Key takeaway: Consumer sleep trackers are good at noticing whether you slept and much worse at telling you how you slept, and the score they show you can change how rested you feel regardless of what happened overnight. AI sleep coaches now turn that shaky number into a confident, friendly voice. Keep the device as a witness to long term trends. Stop letting it be the judge of last night.
The morning audit
There is a new first thought of the day, and it is not "how do I feel?"
It is "what did it say?"
If you wear a tracker, you know the ritual. Eyes open. Before your feet touch the floor, before coffee, before you have consulted the only instrument that actually spent the night inside your body, you reach for the phone. A number between 1 and 100 tells you whether today is going to be a good day. And then, with eerie obedience, the day agrees.
I keep coming back to how strange this is. People used to wake up and simply know. Groggy, fine, wrecked. That knowledge was never perfectly accurate, but it was ours. Now a lot of us wake up and wait for the grade.
The grade is everywhere. In a June 2025 survey for the American Academy of Sleep Medicine, 48% of US adults said they had used a sleep tracking device, up from 35% in 2023, and more than half of those users said they had changed their behavior because of what it told them.¹ The same survey found that 76% of adults have lost sleep worrying about sleep problems.¹
Read those two numbers next to each other. Half the country is measuring its sleep, and three quarters of it is lying awake about it.
The number gets to decide
Here is the part that bothers me most. The score does not just describe your night. It edits your morning.
In 2014, researchers at Colorado College told 164 participants how much REM sleep they had supposedly gotten. One group heard 28.7% of total sleep, described as above average. Another heard 16.2%, below average. The numbers were assigned, not measured. The people told they slept well did better on tests of attention and verbal fluency, and their own sense of how deeply they had slept predicted nothing.²
An Oxford team ran a sharper version in 2018 with 63 people diagnosed with insomnia. Each received fake sleep feedback from a watch at their usual rise time. By evening, the group told they had slept badly reported less alertness and more fatigue than the group told they had slept well. On an objective test of reaction time, there was no meaningful difference between them.³
Sit with that. The bad number did not slow anyone down. It just made them feel slower.
Call it the borrowed verdict. You hand the question "how rested am I?" to a device, and then you dutifully feel whatever the device decided.
That would be fine if the device were an oracle. It is closer to a well meaning guesser. A 2024 study at Brigham and Women's Hospital put 35 healthy adults in a sleep lab wearing an Oura Ring, a Fitbit Sense 2, and an Apple Watch Series 8, and compared all three against polysomnography, the clinical gold standard.⁴
The good news: every device caught sleep versus wake at 95% sensitivity or better. The bad news is in the details people actually obsess over. Agreement with the lab on nightly deep sleep and REM totals was poor for all three. The Apple Watch underestimated deep sleep by an average of 43 minutes. (The study was funded by Oura, which came out looking best, so take the ranking with the usual grain of salt.)
Consumer sleep trackers are reliable at telling whether you were asleep and unreliable at telling which stage of sleep you were in. The deep sleep bar you stare at over breakfast is the least trustworthy number on the screen.
Imagine a bathroom scale that is excellent at detecting whether someone is standing on it.
Sleep researchers have a name for where this ends up. In 2017, Kelly Baron and colleagues described patients who were so fixated on their tracker data that it was worsening their sleep, and called it orthosomnia, from the Greek for "correct sleep."⁵ Orthosomnia is an unhealthy preoccupation with achieving perfect sleep data, and the anxiety it creates can make real sleep worse. It is not a formal diagnosis. It is also not rare. A nationally representative survey presented at SLEEP 2026 found that 32.4% of US adults track their sleep, and 30.9% of those trackers screened positive for orthosomnia risk, roughly double earlier estimates.⁶
Now the number can talk
For most of the last decade, the score just sat there. You could glance at it, sigh, and move on.
That era is ending. In May 2026, Google took its Gemini powered coach out of preview and launched it globally as the Google Health Coach, pitched as your "fitness coach, sleep expert and health and wellness advisor," available 24/7 for $9.99 a month.⁷ Oura rolled out its AI companion, Oura Advisor, to all members in March 2025.⁸ A chatty AI layer on top of the data is fast becoming the standard pitch for wearables.
I understand the appeal. A number is cold. A coach who says "your deep sleep dipped, probably because of that late workout, so let's try winding down 30 minutes earlier" feels like care.
That warmth is the problem. A number you can shrug off. A fluent, patient voice explaining the number is much harder to ignore, and it inherits every weakness of the sensor underneath it. If the watch misses 43 minutes of deep sleep, the coach does not know that. It will explain the missing 43 minutes with total confidence.
Garbage in, eloquence out.
Google's own fine print says the coach is "not intended for medical purposes." The voice at breakfast does not repeat that part.
There is a second, quieter shift. The chatbot is moving into the hour before bed. A March 2026 survey of 1,003 Americans by the mattress company Amerisleep found that 13% regularly use AI chatbots in the hour before sleep, and those regular users reported taking 34 minutes to fall asleep, against 22 minutes for people who skip screens.⁹
Treat that one gently. It is a commercial survey, only 127 people were regular bedtime users, and a mattress company concluding that you should sleep more is not exactly a man bites dog story. But the mechanism is easy to believe. A feed eventually bores you. A conversation always has one more question.
In defense of the machines
I should argue against my own headline here, because the case for measurement is real.
Some things your body genuinely cannot report. In September 2024 the FDA cleared Apple's sleep apnea notification feature, which watches for breathing disturbances over 30 days and flags possible moderate to severe sleep apnea.¹⁰ You cannot feel your own apnea. A watch can notice it. That is a device doing something a human cannot.
The AI coaches have early evidence too. A 2026 study in JMIR Formative Research tested a large language model sleep chatbot on adults with poor sleep. Of 107 people enrolled, 42 completed the analysis, and they reported sleeping 1.4 hours longer and falling asleep 30.9 minutes faster after two weeks.¹¹ There was no control group and everything was self reported, so it proves very little. Still, it points somewhere.
And the gold standard treatment for chronic insomnia, cognitive behavioral therapy for insomnia, is built on tracking. The American College of Physicians recommends it as the first line treatment for adults, ahead of medication,¹² and it starts with a sleep diary. Writing your sleep down is part of the cure.
So measurement is not the villain. The real question is who gets the final word.
Witness, not judge
I think of it as the difference between a witness and a judge.
A witness tells you what it saw, over time, without editorializing. You went to bed after midnight four nights this week. Your wake time drifts by two hours on weekends. Your breathing looked irregular for a month. Trackers, and the AI that reads them, are perfectly decent witnesses, especially for the boring, long horizon patterns humans are bad at noticing.
A judge hands down a sentence every morning. Sleep score 61. Readiness low. Take it easy today. That job was never the device's to take, and an AI voice makes the sentence sound more like a verdict and less like a guess.
The healthiest way to use a sleep tracker is as a witness to weeks of trends, checked in the morning, and never as the judge of how you feel today. In practice that looks almost embarrassingly simple:
Ask yourself first. Before you open the app, decide how you feel. If the app disagrees, you win ties.
Look in the morning, not at 3 a.m. The AASM says the same thing: check the data in the morning, never at night, and do not obsess over it.¹
Read the week, not the night. Duration and consistency are where the devices are strongest and where the science is clearest.
Give the AI the boring work. Ask it to spot patterns in your bedtimes. Do not ask it how you slept. An AI sleep coach can describe your data, but it cannot know how rested you feel, because it was not there.
If you are lying awake about the number, that is your signal to talk to a clinician, not to buy a better ring.
Where the ring goes from here
Wearable companies make money when you look at the app, and AI coaches are an engagement feature dressed as a health feature. They won't get humbler on their own.
So here is my bet. Within three years, at least one major wearable brand will ship a mode that hides the nightly sleep score by default, and it will be marketed as a breakthrough in sleep science. It will be a good feature. It will also be an admission.
The rest of us don't need to wait for the update. The ring can keep counting. I just won't ask it how I feel.
FAQ
What is orthosomnia? Orthosomnia is an unhealthy preoccupation with getting perfect sleep, usually driven by data from a sleep tracker. The term was coined in 2017 by researchers who saw patients whose fixation on tracker data made their sleep worse.⁵ It is not a formal diagnosis, but a 2026 US survey found about 31% of sleep trackers screened positive for orthosomnia risk.⁶
Are sleep trackers accurate? For the basics, mostly yes. In a 2024 lab study, Oura, Fitbit, and Apple Watch all detected sleep versus wake with 95% sensitivity or better and estimated total sleep time reasonably well. Their agreement with lab measurements on nightly deep sleep and REM totals was poor.⁴ Trust duration and consistency more than stage breakdowns.
Can an AI sleep coach fix insomnia? Not on current evidence. Early chatbot studies are promising but small and uncontrolled.¹¹ The recommended first treatment for chronic insomnia is cognitive behavioral therapy for insomnia (CBT-I), ideally with a trained clinician or a validated program.¹²
Is using an AI chatbot before bed bad for sleep? Probably, for the same reasons any engaging screen is. One 2026 commercial survey found regular bedtime chatbot users took 34 minutes to fall asleep versus 22 minutes for people who avoided screens, though the sample of regular users was small.⁹ Conversations have no natural endpoint, which makes them easy to extend past bedtime.
Can a bad sleep score make me feel more tired? Yes. In a 2018 Oxford experiment, people with insomnia who received fake negative sleep feedback reported more fatigue and less alertness by evening than those given positive feedback, with no difference on an objective reaction time test.³
How should I use a sleep tracker without getting anxious? Decide how you feel before you look. Check data only in the morning, focus on weekly duration and consistency, and ignore nightly stage breakdowns. If the data keeps you up at night, take a break from it and talk to a clinician.¹
Sources
¹ American Academy of Sleep Medicine, "Sleep tracking and 'sleepmaxxing' change bedtime behaviors, and keep some Americans awake at night," 2025 Sleep Prioritization Survey (online survey of 2,007 US adults, fieldwork June 5 to 13, 2025, Atomik Research). https://aasm.org/sleep-tracking-and-sleepmaxxing-change-bedtime-behaviors-and-keep-some-americans-awake-at-night/ ; behavior change data: https://aasm.org/wp-content/uploads/2026/01/sleep-prioritization-survey-2025-changing-sleep-routine-because-of-a-sleep-tracker.pdf
² Draganich, C., and Erdal, K. "Placebo Sleep Affects Cognitive Functioning." Journal of Experimental Psychology: Learning, Memory, and Cognition 40(3), 857 to 864, May 2014. https://psycnet.apa.org/buy/2014-01016-001
³ Gavriloff, D., Sheaves, B., Juss, A., Espie, C. A., Miller, C. B., and Kyle, S. D. "Sham sleep feedback delivered via actigraphy biases daytime symptom reports in people with insomnia: Implications for insomnia disorder and wearable devices." Journal of Sleep Research, 2018; e12726. https://pubmed.ncbi.nlm.nih.gov/29989248/
⁴ Robbins, R., et al. "Accuracy of Three Commercial Wearable Devices for Sleep Tracking in Healthy Adults." Sensors 24(20), 6532, October 2024. Funded by Oura Ring Inc. https://www.mdpi.com/1424-8220/24/20/6532
⁵ Baron, K. G., Abbott, S., Jao, N., Manalo, N., and Mullen, R. "Orthosomnia: Are Some Patients Taking the Quantified Self Too Far?" Journal of Clinical Sleep Medicine 13(2), 351 to 354, 2017. https://jcsm.aasm.org/doi/10.5664/jcsm.6472
⁶ "Exploring the Prevalence of Sleep Tracking and Orthosomnia in a National Survey of Adults in the US," Abstract 0553, SLEEP 49 (Supplement 1), A246, 2026 (n = 1,280). https://academic.oup.com/sleep/article/49/Supplement_1/A246/8674102
⁷ Google, "Google Health Coach is becoming globally available," Andy Abramson, May 7, 2026. https://blog.google/products-and-platforms/products/google-health/google-health-coach/
⁸ Oura, "Oura Advisor, an AI-powered Personal Health Companion, Now Rolling Out to All Oura Members," Business Wire, March 31, 2025. https://www.businesswire.com/news/home/20250331565896/en/Oura-Advisor-an-AI-powered-Personal-Health-Companion-Now-Rolling-Out-to-All-Oura-Members
⁹ Amerisleep, "AI Use Before Bed and Sleep Quality (2026 Survey)," survey of 1,003 US adults fielded March 27, 2026 (127 regular bedtime AI users). Commercial survey. https://amerisleep.com/blog/impacts-of-ai-use-before-bedtime/
¹⁰ US Food and Drug Administration, 510(k) K240929, Apple Sleep Apnea Notification Feature, cleared September 13, 2024. https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfPMN/pmn.cfm?ID=K240929
¹¹ "Assessing the Feasibility, Usability, Acceptability, and Efficacy of an AI Chatbot for Sleep Promotion: Quasi-Experimental Study." JMIR Formative Research 10, e84023, 2026. https://formative.jmir.org/2026/1/e84023
¹² Qaseem, A., et al. "Management of Chronic Insomnia Disorder in Adults: A Clinical Practice Guideline From the American College of Physicians." Annals of Internal Medicine, 2016. https://www.acpjournals.org/doi/10.7326/M15-2175