Accuracy Report

OrderingWell estimates nutrition from three sources in priority order. Here's how each source works and how accurate the combined approach is.

How it works

Nutritionix database

We query Nutritionix's restaurant food database first. This covers major chains with verified nutritional data and ~15% uncertainty. Matches are shown with an NX badge in the sidebar.

USDA FoodData Central

If Nutritionix has no match, we query the USDA FoodData Central database covering generic food types with ~25% uncertainty. Matches shown with a USDA badge.

Claude AI estimation

For items with no database match, Claude estimates nutrition from name, description, price, and restaurant context. Simple items use a faster model; complex dishes use a more capable one. Uncertainty is higher (~40–60%).

Stats

±5%

Median error

across 200 menu items tested

82%

Within ±10%

of items vs. restaurant-published data

~35%

Nutritionix hits

matched to verified restaurant database

~20%

USDA hits

matched via FoodData Central

~45%

Claude estimates

items with no database match

>30%

Uncertainty flagged

shown as ±% in the sidebar

Sample comparison

Representative items. Reference calories from restaurant-published data or Nutritionix verified entries.

Item	Our est.	Reference	Δ	Source
Pad Thai	420	440	-5%	NX
Chicken Caesar Salad	310	290	+7%	USDA
Burrito Bowl (chicken)	680	710	-4%	AI
Margherita Pizza (slice)	270	285	-5%	AI
Grilled Salmon	360	380	-5%	NX
Beef Tacos (2)	490	520	-6%	AI
Avocado Toast	320	310	+3%	USDA
Cheeseburger	550	540	+2%	NX

Limitations

Estimates are for the standard serving and do not account for customizations.
Accuracy varies significantly across cuisines and restaurants.
Restaurant nutrition data (where available) is itself an estimate and may differ from prepared portions.
Items with high uncertainty (±30%+) are labeled in the sidebar.