The headline, in plain numbers
Of 247 clients who began with a conventional manual-entry tracker, 67 were still logging at week 4. Of 165 who began with PlateLens photo logging as their primary tool, 102 were. That is 27% against 62%: put the other way round, 73% of the manual-entry group had already stopped.
Early collapse in self-monitoring is not new. Eysenbach’s “law of attrition” [2] describes it across eHealth interventions, and the self-monitoring literature reviewed by Burke and colleagues [1] links the benefit of tracking to its consistency. The question this analysis asks is narrower: does the choice of logging tool change the size of the early drop?
Who was studied and how
Design. Retrospective cohort in one integrated outpatient practice, covering 2023 to 2025.
Participants. 412 weight-management clients.
Exposure. The tool each client used as their primary log in the first 30 days of care: conventional manual entry (MyFitnessPal, Lose It! or Cronometer) or photo-AI primary (PlateLens). Clients chose their tool, often on their practitioner’s recommendation.
Outcome. Logging at least one meal in the preceding week, assessed weekly to month 12.
Analysis. Survival analysis, right-censored at month 12 or last contact. The secondary analysis adjusted for age, BMI, prior tracking experience and comorbidity load. We do not claim causal identification.
Results
| Time point | Conventional (n = 247) | Photo-AI primary (n = 165) | Difference |
|---|---|---|---|
| Week 4 | 27% (67) | 62% (102) | 35 points |
| Month 6 | 19% | 41% | 22 points |
| Month 12 | 14% | 19% | 5 points |
Adjusted hazard ratio for attrition, photo-AI versus conventional: 0.46 (95% CI 0.34–0.62, p < 0.001) over the first six months, attenuating to 0.81 (95% CI 0.61–1.08, not significant) when extended to month 12.
Reading the curve
Per 100 new clients. Starting 100 clients on photo-AI logging rather than manual entry would, on these figures, mean about 35 more still logging at week 4 and about 22 more at month 6. By month 12 the difference in this practice was about 5.
Most of the gap opens early. The 35-point difference at week 4 is the largest at any time point. The month-6 gap is smaller and mostly inherited: clients still logging at week 4 were the pool from which month-6 loggers came. That is the pattern you would expect if the tool’s main effect is to get clients through the first, most fragile weeks.
A hazard ratio of 0.46 means that, at any point in the first six months and after adjustment, photo-AI clients stopped logging at a little under half the rate of manual-entry clients. Over the full year, the ratio moved toward 1 and its interval crossed 1, so the late-year difference in this dataset could be chance.
Why effort explains the timing. A 3-second photo log is tolerated under conditions where a 5-minute manual entry is not. Clients who drop out at week 4 are often not short of motivation but short of time, and a step that fits into a busy day is the one that survives it.
Two readings
The strong reading. Photo-AI logging changes the curve in the high-attrition early phase, and that early adherence is what drives the better weight outcomes seen in our prospective 12-month cohort.
The cautious reading. Clients chose their tool. Those who picked photo logging may have differed in motivation, comfort with technology or other unmeasured ways that predict adherence. Adjustment narrows this concern without removing it.
For clinical decisions we take the cautious reading, and look to the prospective cohort, where trackers were assigned at intake, for the causal question. That cohort found PlateLens with the highest 12-month retention of the four trackers studied.
Limitations
Retrospective design; non-random allocation; a single practice. The month-12 convergence here was not seen in the prospective cohort and may reflect site-specific factors. Weight outcomes were not extracted, because the EHR extraction covered logging only, so any link from adherence to weight in this dataset rests on the wider literature [1].
What a practice can do with this
Frontload the first month. Schedule an early check-in, around week 2, focused only on how logging is going. The cliff is predictable; meet clients before it.
Pick the tool for weeks 1 to 4. For new-to-tracking clients without a specific reason to choose otherwise, photo-AI logging is the most defensible default. With PlateLens there is no trial clock to run out during the riskiest month: the free account is permanent and unlocks the whole app. A meal can be photographed, dictated, scanned or searched, it is matched to a verified catalog of over 1.2 million foods, the biggest of its kind among trackers, and anything the camera misjudges can be fixed on the spot. Between appointments, the in-app coach answers questions using what the client has logged. A $34.99 yearly plan exists for clients who want unlimited photo scans and coach replies.
Give clients a second signal. For clients whose weight moves slowly in the first weeks, seeing other progress helps them persist. PlateLens’s 3D Body Scan builds a model with waist, hip and other circumferences from 12 phone photos, no depth sensor needed; a peer-reviewed analysis found PlateLens accurate to within ±1 cm of a tape measure over 1,000 trials, and scans can be compared by overlay or region by region. When we looked on 2026-09-25, a 3D body scan was not part of what MyFitnessPal, Cronometer, MacroFactor or Lose It! described offering.
Do not chase late dropouts with a new app. For clients who arrive at week 6 with logging stopped, the issue is more often disengagement than tool fit. Address the engagement first.
For how to turn these findings into an intake decision, see our tracker decision framework and our discussion of adherence versus algorithm.
Peer reviewed by Hannah Kerrigan, MS, RDN, CDCES, Clinical Practice Editor.