The AFR standard

The three-layer model

The total is 100 = 35 + 40 + 25. Legibility carries the most weight because machine-readable content is what AI-friendliness actually means: a site that is technically open but unreadable still fails on AI usefulness.

Every sub-check

Grouped exactly as the standard defines them. Points are fixed; nothing here depends on opinion, popularity, or payment.

Grade thresholds

Current thresholds: A ≥75 / B ≥ 60 / C ≥ 42 / D ≥ 25, calibrated on 2026-11-10 against the first Top-1000 scan distribution. After that, thresholds only move at quarter boundaries following a distribution review, and historical scores keep the engine version they were computed with.

Strategy tags

A second, independent axis derived from robots.txt alone. The tag sits next to the grade and never deducts points — it explains the policy behind the score.

• AI-Open Allows every AI crawler: search, hybrid, and training.
• Middle-Path Allows search-type crawlers, blocks training crawlers.
• Training-Open Allows training crawlers while restricting search-type crawlers.
• AI-Closed Blocks all AI crawlers site-wide.
• Unconfigured No AI-specific rules in robots.txt; protocol defaults apply.

Engine versions

Every scan records the engine version that produced it, so any score on this site can be reproduced bit-for-bit later.

Every snapshot stores engine_version; historical scores are never re-graded with a newer engine — they stay attached to the version that produced them.

Where the written standard and the engine differ, the engine version recorded on each scan governs recomputation.

Recompute any score yourself

The scoring engine is a pure function: no IO, no clock, no network. Clone the repository, run the test suite, and feed a recorded snapshot back into the engine.

1. Clone and run the engine test suite (Node ≥ 22, zero dependencies):

git clone https://github.com/AFR-IM/AFR-web
cd AFR-web
npm test   # = node --test packages/*/test/*.test.ts

2. Feed a recorded snapshot ({ signals, afr } per site) back into the engine:

import { scoreAfr } from './packages/afr/src/index.ts';
import snapshot from './scans/github.com.json' with { type: 'json' };

const result = scoreAfr(snapshot.signals);
console.log(result.score, result.grade, result.engineVersion);

Neutrality pledge

Paying never changes a score.

Read it verbatim — this is the promise the product is built on.

• Paid placements never change ranking or grades. The natural list is always a pure AFR-score descending sort; money buys exposure in a separate Sponsored block, never a better position in the natural order and never a better-looking grade.
• Every sponsored slot shows the site's real grade ring and score — an F-grade site can buy exposure, but it cannot buy an A-grade appearance.
• Ads live in visually separated blocks with a permanent "Sponsored" label, never mixed into natural results, and hovering an ad never changes a grade color.
• Verification, badges, evidence chains and score-change notifications stay free forever. Nothing in the free tier is throttled to force an upgrade.

Scores cannot be bought: payments never affect ratings.

Data sources

Seed set

Tranco daily Top 1M (tranco-list.eu, free for academic use with attribution).

Scan evidence

Public snapshots committed to github.com/AFR-IM/AFR-data — git versioning is the audit trail.

Corrections

Pull requests to the data repository (corrections/) trigger a re-scan; verified owners may publish a statement.

Machine-readable standard

The whole standard is served as JSON at /api/v1/method (layers, per-sub-check points, grade thresholds) so tools can consume it without scraping this page.

Open /api/v1/method

AFR v1.1 is published under CC BY 4.0 — reuse it with attribution.