What the AI analyst sees

The AI briefing is generated by Google Gemini. We keep the prompt deliberately narrow: the model sees aggregated numbers, not people.

What the model receives

• Department-level averages for each metric in the current period (mood, stress, sleep, energy, vibe score).
• The same averages for the previous comparable period, for delta calculations.
• Response rates per department.
• Survey question metadata — the question text itself, so the narrative can reference it.

What the model never receives

• Individual responses, names, emails, or user IDs.
• Open-text answers from surveys or journal entries.
• Raw timestamps that could identify a specific submission.
• Any data from teams with fewer than 5 responses in the period.

Why a narrow prompt matters

Passing only aggregated numbers protects anonymity even if inference traffic were ever compromised. It also keeps the output grounded — the model can't 'hallucinate' a quote from a teammate because it never had one.