metzia ranks the products in its catalog. This page states what the ranker reads, and what it never reads.
metzia earns a commission from the seller. Read the affiliate disclosure for what that means for you. The commission does not change the order of the results.
What the ranker reads
The ranker reads the structured fields a seller declares about a product, and the outcome signals the platform computes from purchases. It reads no prose. A seller writes a summary and a description, and neither one orders a result.
category. The product category the buyer asks for. A product in the requested category scores, and a product in another category does not.pricing_model. The way a product charges: a flat fee, per seat, per usage, or tiered. A product whose pricing model matches the buyer's request scores, and another model does not.price_points. The prices a product lists. The buyer names one unit and the most they will pay for it each month, and a product whose price for that unit is lower scores higher. A yearly price counts as one twelfth of itself, so it compares with a monthly one.limits. The limits a product declares, such as seats or storage. The score is the fraction of the buyer's minimums the product meets.delivers. What the buyer receives when the purchase completes: seats, credits, an API key, a license, access, or a shipment. A product that delivers exactly what the buyer asked for scores.integrations. The integrations a product declares. A product scores only when it lists every integration the buyer asked for.provisioning. How much work the buyer does to start using the product, ordered delegated, then keyless, then claim. Less buyer work scores higher.not_a_fit_when. The seller's answer for a buyer the product does not fit. A confirmed match scores zero, a confirmed non match scores one, and an unknown answer scores zero. The ranker reads the structured answer and never the prose.outcome_refund. The seller's refund rate over the latest 90 day window. A seller whose buyers refund less than the average seller scores above zero, and a seller with too little volume takes the neutral value of the average seller.outcome_cancellation_30d. The seller's rate of purchases cancelled within 30 days, over the latest 90 day window. A seller cancelled less often than the average seller scores above zero, and a seller with too little volume takes the neutral value of the average seller.outcome_repeat_purchase. The seller's rate of buyers who purchase again, over the latest 90 day window. A seller whose buyers return more often than the average seller scores above zero, and a seller with too little volume takes the neutral value of the average seller.outcome_oversell. The share of the seller's purchases it could not supply, over the latest 90 day window. No writer computes it yet, so it stays neutral for every seller.outcome_pick_kept. The seller's rate of buyer reports that the buyer kept the pick, over the latest 90 day window. A seller kept more often than the average seller scores above zero, and a seller with too little volume takes the neutral value of the average seller.
What the ranker never reads
A commercial signal never orders a result. The list below names each signal we exclude. The rule is the same for all of them. The signal never affects retrieval, eligibility, or position.
seller_plan. The seller's plan: free, pro, or scale. A paid plan buys product capability and never placement. The plan is not a ranking input, and it never affects retrieval, eligibility, or order.affiliate_commission_rate. The percentage or fixed amount a seller pays on an affiliate sale. The rate is not a ranking input, and a higher rate never buys retrieval, eligibility, or a higher position.referral_commission_rate. The flat commission on an assisted referral. The rate is not a ranking input, and it never affects retrieval, eligibility, or order.
The same question gives the same answer
Each ranked answer names the ranking version that produced it. metzia records the question, the version, and the order. We can replay the record and reproduce the answer.
How this list changes
A change to the input list is a new ranking version. The old version keeps the order it produced, so a recorded answer stays reproducible.
The two lists above are the text of this policy at this version. A change to either list is a new version of this document, and the version history keeps the list that was in force on a date.
GET /api/public/ranking-inputs returns the same two lists and the active
ranking version, for a reader who wants them as data.
Questions
Write to support@metzia.ai with a question about this document.