Methodology
How SaidTrue Scores AI Answers
SaidTrue asks multiple AI engines a fixed set of memory and live-search questions about a local business. We compare answers with facts extracted from the business's own website, then score presence, accuracy, completeness, recommendations, and source quality.
We currently probe ChatGPT/OpenAI, Google Gemini, Perplexity and Claude because they represent almost all measured AI-chatbot usage: StatCounter's July 2026 worldwide AI chatbot share reports ChatGPT 77.89%, Google Gemini 9.93%, Perplexity 5.86% and Claude 3.20%, about 96.88% combined. Check StatCounter Global Stats.
For OpenAI memory and live-search probes, SaidTrue uses the current cost-efficient GPT-5.6 Luna model. That is a deliberate choice: it favours current platform behaviour over older free-tier defaults, and model IDs/prices are reviewed monthly because provider model names are perishable.
Scores are directional, not endorsements. They show what AI systems said during a specific scan, including uncertainty, stale sources, competitor recommendations, and possible confusion with similarly named businesses.
Public scorecards hide competitor names. Private reports will include the full evidence trail, raw quotes, citations, and suggested fixes.