Claim: Popular AI calorie-tracking apps underestimate meal calories by approximately one-third

The claim that popular AI calorie-tracking apps underestimate meal calories by approximately one-third is mostly false. While some studies indicate significant inaccuracies, particularly in specific cases, the overall consensus suggests that the average underestimation is lower than one-third. Mainstream sources like systematic reviews report calorie estimation errors ranging from 0.10% to 38.3%, but many independent tests show errors around 10-15% for single foods and 25-35% for mixed meals, not consistently reaching one-third. Disputing this claim, some sources argue that the inaccuracies are overstated and vary widely across different apps and meal types. The models diverge sharply — treat this as higher-uncertainty. Perplexity comes in highest (68%), while OpenAI is lowest (40%). Gemini expresses higher confidence than Perplexity on this claim. While there is evidence suggesting that some AI calorie-tracking apps can underestimate calories significantly, the extent of this underestimation varies. Some studies report errors close to one-third, but other sources indicate that typical inaccuracies are lower, often around 10-15% for single foods and 25-35% for mixed meals. This discrepancy in findings means that while there are instances of substantial underestimation, it does not uniformly apply across all apps or meals, leading to uncertainty about the claim's accuracy.