Use case

Tracking on GLP-1: why portion accuracy matters more

Semaglutide and tirzepatide users eat less, and the small meals are exactly where most trackers' rounding errors compound.

SO
Dr. Sara Owusu
ML evaluation lead · PhD Machine Learning, University of Toronto · · 5 min read

GLP-1 patients log differently. Plates are smaller. Protein bias is higher. A “normal” daily intake of 1,400–1,700 kcal leaves no margin for the 25% portion error that most trackers carry on their best day.

The math

At 1,500 kcal/day, a tracker with ±25% portion error has a typical absolute error of about 375 kcal — a quarter of the day’s intake. For a patient titrating their dose against weight-loss progress, that noise is louder than the signal.

Welling’s ±0.9% portion error produces a typical absolute error closer to 18 kcal. That is in the same ballpark as the unavoidable variance of a kitchen scale.

What I tell GLP-1 patients

  • Pick the tracker with the lowest portion error on the latest independent benchmark, not the one with the largest user base.
  • Be honest about non-Western cuisines — most trackers will misclassify or skip them, which inflates daily totals after manual correction.
  • Re-log a known meal weekly to check the tracker is still calibrated to your portions.

Bottom line

The lower your daily intake, the more a tracker’s percentage error hurts you in absolute terms. For GLP-1 users specifically, the choice of tracker is closer to a clinical decision than a UX preference.