The service sends Gemini chat text and writing instructions. The current implementation does not train a separate relationship diagnostic model or run a quantitative assessment.
The writing instructions
The request asks for a nickname, a short comment about each participant and an overall summary. Mild mode emphasizes positive aspects; spicy mode asks for criticism. The requested output language follows the chosen page language.
Information the request cannot provide
The tool does not know the participants’ actual personalities, offline relationship or read receipts. Timing is also removed during preparation. Confident background statements from AI are not facts independently checked by the service.
Limits of short output
A long exchange can be compressed into one or two phrases. Counterexamples may be omitted and speakers may be confused. Try a short exchange with consistent aliases and compare each comment with the text.
Failure versus a bad interpretation
A failed request may involve connectivity, provider availability or input format. A successful response can still contain wrong information. For an error message, retry or contact support; for an overconfident interpretation, review the source. Fluency is not verification.