Confidence is not accuracy
ChatGPT is designed to produce fluent, helpful-sounding text. It does not signal uncertainty the way a human expert would. When it does not know something, it does not say "I'm not sure" -- it generates a plausible-sounding answer with the same tone and structure as a correct one. This makes it impossible to judge reliability by reading the output alone.
In October 2025, OpenAI updated its usage policies to ban personalised professional advice across legal, medical, and financial domains. This was not a hypothetical concern -- it was a response to documented cases where users relied on AI outputs that turned out to be fabricated. The model itself has no mechanism to verify its own claims.
How to actually check
Blackbird Scope scores every AI response across three metrics: Fidelity (does the response match verifiable sources?), Precision (are specific claims, names, dates, and figures accurate?), and Recall (what relevant information has the AI left out?). These scores expose the gap between how confident an answer sounds and how accurate it actually is.
The browser extension analyses ChatGPT conversations in real time. The Blackbird Scope lets you paste any AI interaction for a full breakdown -- flagging hallucinated citations, unsupported statistics, and missing context that the model silently omitted. Every flag links to the methodology behind it. No black boxes.