# A research framework before any result is claimed.

> Xindar's public measurement specification defines the fields required for repeatable AI visibility research. It contains no approved observations or performance claims.

Canonical URL: https://www.aixindar.com/research-methodology

Last reviewed: 2026-09-02

## Summary

A versioned, brand-neutral specification for AI visibility sampling, prompt panels, source IDs, scoring, limitations, and citation.

## Status

**Template only — no approved observations.** This specification contains no collected responses, scores, citation uplift, recommendation uplift, or commercial result. External platform collection is not authorized by this file.

## Research question

How consistently do selected AI answer engines identify, describe, compare, cite, and qualify an entity for a defined category, audience, and market?

## Sample

One engine response to one frozen prompt in one declared market-language context.

Inclusion rule: Include every completed response returned during the declared collection window, including responses with no entity mention.

Exclusion rule: Exclude interrupted, policy-blocked, empty, or technically failed responses and record each exclusion with a reason.

## Brand-neutral prompt panel

### Q-CATEGORY-001 — category_discovery

Which providers should a [buyer role] consider for [category] in [market], and what evidence should they compare?

### Q-CRITERIA-001 — selection_criteria

What criteria should a [buyer role] use to evaluate a [category] provider for [use case] in [market]?

### Q-COMPARISON-001 — provider_comparison

Compare credible approaches to [use case] for [buyer role] in [market]. Include limitations and verifiable sources.

### Q-EVIDENCE-001 — evidence_verification

What public evidence would verify that a [category] provider can support [use case] in [market]?

### Q-RISK-001 — risk_review

What risks, exclusions, or unsupported claims should a buyer check before selecting a [category] provider?

### Q-IMPLEMENTATION-001 — implementation

What implementation steps and evidence are required for a [buyer role] to adopt [category] for [use case]?

### Q-REGIONAL-001 — regional_fit

How should requirements for [category] differ between [market A] and [market B]?

### Q-ALTERNATIVES-001 — alternatives

What alternatives exist to [approach] for [use case], and when is each option appropriate?

## Engines, models, region, and language

Must be selected before collection and recorded exactly as shown by the product or API. No engine or model has been approved for this template yet.

Must be declared per run. Do not merge markets or languages into one score without a documented aggregation rule.

Collection date: Not set.

Planned repeats per query: 3. No repeats have been collected.

## Scoring method

- **entityMention:** 0 = absent; 1 = present.
- **descriptionAccuracy:** 0 = materially wrong or unsupported; 1 = mixed or incomplete; 2 = supported by approved entity facts.
- **recommendationPosition:** Record ordinal position only when the answer presents an explicit ordered recommendation. Otherwise record null.
- **citationPresence:** 0 = no inspectable source; 1 = at least one inspectable source.
- **citationSupport:** 0 = source does not support the adjacent claim; 1 = partial support; 2 = direct support.
- **qualificationCoverage:** Count declared buyer criteria that the answer addresses with evidence; store the denominator separately.
- **riskDisclosure:** 0 = relevant limitation omitted; 1 = limitation mentioned; 2 = limitation linked to decision impact or evidence.
- **aggregationBoundary:** Do not publish a composite score until weighting, missing-value handling, confidence intervals, and minimum sample size are approved.

## Source IDs

SRC-[collection date YYYYMMDD]-[query ID]-[engine code]-R[repeat number]

## Limitations

- AI answers can change across time, sessions, products, model versions, account context, and geography.
- A fixed prompt panel measures only the declared questions; it does not represent every buyer query.
- Citation presence does not prove that a source supports a claim; support must be checked separately.
- Entity mention and recommendation are observations, not proof of commercial impact.
- This specification contains no collected responses, scores, uplift claims, or approved research result.

## Version

1.0.0. Any change to a frozen field creates a new study version and must not be silently combined with earlier observations.

## How to cite

Xindar. AI Visibility Query Panel and Measurement Specification, version 1.0.0. Template status: no approved observations. https://www.aixindar.com/research-methodology