Feedmind
How it’s scored →
FM-score methodology v1.0 · July 2026

How the Feedmind Score is calculated

Shopping is moving into assistants, and assistants read product content differently than people do. The Feedmind Score measures one thing: how well a catalog can be read, understood and answered from by an AI agent. We publish the method in full, because a score nobody can audit is a score nobody should trust.

++01 · Dimensions

Four dimensions, weighted.

Agent accessibility
25%
Structured data
25%
Content answerability
30%
Multi-language parity
20%
Agent accessibility
25%
Agent accessibility
Can an assistant reach and read the page at all? We check crawler permissions, whether content survives without JavaScript, and whether the page states what it is before a human ever scrolls.
EXAMPLE CHECKS
robots.txt — GPTBot, PerplexityBot, Google-Extended
Server-rendered content without JS execution
A single, clear H1 and a declared document language
Structured data
25%
Structured data
Machine-readable facts beat prose. Product and Offer markup with real identifiers is what lets an assistant match your item to the one a shopper is asking about, and quote a price it trusts.
EXAMPLE CHECKS
schema.org Product + Offer JSON-LD validity
GTIN / MPN / brand coverage across SKUs
Price, currency and availability freshness
Content answerability
30%
Content answerability
The heaviest weight, because it decides recommendations. We measure whether your copy contains the specifics buyers ask about — fit, compatibility, materials, what is in the box — and whether those facts survive summarization.
EXAMPLE CHECKS
Description depth and attribute density
FAQ content and FAQPage markup
Buyer-question answer rate in simulation
Multi-language parity
20%
Multi-language parity
European catalogs answer in several languages. A thin German page is a German shopper you lose silently, so we compare depth, attributes and markup across every locale you publish.
EXAMPLE CHECKS
Word-count and attribute parity per locale
hreflang correctness and locale routing
Localized structured data, not just copy
Weights sum to 100%.
++02 · Two layers

A rule engine you can argue with, and a simulation you can’t.

LAYER 01 · RULE ENGINE
~15 deterministic rules, whole catalog
Markup validity, identifier coverage, crawler access, JavaScript dependency, description depth, FAQ presence, locale parity. Instant, free, ungated.
LAYER 02 · AI SIMULATION
Buyer questions + summarization survival
A model is given only what an agent can read, then asked real purchase questions. We also test what survives when your page is summarized to three sentences.
++03 · Grade scale

From 0–100 to a letter.

GRADEBANDREADING
A85 — 100Agent-ready. Assistants can answer almost anything about the catalog.
B65 — 84Solid. Gaps appear on comparison and compatibility questions.
C50 — 64Readable but thin. Frequently skipped for better-documented rivals.
D35 — 49Structural problems. Most buyer questions go unanswered.
Fbelow 35Effectively invisible to assistants.
Each band splits in thirds: the bottom third takes a −, the top third a +. F is terminal.
++04 · Benchmarks

Why every score carries a comparison.

A category median is the anonymized aggregate of every store we have scanned in that category. No individual shop’s data is exposed by the median, and a category needs at least 20 scanned stores before we publish one. The named competitor is a public scan of a public storefront, run with exactly the same rules as yours — if someone scans you, you appear in their comparison the same way.

++05 · What it is not

What the score is not.

It is not a guarantee of placement. No one outside the model providers can promise that a change makes an assistant recommend you, and anyone who does is selling something.

It is not attribution. We do not measure traffic or revenue from assistants; we measure whether your content can be used at all. Readability is a precondition, not a conversion metric.

It is not a ranking. Two stores with the same score are equally readable; which one gets named depends on price, availability, reviews and the shopper’s question.

© 2026 Feedmind · Luxembourg, Grand Duchy of Luxembourg