# AI Readiness Benchmark 2026: protocol before leaderboard.

> This page publishes the study design. It does not claim benchmark results that have not yet been collected, reviewed and calibrated.

- Canonical page: https://winin.ai/en/benchmarks/ai-readiness-2026/
- Language: English
- Last updated: 2026-09-04

## Research objective

Estimate the distribution of controllable AI readiness capabilities and observed AI representation outcomes across enterprises, while testing which capability measures remain reliable across sectors, locales and model changes.

## Planned strata

### Sector

B2B technology, professional services, finance, healthcare, manufacturing and consumer services, reported separately where evidence rules differ.

### Market

English and Simplified Chinese tasks with country effects recorded rather than silently pooled.

### Enterprise size

Size bands based on public operating scale, never inferred from website polish.

### Evidence maturity

From fragmented facts to governed, versioned evidence systems.

### Representation

Repeated task samples across multiple engines, dates and retrieval modes.

### Agent surface

Presence and governance of documented actions, APIs and machine-readable interfaces.

## Calibration design

1. **Pilot:** Test rubrics on a small set with known positive and negative controls.
2. **Dual review:** Measure inter-rater agreement and adjudicate material claims.
3. **Stability:** Repeat samples to estimate within-site and within-model variance.
4. **External validity:** Test whether capability measures predict later observed outcomes.
5. **Version release:** Freeze the method, publish limitations and preserve raw lineage.

## Participation and publication

Participation will require permission to test public properties, a fact owner for adjudication and agreement on anonymization. Commercial participation will never purchase a better score. Conflicts, sponsorship and exclusions will be disclosed with the release.

> **Status:** protocol design and pilot recruitment. Benchmark values are not yet available.

## Source

This Markdown document is the machine-readable counterpart of https://winin.ai/en/benchmarks/ai-readiness-2026/. The canonical HTML page remains the public presentation source.