Methodology

How we measure American trust in AI — transparently and reproducibly.

How the Trust Score Works

The USA AI Trust Score is a weekly composite metric (0–100) derived from thousands of public data points across social media, news, and search. Higher scores indicate greater public trust; lower scores signal concern or distrust.

5+ platforms
Data Sources
Reddit, HN, Search, News, Trends
Weekly
Update Frequency
Every 7 days, automated
300–500+
Data Points
Per weekly run

The 7 Research Agents

Reddit Crawler
Analyzes AI-related subreddits for sentiment, emotion scores, and recurring themes.
Hacker News Monitor
Tracks HN discussions about AI products, safety, and trust to capture the technical community perspective.
Brave Search Agent
Gathers search results and snippets for AI trust queries across diverse demographics.
Tavily Research Agent
Retrieves news articles, blog posts, and opinion pieces about AI trust and safety.
Google Trends Tracker
Monitors search trend volume for AI-related keywords to detect emerging concerns and interest spikes.
Sentiment Analyzer
Runs Claude AI-powered emotion and sentiment scoring on every collected data item.
Trust Score Engine
Aggregates all signals into a weekly trust score using a proprietary weighted model.

Scoring Model

Emotion Scoring: Each piece of content is scored across 6 emotions — trust, distrust, fear, excitement, curiosity, and skepticism — using Claude AI.

Sentiment Labels: Content is classified as positive, negative, neutral, or mixed, contributing to platform-level sentiment breakdowns.

Trust Score Formula: The composite trust score weights positive emotions (trust, excitement, curiosity) vs. negative emotions (distrust, fear, skepticism), normalized to a 0–100 scale.

Demographic Tagging: Content is tagged by inferred demographic group based on source community and context signals.

Confidence Level: Reported as a 0–1 value reflecting data volume and consistency. Low confidence means fewer data points were collected.

Limitations & Caveats

  • • This data reflects expressed sentiment in public online spaces — not a scientific random sample.
  • • Demographic attribution is inferred from community context, not self-reported.
  • • AI analysis may occasionally misclassify nuanced or sarcastic content.
  • • Low-volume weeks may show higher variance in trust scores.