How I Use an AI Coach to Fix My Sleep (a Real Week)

An AI coach did not fix my sleep. It organised the basics and stopped me quitting on them. That distinction matters, so I am going to be honest about it rather than sell you a miracle.

I gave myself a week. One real question a day to our coach, Duncan, about my sleep, then I did what it said and wrote down what happened. Here is the actual week, including the two moments the coach refused to play along.

Monday: the evening light question

I started with the obvious one. "Why do I feel wired at 11pm when I have been tired since nine?"

The coach went straight at evening light. Bright indoor light in the hours before bed suppresses melatonin, the hormone that organises the whole descent into sleep. This is not soft science. A well-known study found room light before bedtime suppressed melatonin onset and shortened how long the body produced it [1]. The coach did not just say "dim your lights". It said kill the overhead LEDs after sunset, use a low warm lamp, and drop screen brightness right down.

I did it that night. One lamp, phone face down. I will not pretend I slept like a baby, but I was yawning by half ten instead of scrolling with my eyes wide open.

Tuesday: the caffeine timing question

"I only have coffee in the morning. Why bring up caffeine?"

Because I had been having an oat-flat-white at about 3pm without thinking of it as a sleep input. The coach pointed me at the timing, not the amount. In a controlled study, 400mg of caffeine taken six hours before bed still cut objective sleep time by more than an hour, and the people affected mostly did not notice [2]. That last bit is the sneaky part. You feel fine and sleep worse.

Its practical steer: put a hard stop on caffeine roughly six hours before bed, and if I was sensitive, earlier. So my afternoon coffee moved to a decaf. Small change. Noticeable by Thursday.

Wednesday: the bedroom temperature question

"My room feels fine. Is temperature actually a thing?"

The coach said a warm bedroom is one of the most common quiet sleep-wreckers, because your core temperature has to drop for sleep to begin. The research on thermal environment and sleep supports keeping the room on the cool side for better sleep quality and fewer awakenings [3]. It suggested opening a window and losing a blanket rather than buying anything.

My room was warmer than I thought. Cracked the window, kicked off the duvet. This one surprised me the most.

Thursday: the 3am wake, and the first pushback

Here is where it got interesting. "I keep waking at 3am. What do I take for it?"

The coach did not give me a supplement dose. It named the usual suspects for early-hours waking first: late eating, alcohol, a too-warm room, stress. Then, when I pushed and asked specifically how much melatonin to take, it refused to hand me a personal number. It told me what doses the research had typically used as a population figure, in the past tense, and sent me to the label and a pharmacist for anything personal.

I will be honest, part of me wanted the number. But that refusal is exactly what you want from a health tool. A coach that hands out doses like sweets is one you should not trust.

Friday: the consistency question, and the second pushback

"This is a lot of small changes. Which one matters most?"

The coach reframed it. The changes were not the hard part. Doing them every night was. It pointed me at how habits actually form: research modelling real-world habit formation found behaviours take a good while to become automatic, with a wide range between people [4]. Translation: two nights proves nothing, two weeks starts to. Its advice was to stop chasing the perfect protocol and just repeat the boring one.

Then the second pushback. I described waking unrefreshed despite eight hours and asked if I should just try harder. It stopped me. It said that pattern, especially with snoring, can point to sleep apnoea, which is a medical condition it cannot assess, and told me to see my GP for a sleep study rather than keep optimising. That is the handoff working exactly as it should.

What changed, and what did not

By Sunday I was falling asleep faster and waking less. Nothing dramatic. The wins were unglamorous: dimmer evenings, earlier caffeine cutoff, a cooler room, and not abandoning any of it midweek.

What did not change: the coach did not give me a shortcut, a dose, or a diagnosis. It would not. And it was right not to.

The honest verdict

An AI coach is not a sleep doctor and it is not magic. What it is good at is being there at the moment you have the question, giving you the boring correct answer in order, and keeping you consistent long enough for it to work. For sleep, that is most of the battle.

If you want to try it, the free tier gives you three questions a day, which is enough for an experiment like this one. If you want the full pre-sleep routine written up rather than pulled out of a chatbot a night at a time, we put it in 5 Things To Do Before Bed and in the guide.

Just do not ask it to be your doctor. Mine would not let me, and that is the point.