Ask a Friend logo

The Research Repository for the Consumer AI Validation Analysis

Published protocol, registered methodology, and demonstration artifacts for a systematic study of how AI performs on the questions consumers actually ask — across legal, insurance, home buying, consumer protection, remodeling, and health.

Read the ProtocolView the Paper

Universal Core Framework v1.0

The methodological foundation — defining six evaluation dimensions, anchored scoring rubrics, study classification rules, and reproducibility standards applicable across any consumer-facing AI domain.

Consumer Guidance Reference Implementation

The registered study protocol — 270 prompts across six consumer domains, five leading LLMs, dual-track expert and automated evaluation, and five pre-registered hypotheses awaiting data collection.

GitHub Repository

The open repository for all study materials — prompt library with difficulty metadata, analysis scripts, scoring instruments, and anonymized dataset — released to support replication and independent review.

The Study Is Underway — Results Coming

The 270-prompt evaluation is in active development. When data collection is complete, this site will publish domain-specific performance scorecards, scored Q&A artifacts across all six domains, and the full anonymized dataset — openly, with no paywalls. See an example artifact here.
For commercial applications of the CAVA evaluation framework, visit CAVASuite.com

responsible-ai-badge-square-300x300

Support the research

This project receives no funding from AI companies. Independence is foundational to the methodology — an evaluation funded by the companies it evaluates cannot produce findings that consumers, regulators, or the research community can trust.

If you believe independent AI evaluation matters, there are several ways to help — from micro-grants to institutional sponsorship.

AskAFriend.com provides general educational information only and is not a substitute for professional legal advice.

Github Repository