, additional images, price, currency, brand, category, attributes, shipping and review summary.
How to raise it: Fill every empty field on the listing. The console lists the missing ones per listing.

Freshness

weight 0.15

What it measures: Days since the listing was last updated, on a decay curve. Under a week scores near full; several months scores near zero.
How to raise it: Push feed updates when anything changes, or use POST /api/v1/feed/update.

Availability

weight 0.10

What it measures: Availability state plus whether a real stock quantity is published. In stock with inventory beats in stock without.
How to raise it: Keep availability and stock_quantity in sync with your inventory system.

Trust

weight 0.10

What it measures: Merchant domain verification, published shipping and returns policies, listing certifications, and review count and rating.
How to raise it: Verify your domain in the console, publish policies, and add certifications.

Performance

weight 0.15

What it measures: Click-through rate on past AI answers that included this listing, smoothed for low impression counts.
How to raise it: Improve the fields agents show buyers — accurate price, clear title, real images — so the listing gets picked when shown.

Eligibility

A listing is only eligible for AI answers when it is marked searchable, has a title and a price (or a pricing model for services), and its merchant record exists. Ineligible listings are never ranked and never returned to agents.

Seeing your own scores

In the merchant console, every listing shows its component scores, the explanation behind each one, and a ranked fix list with the estimated point gain per fix. The AI Query Simulator runs a real query through the live ranking engine so you can see where you place and why. Programmatically, the same data is available from explain_ranking over MCP and GET https://thellmrank.com/api/v1/products/{id}?explain=your+query.

When scores change

Scores are recomputed when you edit a listing, when a feed import or incremental update touches it, and on a scheduled background pass that also refreshes freshness and performance. Each recomputation is stored as a score event, so you can see history rather than a single moving number.