WHOOP's new Custom Journal Behaviors let people describe a routine in their own words and turn it into something they can track. Members review the suggested name and daily question before adding it; some behaviors include an additional measurement. WHOOP announced the feature on September 18, 2026.
We like this use of AI. A preset list is a sensible starting point for a journal, but the person wearing the band will eventually have something more specific in mind. They know why they kept answering work messages after dinner, or which part of their evening they want to change. Making them hunt for the closest available checkbox loses some of what made the observation interesting.
There is an appealing division of work here. The person supplies the context; the software helps turn it into a question they can answer consistently. That is a useful ambition for an AI feature. It also sounds considerably less tiring than designing your own questionnaire.
Custom behaviors are here; the wider rollout is less tidy
WHOOP's timing takes a little untangling. Ed Baker's September 3 roadmap put custom behaviors, a unified daily flow, and natural-language voice/text logging in a September/October window. The September 18 announcement confirms custom behaviors specifically, not the entire package.
Complicating matters, WHOOP's May 8 announcement already described voice/text Journal logging in present tense. These pages don't establish a clean launch date for that capability. Clearer release notes would save everyone some work.
The distinction is worth keeping because the broader idea is promising. Letting someone describe a routine and making daily logging easier address related problems. One expands what the Journal can capture; the other could make it less of a chore to maintain. A beautifully personalized list of questions is still a list of questions someone has to answer.
The WHOOP app and the public API have different boundaries
For developers, the immediate catch is access. As of September 18, 2026, Journal entries, custom behaviors, behavior-impact results, Coach conversations, and My Memory are not documented in WHOOP's public API reference. That qualification concerns the published interface; it says nothing about private arrangements.
The documented user-data resources cover physiological cycles, Recovery, sleep, workouts, basic profiles, and body measurements. They include useful measurements such as heart rate variability, resting heart rate, sleep-stage durations, and Strain. There is plenty to build with, but those records don't give you the person's explanation of their evening.
Our WHOOP integration walkthrough shows those measurements arriving through Terra. Sleep, Recovery and workout data can reach your backend in a common format, alongside supported records from the other wearables your users connect. WHOOP still requires your own app credentials; Terra handles ingestion, normalisation and delivery. Journal entries and Coach conversations remain outside that documented data access.
An app using these endpoints needs its own collection flow for habit context. We would start with a small optional check-in tied to something the person wants to understand, rather than immediately reproduce a comprehensive Journal. Nobody connected a wearable because they were short of forms.
WHOOP's OAuth flow requires consent to the requested scopes. Storing personal reflections and using them in an external coaching model deserve their own clear explanation and controls. A connection screen is a poor place to bury every subsequent use of someone's data.
The useful distinction inside your app is between a WHOOP measurement, a person's answer, and the model's interpretation. Keep their origins attached. An interpretation that gets saved as a confirmed habit can make the next answer look better supported than it is.
A custom habit still needs to mean something
The review step in WHOOP's custom-behavior flow is particularly welcome. Natural language is good at accommodating how people speak. Analysis needs a little more agreement.
Take “worked late.” That might mean after dinner, after a particular time, or beyond someone's usual hours. All are reasonable questions, but their answers are not interchangeable. A custom habit becomes useful when the person recognizes the question and can answer it the same way next week.
For an app collecting its own context, this is where a stable habit identifier, a version of the question, an answer type, and units earn their keep. If the definition changes, preserving that version prevents the old answers from silently acquiring a new meaning. These are design choices for your own records, not undocumented WHOOP Journal fields.
Dates are equally easy to flatten. A morning entry can describe the previous evening, so the time an answer arrives is not necessarily the time the behavior happened. Keeping the event time, local date and time zone, and entry timestamp separately gives you a chance of matching the right things.
WHOOP's own cycle documentation explains why physiological cycles and calendar days can differ. A night shift is enough to expose the weakness in grouping everything by calendar date. Matching an evening behavior to the following main sleep period can make sense, provided the convention is explicit and ambiguous entries remain open to correction.
Then there is what people actually say. “I didn't drink coffee” reports an absence. “I should stop drinking coffee” expresses an intention. “Coffee may be why I slept badly” offers a theory. A model that turns all three into the same record has made logging easier at a fairly unfortunate cost.
A short preview of the proposed answer and resolved date is worth the interruption. So is a correction flow that updates the existing record and revisits affected comparisons. We would happily trade a little conversational elegance for fewer invented cups of coffee.
Please don't turn a pattern into a verdict
WHOOP explicitly describes Journal findings as associations and says they do not establish causes. Good. The temptation to supply a satisfying explanation is exactly where a coaching interface needs some restraint.
Suppose someone's evening walks mostly happen on weekends, when their work schedule also changes. Better sleep after those walks would be interesting. It would not tell us how much of the difference belonged to walking, the weekend, or something nobody thought to log. A smooth paragraph explaining that walking “improved your sleep” would outrun the evidence.
Missing answers make this harder. Someone skipping a check-in has not said they skipped the behavior. Unanswered, deliberately skipped, and explicit “no” responses need to stay distinct. Otherwise, the comparison starts describing the person's logging habits as much as their life.
There is a similar trap on the measurement side. WHOOP's Recovery model distinguishes scored, pending, and unscorable records. An unavailable score is not a zero to drop into an average.
We would rather see a modest comparison with its dates, usable observations, and missing entries than a confident ranking of habits based on thin evidence. “There isn't enough comparable information yet” can be a perfectly useful answer. It gives the person a reason to keep observing without pretending the result is already settled.
Coaching that remembers what still matters
WHOOP's April 30 description of AI guidance combines biometric data with goals and routines. My Memory lets members inspect, add, edit, or remove that personal context. The ability to remove things deserves as much attention as the ability to remember them. A finished trip should not remain the explanation for every difficult week.
The measurements can change, too. WHOOP documents update and delete webhooks for sleep, Recovery, and workouts. Update notifications prompt the app to fetch the latest record; delete notifications identify records it should stop using. WHOOP also recommends reconciliation because events can be missed.
Our webhooks and polling explainer covers the delivery pattern. Through Terra, the application receives normalised data payloads, so it can update its own view without implementing each provider’s notification-and-fetch sequence.
For an app adding its own coaching, that plumbing reaches all the way to the answer on screen. A corrected sleep record can invalidate yesterday's comparison. A deleted habit entry needs to disappear from future retrieval and any affected summaries. Keeping the old explanation because it reads well would be a strange priority.
This is what makes the custom Journal interesting beyond its launch announcement. Someone can choose a small part of their routine they are curious about, give it a workable definition, and see what accumulates. WHOOP has made more room for that curiosity. The coaching is most useful when it leaves room for the result to be inconclusive, too.
Building those check-ins is work specific to your product. With the wearable feeds handled through Terra, you can extend it to another supported device without first writing another provider-specific importer. That leaves more time for the question someone wants to explore, and less for the format their watch happens to use.
