AI Commerce Index
A 0–100 index shown prominently to merchants, with the underlying Readiness and Visibility components always visible instead of hidden.
For the curious
How Burnish actually works.
You don't need to read this page to use Burnish. It's here for the people who want to see the work behind the score — what we measure, what we change, and how we know it's paying off.
Honest movement, not fake movement
A single score that jumps the moment a fix is applied would be misleading. Burnish keeps the split visible so merchants know what changed now and what will show up later.
Rubric-based and merchant-actionable
Readiness is calculated directly from catalog state. It moves when merchants improve fields, structure, and data quality inside Shopify.
- Readiness is rubric-based and recalculates immediately after approved catalog fixes.
- The rubric is weighted toward merchant-controlled fields rather than vanity metrics.
- The goal is a score merchants can improve through real work inside Shopify.
| Factor | Weight |
|---|---|
| Metafield completeness | 20% |
| Description quality | 15% |
| Alt text coverage | 10% |
| Structured data / schema.org | 15% |
| Category / taxonomy correctness | 10% |
| Title optimization | 10% |
| Google Merchant Center readiness | 10% |
| Variant consistency | 5% |
| llms.txt presence | 5% |
Empirical and anchored in real sampling
Visibility is measured through sampled shopper-style prompts and model responses. It reflects how AI engines actually surface the merchant, not how good the catalog looks in isolation.
-
01
Generate shopper-style prompts grounded in the merchant catalog, category, and competitors.
-
02
Query official APIs across ChatGPT, Perplexity, Gemini, Claude, and Grok depending on tier.
-
03
Multi-sample responses to reduce non-deterministic noise.
-
04
Detect mentions against catalog-grounded brand and product context.
-
05
Normalize results into per-engine visibility measurements with confidence intervals.
Visibility is reported with confidence intervals so the merchant sees signal quality, not just a naked score.
Measurements are tied to model versions so merchants know when provider changes require a fresh baseline.
The methodology is designed to support explainability, including the ability to trace back into sampled queries and response evidence.
- Visibility changes lag behind fixes. Merchants should expect a 2–6 week delay as engines re-index.
- Visibility is a statistical estimate, not a promise of placement.
- Attribution is directional and operational, not a randomized controlled-trial claim.
- Only approved model providers are used, through commercial API agreements.
- Provider access is limited to the product and catalog data required for scoring and fix generation.
- Merchants are given clear disclosure around data handling and provider use before the workflow expands.
Benchmark publication
F1 publication is part of the product promise
Burnish is designed to publish brand-disambiguation benchmark results and refresh them over time. The production benchmark is still being finalized, so this page intentionally sets the framework first and the measured number second.
Leave your store if you want to follow the methodology as it matures alongside the product.