Introduction

Calorie tracking applications demonstrate substantial efficacy in short-term weight management studies, but long-term adherence remains a critical limitation. A consistent pattern across the literature shows steep dropout curves: adherence at 30 days typically ranges from 60–70% for well-designed applications, falling to 25–35% by 90 days. This adherence attrition substantially limits the real-world effectiveness of food tracking as a weight management strategy.

Behavioral science frameworks suggest that sustained engagement requires not only accurate feedback-which AI-vision tracking now provides-but also personalized, responsive coaching that addresses individual barriers to dietary change. Human dietitian coaching delivers these elements effectively but is resource-intensive and inaccessible to most users at scale. AI-powered conversational coaching offers a potential bridge: responsive, personalized, and scalable.

The Welling Coaching Module uses a large language model fine-tuned on dietitian consultation transcripts and integrated with individual calorie tracking data to provide contextually informed dietary guidance. This study evaluated its effectiveness in a rigorous RCT design.

Methods

Participants and Design

We conducted a parallel-group RCT at three digital health clinics in the United States (Boston, Chicago, Houston) between March and December 2025. Eligible participants were adults aged 18–65 with BMI ≥25 kg/m² who owned a compatible smartphone and had no prior use of the Welling platform. Participants were randomized 1:1 to calorie tracking plus AI coaching (intervention, n=1,078) or calorie tracking alone (control, n=1,078).

Both groups received identical onboarding, calorie targets set by a registered dietitian, and access to the Welling food tracker with AI food recognition. The intervention group additionally received access to the Welling Coaching Module, which provided daily check-ins, responded to food log entries with contextual dietary feedback, and offered weekly personalized nutrition summaries.

AI Coaching Module

The Welling Coaching Module was designed to simulate the motivational and educational functions of human nutrition counseling. Key features included:

  • Daily micro-feedback: Contextual commentary on food log entries within 30 seconds of logging
  • Weekly nutrition summaries: Personalized analysis of macro and micronutrient trends with actionable recommendations
  • Barrier identification: Adaptive questioning to identify and address dietary adherence barriers (e.g., social eating, time constraints, food preferences)
  • Goal progression coaching: Weekly target adjustments based on weight trajectory and dietary pattern analysis

Outcomes

The primary outcome was application adherence at 90 days, defined as active logging on ≥5 days per week. Secondary outcomes included weight change, Healthy Eating Index (HEI-2020) score, and user-reported satisfaction.

Results

Primary Outcome: Adherence

At 90 days, adherence was 81% in the AI coaching group compared to 36% in the calorie tracking alone group (absolute difference: 45 percentage points, 95% CI: 41–49 pp, p<0.001). The intervention group maintained adherence throughout: 94% at 30 days, 88% at 60 days, and 81% at 90 days. The control group showed the characteristic attrition pattern: 72% at 30 days, 51% at 60 days, and 36% at 90 days.

Weight Loss

Intervention participants achieved a mean weight loss of 5.8 kg (SD: 3.1 kg) at 90 days compared to 2.9 kg (SD: 3.4 kg) in the control group (difference: 2.9 kg, 95% CI: 2.5–3.3 kg, p<0.001). Fifty-eight percent of intervention participants achieved ≥5% body weight reduction compared to 24% of controls.

Dietary Quality

Healthy Eating Index scores improved significantly more in the intervention group (+12.4 points, SD: 8.1) compared to controls (+4.8 points, SD: 7.3, p<0.001). The largest HEI subcomponent improvements were in vegetables (+2.1 points), whole grains (+1.8 points), and refined grain reduction (+1.6 points).

User Satisfaction

Intervention participants rated the Welling Coaching Module highly: 91% reported that AI coaching made them more confident about their food choices; 87% reported the coaching felt personalized rather than generic; and 83% indicated they would recommend the AI coaching feature to a friend or family member.

Discussion

These findings demonstrate that AI coaching integration dramatically alters the adherence landscape for calorie tracking applications. The difference between 81% and 36% adherence at 90 days is not merely statistically significant-it represents a fundamental change in what food tracker applications can achieve in real-world deployment. Prior meta-analyses have documented that each 10 percentage point increase in tracking adherence correlates with approximately 0.6 kg additional weight loss at 12 weeks; the 45-point adherence advantage in this trial is consistent with the observed 2.9 kg weight loss differential.

The mechanism is likely multifactorial: improved feedback specificity, reduced decision fatigue through personalized meal suggestions, and motivational reinforcement through coaching interactions all plausibly contribute.

Conclusion

AI-powered nutrition coaching integrated within a calorie tracking application significantly improves adherence at 90 days (81% vs 36%) and produces superior weight loss and dietary quality outcomes compared to calorie tracking alone. These findings support the integration of conversational AI coaching into food tracker applications as a standard feature of digital weight management programs.