I used AI-powered calorie counting apps, and they were even worse than expected

Original Article ↗ Hacker News Discussion ↗

Scope & core reaction

  • Commenters generally agree the tested “AI calorie from photo” apps perform poorly and are oversold.
  • Many say they expected this: there simply isn’t enough visible information in a picture to estimate calories and macros reliably.

Why photo-based calorie estimation is fundamentally hard

  • Photos can’t reveal:
    • Cooking fats (oil, butter), sugar in sauces, or hidden ingredients.
    • Food variants (whole vs skim milk, lean vs fatty meat, low‑ vs high‑sugar yogurt, Coke vs Coke Zero).
  • Volume estimation is shaky: 2D images, inconsistent scale, and lack of depth data. Some note iPhones have depth/LiDAR, but say most apps either don’t use it or exaggerate their use of it.
  • Even in best case (standard containers, homogeneous foods), commenters doubt accuracy is good enough for the ~200–300 kcal precision needed for meaningful weight change.

Manual and LLM-assisted tracking vs “AI camera”

  • Several people report success with:
    • Traditional apps (MyFitnessPal, Cronometer, Macrofactor, Lose It, FoodNoms).
    • Using ChatGPT directly with detailed text/voice descriptions, weights, and labels.
  • Consensus: AI is useful as an assistant (parsing text, reading labels, logging meals, suggesting macros), not as a magic one-shot from photos.
  • Some say the effort of manual logging is part of why calorie counting works: it increases awareness and introduces friction before eating.

Debate on accuracy and usefulness of calorie counting itself

  • One camp: calorie labels and expenditure estimates are noisy (±20% or more), digestion varies, and CICO is oversimplified.
  • Another camp: despite imprecision, systematic tracking clearly works for many; not tracking is worse, and it’s especially useful for education (e.g., learning oil, restaurant meals, and alcohol are calorie-dense).

Business models, ethics, and user impact

  • Strong suspicion that some apps are hype-driven “snake oil”:
    • Heavy marketing, questionable revenue claims, likely paid/fake reviews.
    • Paywalls, upsells, and poor UX suggest quick money grabs riding “AI” branding.
  • Concerns:
    • Users may blame “calorie counting doesn’t work” when the tool is wildly off.
    • Risk of disordered eating if apps systematically under/overestimate.
    • Data-mining potential from detailed food-photo logs.
  • Some note there are more careful apps (e.g., SnapCalorie, Macrofactor, text-first tools) that stress education, databases, and clear communication of estimates, but even these admit substantial limitations.