Early Engagement and Recorded Weight Change with an AI Accountability Coach
A retrospective observational analysis
Francis John, BodyBuddy
Data through August 21, 2026. Aggregate results only; no individual member data is reported.
Abstract
Background. BodyBuddy is an AI health accountability coach. Members talk to it by text message, and coaching replies are generated by AI; BodyBuddy does not assign human coaches. This report asks two questions: Does messaging in week one predict later engagement? What weight change did members with a weight-loss goal record?
Methods. We reviewed members who completed onboarding. We counted a day of engagement when a member sent at least one message. For weight change, we compared each member's first recorded weight with the median of their weights near each milestone.
Results. Members who messaged on five or more days in their first week were 5.0 times as likely to still be active in week 8. The comparison groups included 106 and 85 members. Messaging on three or four days did not predict more week-8 activity than messaging on two days or fewer. Among 25 members with a weight-loss goal who kept recording weight through four weeks, median recorded change was -1.3%, and 18 recorded some loss. Smaller eight- and twelve-week cohorts showed larger median changes (-2.9%, n=13; -3.9%, n=6) and are too small to estimate typical results.
Interpretation. The behavioral signal is strong: heavy use in week one is associated with much higher use two months later, with a step at five days rather than a gradient. The health-outcome signal is promising and preliminary: among members who kept recording weight, recorded weight generally moved downward. What this analysis cannot tell us is how much of that change BodyBuddy caused, or what a new member should expect.
The first week predicts the second month
We counted how many days each member messaged BodyBuddy in week one. Then we compared each group at weeks 4, 8, and 12. The 0-to-2-day group is the baseline for every ratio.
How much the first week matters
Relative likelihood of being active in a later week, by days messaged in week 1, indexed to the 0-to-2-day group.
- All 7 days
- 5 to 7 days
- 3 to 4 days
- 0 to 2 days
| Days messaged in week 1 | Week 4 | Week 8 | Week 12 |
|---|---|---|---|
| 7 | 10.5× | 6.4× | 3.8× |
| 5 to 7 | 8.9× | 5.0× | 2.8× |
| 3 to 4 | 0.8× | 1.2× | 0.3× |
| 0 to 2 | 1.0× | 1.0× | 1.0× |
The relationship is a step, not a slope. Three or four days did not predict more later activity than zero to two days. Five to seven days did. The five-day threshold is a post hoc observation, not a hypothesis we set out to test. Its shape is what makes it worth reporting: if the pattern were only motivated members sorting themselves, we would expect a smooth gradient from two days to seven, and the data shows a discontinuity instead.
Weight change among members who kept weighing in
Each dot represents one member with a weight-loss goal. Weight was lower for 18 of 25 members at four weeks, 9 of 13 at eight weeks, and 5 of 6 at twelve weeks. Weight milestones begin at day 28, so every member in the weight analysis had continued past the free trial.
Recorded weight change for each eligible member
Percent change in body weight from each member's first recorded weight. Each dot is one member with a weight-loss goal. The vertical bar marks the median.
| Milestone | n | Median | Mean | Lost any | Lost 3%+ | Lost 5%+ |
|---|---|---|---|---|---|---|
| 4 weeks | 25 | -1.3% | -1.3% | 18 (72%) | 4 (16%) | 1 (4%) |
| 8 weeks | 13 | -2.9% | -2.8% | 9 (69%) | 6 (46%) | 3 (23%) |
| 12 weeks | 6 | -3.9% | -4.4% | 5 (83%) | 4 (67%) | 2 (33%) |
The comparison that rules out drift
Members without a weight-loss goal (maintain, gain, or no target on file) recorded a median change of -0.6% at four weeks (n=11), against -1.3% for goal-setters. The gap is small, but it answers the most obvious objection: the loss among goal-setters is not simply everyone who weighs in drifting downward. The two groups were not randomized or matched, so the gap is context, not an estimate of BodyBuddy's effect.
The same people over time
The three milestones are overlapping subsets, not one cohort followed from start to finish, so the table should not be read as a trajectory on its own. Within the same people it holds: 11 of the 13 eight-week members also had a four-week value. Their median was -1.9% at four weeks and -2.9% at eight weeks, and 9 of the 11 were lower at eight weeks than at four.
The 2013 AHA/ACC/TOS guideline on the management of overweight and obesity describes a sustained loss of 3% to 5% of body weight as clinically meaningful. This sample is too small and too selected to claim that members reach it. It is the benchmark the larger cohort in Research Note 2 will be measured against.
Daily logging and four-week weight change
At four weeks, 11 members had logged a meal or workout every day. All 11 had lost weight, and their median change was -1.8% in that group. Among the 14 members who logged less often, 7 had lost weight. Their median change was -0.1%. This is an exploratory pattern worth testing, not a result. Daily logging may also identify more motivated or adherent members, so the comparison cannot show whether logging itself made the difference.
Engagement and self-monitoring move together
We set out to compare weight change between high- and low-engagement members and could not. Nearly every member with a four-week weight had messaged on at least five days per week, and members who disengaged also stopped recording weight. That is a limitation for this analysis, and it is also a finding: in this data, weighing in and talking to the coach are the same behavior. There is no low-engagement group with observed weight outcomes, and we do not report one.
Methods
Population
The engagement analysis includes members who completed onboarding before August 21, 2026. We excluded internal and demonstration accounts. The weight analysis also required at least two observed weights. A member could type a weight in chat, enter it in the app, or sync it from Apple Health. We excluded imported history and weights recorded before signup.
Engagement
We counted a member as active on a day when they sent at least one message to BodyBuddy. Week 1 is days 0 through 6 after signup. Week 2 is days 7 through 13, and the same pattern continues. A member was active in a later week if they messaged on at least one day. We included a member in a week-N measure only if their account was at least N+1 weeks old on the analysis date. App opens, Apple Health syncs, and phone calls did not count.
Weight
Baseline is each member's first recorded weight. For each milestone, we used the median weight from a window that began 14 days before and ended 3 days after day 28, 56, or 84. This reduces the effect of daily weight variation. We excluded a member from a milestone when the window had no weight. We counted a weight-loss goal when the target was at least 1 kg below baseline. 5 of the 25 four-week weights came from a connected scale; the rest were typed in by the member.
Analysis
These results are descriptive. We use medians because the weight distributions are skewed. Each relative likelihood compares the share of one group active in a later week with the share of the 0-to-2-day group. This note reports relative likelihoods only; absolute activity rates are not reported. We did not run hypothesis tests.
Limitations
- Observational designThere is no control group or random assignment. Early engagement predicts later engagement, but this analysis cannot show that it causes it. The weight analysis also cannot separate BodyBuddy from medication, other programs, or natural variation.
- Week one overlaps the free trialFirst-week messaging is also the period in which a member decides whether to continue, so the engagement ratios reflect both the habit and that decision.
- SurvivorshipWeight is observed only for members who continued to weigh in. These members may differ from people who stopped recording.
- Small weight samplesThe three milestones include 25, 13, and 6 members. A few more observations could change the twelve-week result.
- Most weights were self-reportedEntries were not collected under a standardized measurement protocol. Five of the four-week weights came from a connected scale.
- Single program, single periodAll members used one product during a five-month period in 2026. Other programs and populations may produce different results.
Conclusion
Week one produced the clearest signal. Members who messaged on at least five days were several times as likely to be active two months later. Three or four days did not predict the same lift. Median weight moved lower at each milestone among members who kept weighing in, but that sample is small. The next report needs a larger weight cohort, fixed weigh-in milestones, more connected-scale data, and a target weight for every member. It will also define the analysis before looking at outcomes, report outcomes for every member who starts rather than only those who keep weighing in, and record GLP-1 and other weight-loss treatment at onboarding.
Research Note 2 will re-run these definitions on the cohort through November 2026 and will be published whatever the numbers show.
Questions people may have
- Does this prove that BodyBuddy caused weight loss?
- No. This is an observational report without a control group. It describes weight changes among members who kept weighing in, but it cannot establish what caused those changes.
- Are these typical results?
- No typical-results estimate is possible from this report. The weight sample is small, and people who kept recording may differ from people who stopped.
- Why are there fewer people at later milestones?
- A member needed enough time in the program and a recorded weight near the milestone. Fewer members met both requirements at eight and twelve weeks.
Disclosure
This analysis was prepared and published by the creators of BodyBuddy, who have a financial interest in the product. No independent researchers reviewed this report before publication.
Data availability and citation
Download the aggregate values in this report as a machine-readable JSON file. Individual member data is not public. We will use the same definitions in the follow-up report so the results remain comparable.
John F. Early Engagement and Recorded Weight Change with an AI Accountability Coach: A Retrospective Observational Analysis. BodyBuddy Research Note 1. Version 1.0. Published August 22, 2026.