Product Recommendation Methodology

Last updated July 15, 2026

Recommendations are produced in two clearly separated stages.

1 · Need classification

Article context (or, in the planner, your selections) is converted into a structured need profile: kitchen zone, product category, style, rough size constraints, budget, installation complexity, renter suitability, and any exclusions. Photo-based classification is not enabled today; when it is, the model will only classify visible needs — it will never pick products or invent details.

2 · Deterministic ranking

A deterministic engine ranks shopping categories by kitchen zone, topic, style, budget, renter suitability, and DIY complexity, and applies a relevance floor so nothing off-topic can surface. On pages defined by a look rather than a single zone — a color scheme, a style, a decor theme, or a layout — the categories come from an editor-curated set for that page, so a coffee-theme page shows coffee-station gear, not random cabinet pulls. Each recommendation explains why it is shown and what to measure or check first, and never invents dimensions. Product-level matching against an approved internal catalog — adding color/finish, exact dimension fit, data freshness, and editorial-approval ranking — is built but not enabled today; until it is, we link to Amazon searches rather than specific products.

Integrity guarantees

We link to Amazon searches built from fit criteria — not specific products or ASINs. You choose from current listings yourself.
No fabricated prices, ratings, reviews, availability, Prime status, dimensions, or “best seller” claims. Specific product data would appear only from an approved, compliant Amazon source (PA-API), which is not enabled today.
Each page only surfaces shopping criteria relevant to its topic — a cabinet-hardware guide never shows shelving, for example.
Development or sample data used while building the site is kept off public pages and is never presented as real product data.