Fitbit’s Improved Sleep Algorithm!
In a Nutshell
The new Fitbit/Google sleep algorithm shows only a minor improvement over the previous version, with Pixel Watch 4 achieving 86% deep, 78% light, and 71% REM sleep sensitivity against EEG reference. Google and Fitbit remain among the top five sleep trackers alongside Apple Watch, Oura Ring, and Whoop. The Google Health app prioritizes weekly overviews and AI coach summaries but presents cluttered text-heavy interfaces rather than clear data-first displays.
These notes were generated by AI and may contain inaccuracies.
The brand new sleep algorithm shipped with all Fitbit and Google devices including the Fitbit Air and Pixel Watch is claimed to be 15% better than the previous version. Independent testing will verify this claim using detailed data analysis. The same sleep stage tracking algorithm is used across Fitbit and Google devices including the Pixel Watch.
Testing uses the ZMAX EEG headband as the reference standard for sleep stage tracking. The ZMAX EEG headband measures brain waves and was designed for sleep stage tracking in scientific studies. The average sensitivity is calculated across deep sleep, light sleep, and REM sleep stages. The horizontal axis shows average sensitivity while the vertical axis shows performance of the worst sleep stage. Devices positioned toward the top right perform best.
The old Pixel Watch testing results (45mm and 41mm versions) are compared against the new algorithm results. The new algorithm shows a minor improvement in the initial test. Google claims a 15% improvement relative to the old algorithm, meaning a 1.15 multiplication of the previous performance score, not a 15 percentage point increase.
The new algorithm on the Pixel Watch 4 shows the following sensitivities compared to the ZMAX EEG reference:
- Deep sleep: 86% sensitivity
- Light sleep: 78% sensitivity
- REM sleep: 71% sensitivity
These values appear on the diagonal of the confusion matrix and are considered quite good compared to much of the competition.
In one example night, the first deep sleep segment showed near-perfect agreement between the reference and Pixel Watch 4. The watch detected some extra deep sleep, mainly one additional segment. The user typically experiences one or two deep sleep segments, usually earlier in the night rather than near the end. REM sleep agreement was quite good despite fragmented sleep with more awake time. The segments detected were roughly the same overall, with only a slight shift in one early REM segment.
Fitbit published a white paper on their website (not in a scientific journal) showing their official testing results. The white paper shows similar overall numbers to the independent testing but with deep sleep and REM sleep performance flipped. Google shows deep sleep as slightly worse than REM sleep, while the independent testing showed the opposite pattern. When plotted on the same sensitivity chart, the official testing results end up at nearly the identical location as the independent results.
Google and Fitbit remain among the top performers in sleep stage tracking. The other four major performers are:
- Apple Watch
- Oura Ring
- Whoop Strap
- Helio Pod
These five brands represent the top performers, with Google and Fitbit potentially showing slight improvement from the algorithm upgrade.
The Google Health app shows a focus on weekly metrics rather than day-to-day variation. The app displays weekly cardio targets achieved and lists different exercises performed, some auto-tracked and some manually tracked. The app can learn from user activity patterns to improve automatic workout classification for activities like elliptical or rowing.
The app opens to a weekly overview rather than a health-first view that would show out-of-range metrics first. The interface is described as somewhat cluttered with a lot of data presented. Textual coach summaries appear with individual logging items. The preference would be for numbers and visuals to appear front and center, with text available on request rather than displayed by default.
Text descriptions appear with most data points. The preference is for shorter summaries with extended text available on click. The new AI models make it natural for companies to prioritize text, but the preferred view would have data and graphs as the main element with text secondary. The sleep stage tracking tab shows a better balance between text and graphs.
The weekly overview is considered nice but not ideal as the first view when opening the app in the morning. The preferred first view would show basic health metrics. The Google Coach feature (available with premium) provides summaries and allows conversational queries. In one test, the coach was asked about marathon training recommendations after heavy running and cycling. The coach recommended 4-5 km of easy running for the weekend, noting that missing the 40 km goal was not a failure. The advice contained some contradiction but ultimately made reasonable sense. The conversational nature allows users to refine recommendations through follow-up questions.
Premium features include the Google Coach summaries. The full review of the Fitbit Air will discuss premium versus non-premium differences in detail.
Keep The Quantified Scientist in your library
Save the videos and channels worth coming back to, and find them again in one place.
