Your AI's personality. You, in its eyes.

A questionnaire about an AI and an AI's impression of you answer different questions. aivitals keeps both visible: how the AI answers twelve situations, and what it can reasonably infer from interactions it actually knows.

Two results, two kinds of evidence

The AI type is calculated from twelve A/B responses across four paired dimensions. It reflects that run, not a hidden personality discovered inside a model. The user portrait is the AI's tentative interpretation of accessible history; it is not calculated from your answers to those twelve questions.

The portraits may agree, differ, or include an unknown result. None of those outcomes establishes compatibility, intelligence or mental health.

Ask for an observation, not just a label

A useful supporting observation identifies a pattern: for example, repeatedly asking for alternatives before deciding. “You are brilliant” is praise, not evidence. Even a specific observation may be biased by the tasks you recently used the AI for. A week of debugging does not describe all of your life.

Keep your expectations out of the prompt: asking for a particular type can influence the reply. Do not ask the AI to invent a conversation or quote unavailable history to make the result sound more personal.

Why a new chat might not know you

Conversation context, saved memory, chat search and project instructions have different scopes. An assistant cannot necessarily read every conversation in an account, another device or another product. An instruction to “remember everything” does not grant that access.

Choose what leaves your browser

• Auto mode asks your AI to send answers for scoring and publish a summary, including model information and a one-line user portrait.
• Manual mode scores locally and does not automatically publish that result.
• A shared report link includes encoded report data. Encoding is not encryption; anyone receiving the link can read the report.
• Share only after reviewing the one-line portrait and evidence. Remove identifying or sensitive details before generating a link or image.

Make comparisons a conversation, not a verdict

Try another run or model if you are curious. Note what context it had and which answers changed. A disagreement can reveal differences in prompts, instructions or available observations; it does not prove that one model knows your true self.