Why generic sleep advice doesn't work
"Go to bed at the same time every night." "Avoid screens an hour before bed." "Keep the room cool." I've read these in approximately forty articles. I know them. Knowing them hasn't helped, because knowing generic advice and knowing which of these things is actually affecting your specific sleep are completely different problems.
What makes AI useful here isn't that it knows more about sleep than a doctor or a good article. It's that when you give it your actual patterns — your specific bedtimes, wake times, what happened that day, how you felt in the morning — it stops giving you generic advice and starts asking questions about the variables that are actually relevant to you.
It took about two weeks of data before the AI started saying things that felt genuinely specific rather than recycled. That's the honest timeline. Don't expect personalised insight from day one.
The manual tracking method — no wearable needed
The data you need is simpler than you'd think. Four numbers and two observations, noted every morning. Takes about 60 seconds.
| Field | What to note | Why it matters |
|---|---|---|
| Bed time | When you actually got into bed (not when you intended to) | Consistency is the single biggest lever on sleep quality |
| Lights out time | When you stopped looking at your phone/book and closed your eyes | Different from bed time — the gap is usually revealing |
| Wake time | When you actually woke up, not your alarm time | Natural wake time tells you a lot about whether your total sleep was sufficient |
| Morning feel | 1–5 score, quickly. 1 = groggy and slow, 5 = alert from the start | This is the actual outcome you're trying to improve |
| One variable | Anything notable from yesterday — late meal, alcohol, exercise, stress, screen time, caffeine timing | This is what the AI will use to find correlations |
| Sleep quality guess | 1–5. Did you feel like you slept well, regardless of duration? | Helps distinguish between short-but-good and long-but-poor sleep |
I kept this in a simple note on my phone — one line per day, same format every morning. After two weeks I had enough data to paste into a prompt and get something useful back.
The AI sleep coach prompt
Run this after you have at least 7 days of data. Paste your log as plain text.
Here's my sleep log for the past [X] days. Each line is: date | bed time | lights out | wake time | morning feel (1-5) | sleep quality (1-5) | notable variable [paste your log here] Based only on this data: 1. Do you notice any pattern between the "notable variable" and morning feel or sleep quality? 2. Is my bed time consistent or irregular — and does it seem to matter? 3. What's the gap between lights-out and bed time telling you? 4. Suggest one specific micro-change to try this week — just one, not a list. Don't give generic sleep advice. Only comment on what you actually see in the data I've provided.
Without that constraint, AI will generate a list of six improvements. Implementing six things at once means you won't know which one actually worked. One change per week lets you isolate cause and effect — the same principle behind any decent experiment.
The weekly sleep review prompt
After four weeks you have enough data to see real patterns. This prompt is for that point.
Here's four weeks of my sleep log: [paste full log] Look across the whole month and tell me: - Which days of the week consistently have the worst morning feel? - Is there any variable that appears before my worst nights? - Is there any variable that appears before my best nights? - Has any pattern changed over the four weeks — is sleep getting better, worse, or staying flat? Give me one conclusion I can actually act on. Not a summary — a specific hypothesis I can test.
What actually moved — and what didn't
After 30 days of tracking and weekly AI reviews, here's what genuinely made a difference for me and what didn't:
What worked: consistent wake time, not consistent bed time. I'd always heard "go to bed at the same time". The AI looked at my data and pointed out my wake time varied by up to 90 minutes on weekends, while my bed time was actually fairly consistent. Fixing wake time first made a noticeable difference in how alert I felt by mid-morning.
What worked: the gap between bed time and lights-out. I was spending 25–40 minutes in bed before actually stopping my phone. The AI flagged this consistently. Reducing it to under 10 minutes — by not getting into bed until I was ready to actually sleep — was the change that moved my quality scores most.
What didn't work: the caffeine timing experiment. Three weeks of cutting off coffee at 1pm instead of 3pm showed no measurable difference in my morning feel scores. That was useful data too — it stopped me from being neurotic about afternoon tea.
The variables that matter for your sleep won't be the same as mine. That's the whole point of tracking your own data rather than following general advice. The AI is only as useful as the data you give it — specific data gives specific insight.
If you do have a wearable
If you have an Oura Ring, Whoop, or Eight Sleep, the manual tracking method above is unnecessary — you already have better data from the hardware. The prompts still apply, just replace the manual log with an export from your app.
Oura has the most useful native AI coaching built into the app itself — the readiness score and daily insights are genuinely well-designed. Using Claude or ChatGPT on top of Oura export data adds a second perspective, particularly for spotting multi-week patterns the Oura app doesn't surface in its default view.
Whoop is better for athletes and people doing serious training — the recovery and strain metrics are more granular than Oura for exercise-related sleep impact. Less useful if your sleep problems are primarily about consistency and habits rather than training load.
Eight Sleep is the most impressive hardware but at that price point it's a different category of commitment. If you have one, the onboard AI is good enough that external prompting adds less marginal value than with the other two.
A note on Indian sleep patterns
A few things about sleeping in India that most Western sleep content ignores completely.
Late dinners are normal and culturally embedded — eating at 9pm or 10pm is not unusual in many Indian households, including mine. Most AI sleep advice will immediately flag this as a problem. In practice, if you've eaten at 9pm your entire adult life, your body has adapted. The AI recommendation to "eat dinner by 7pm" may be genuinely unhelpful for someone whose family sits down to eat at 9pm and that's not changing. Flag this context explicitly in your prompt.
Summer heat in most of India meaningfully affects sleep quality in ways that northern European sleep research simply doesn't account for. Room temperature management, when it's possible, matters a lot — more than most other variables during April through June. Worth noting in your daily log if heat is relevant.
Afternoon naps (the post-lunch rest) are common and often culturally expected. Whether they help or hurt your night sleep depends heavily on timing and duration. The AI is actually useful for tracking this correlation — add "nap yes/no, duration" to your daily log if this applies to you.
Verdict
This habit requires more initial setup than the others in this series — the manual log, the two weeks before the data is useful, the discipline to note it every morning. It's the highest-friction habit here. But the output is also more concrete: measurable morning feel scores that moved from an average of 2.8 to 3.6 over the month, with two specific changes I can actually attribute to the AI's analysis rather than guessing. If your sleep is genuinely affecting your work or mood, the setup cost is worth it. If your sleep is just "a bit irregular sometimes" and it doesn't bother you much, start with one of the lower-friction habits first.