What the AI found
“The AI revealed that her three worst-sleep nights (averaging 5h 12m) consistently followed late-evening work emails—not her pre-bed reading habit, as she had assumed.”
Before
Fragmented Sleep Data, Zero Review
After
One 7-min Actionable Review
The same system, three states — real screens, not a screenshot
| Date | Sleep Duration (Oura) |
| 2024-03-01 | 7h 03m |
| 2024-03-02 | 6h 45m |
| 2024-03-03 | 5h 15m |
Prompt
Analyze the attached Oura sleep data (CSV) and my Apple Health subjective notes for the last 7 days. Identify any significant correlations between sleep duration/quality and daily events I noted, especially focusing on differences between nights with low sleep scores (under 70 Oura readiness, or subjective 'poor' in Apple Health) and high sleep scores (over 85 Oura readiness, or 'excellent' in Apple Health). My subjective notes often mention 'late work', 'evening exercise', or 'reading before bed'.
Analyze the attached Oura sleep data (CSV) and my Apple Health subjective notes for the last 7 days. Identify any significant correlations between sleep duration/quality and daily events I noted, especially focusing on differences between nights with low sleep scores (under 70 Oura readiness, or subjective 'poor' in Apple Health) and high sleep scores (over 85 Oura readiness, or 'excellent' in Apple Health). My subjective notes often mention 'late work', 'evening exercise', or 'reading before bed'.
AI
Your three worst-sleep nights this week, averaging 5 hours and 12 minutes of sleep duration, consistently followed evenings where you noted 'late work' activity (specifically, responding to emails past 9 PM). This correlation was stronger than the impact of 'evening exercise' or 'reading before bed', which showed no statistically significant negative association with sleep duration or quality.6h 38m
Avg Sleep Duration (last 7 days)
Late Evening Work (post 9PM)
Key Sleep Disruptor Detected
3 of 7 nights
Nights Impacted by Disruptor
Weekly Sleep Review: From Scattered Data to Sharp Insights
A busy professional transforms a fragmented sleep tracking routine into a clear, actionable weekly review using AI.
A 48-year-old marketing director, Northern Europe
Tools used
The real tools used here — swap any for your own equivalent. Each links to how we’d set it up.
Starting state
Before anything was set up
Before implementing the AI-guided review system, Sarah’s sleep tracking was a collection of unexamined data points. She wore an Oura ring nightly, and occasionally logged her subjective sleep quality in Apple Health, but the information remained siloed. Without a dedicated process for synthesis, these individual metrics offered little actionable insight, leaving her to guess at connections between her daily routines and sleep quality.
| Date | Sleep Duration (Oura) |
| 2024-03-01 | 7h 03m |
| 2024-03-02 | 6h 45m |
| 2024-03-03 | 5h 15m |
| 2024-03-04 | 7h 20m |
| 2024-03-05 | 5h 10m |
Working state
Resources, doing its job
The transformation began with Sarah feeding her sleep data into Gemini. She exported Oura data as a CSV and manually entered subjective notes. Her prompt directed the AI to analyse patterns, focusing on identifying correlations between her daily activities, perceived stress, and biometric sleep scores. This specific, data-rich input allowed Gemini to pinpoint non-obvious connections, such as the true impact of late-night work.
Prompt
Analyze the attached Oura sleep data (CSV) and my Apple Health subjective notes for the last 7 days. Identify any significant correlations between sleep duration/quality and daily events I noted, especially focusing on differences between nights with low sleep scores (under 70 Oura readiness, or subjective 'poor' in Apple Health) and high sleep scores (over 85 Oura readiness, or 'excellent' in Apple Health). My subjective notes often mention 'late work', 'evening exercise', or 'reading before bed'.
Analyze the attached Oura sleep data (CSV) and my Apple Health subjective notes for the last 7 days. Identify any significant correlations between sleep duration/quality and daily events I noted, especially focusing on differences between nights with low sleep scores (under 70 Oura readiness, or subjective 'poor' in Apple Health) and high sleep scores (over 85 Oura readiness, or 'excellent' in Apple Health). My subjective notes often mention 'late work', 'evening exercise', or 'reading before bed'.
AI
Your three worst-sleep nights this week, averaging 5 hours and 12 minutes of sleep duration, consistently followed evenings where you noted 'late work' activity (specifically, responding to emails past 9 PM). This correlation was stronger than the impact of 'evening exercise' or 'reading before bed', which showed no statistically significant negative association with sleep duration or quality.Use case implemented
The finished system, running on its own
Now, Sarah Dedicates seven minutes each Sunday to reviewing her sleep. Gemini summarises the week, highlighting key trends and surfacing previously hidden influencing factors. This automated analysis provides her with a clear, evidence-based understanding of her sleep patterns, enabling her to adjust her evening routine proactively. The system has moved her from passive data collection to active, informed self-management.
6h 38m
Avg Sleep Duration (last 7 days)
Late Evening Work (post 9PM)
Key Sleep Disruptor Detected
3 of 7 nights
Nights Impacted by Disruptor
What an outside observer would notice
from 30 mins to 7 mins/week
Time spent on sleep review
from assumption to evidence-based
Identification of root cause for poor sleep
from none to specific
Actionable insights for routine adjustment
The stack — build it yourself
Provides reliable, continuous biometric sleep data.
Convenient for quick, qualitative notes about daily activities.
Translates raw data and notes into specific, actionable insights, identifying non-obvious patterns.
These are the tools used in this story. Any can be swapped for an equivalent you already trust.
Go deeper
Do this yourself
See Resources
This story runs on Resources. The tools and prompts above are the real build — swap any tool for your own equivalent and follow the same steps.