Nutritionix Natural Language & Grocery API

It is the verified-nutrition layer of the American and Canadian food supply, published as queryable records. Nutritionix - now owned by Syndigo - operates what it describes as the largest verified nutrition database: 1,266,570 food items, of which 1,053,256 are barcoded grocery items across 48,317 grocery brands, with a stated match rate above 92% for US and Canadian UPC scans. Around them sit 202,837 restaurant items from 860 chains monitored for menu changes, and a curated common-foods layer where an in-house registered-dietitian team maps 10,477 tags onto 38,828 everyday food phrases built on top of USDA data.

Three query families cover the corpus. Natural-language parsing converts a sentence - 'two scrambled eggs and toast' - into structured per-food nutrient rows. Instant search returns autocomplete candidates as you type. Direct lookup resolves a single item by UPC barcode or internal identifier. Scale tells you how battle-tested the plumbing is: roughly 20,000 health apps push about 250 million queries through it every month. On the Other Specialty Retail shelf this record owns the item-level nutrition lane, and it scores 6/10 on our rubric - carried by breadth and verification, docked because response-shape documentation is captured from observed structure rather than a fully published schema.

Three shapes cover the three query families:

# natural-language parse - one sentence becomes per-food rows
query_text     : "2 scrambled eggs and 1 slice wheat toast"
food_name      : Scrambled Eggs, 2 large
serving_qty    : 1          serving_unit: serving
nf_calories    : 199        nf_total_fat: 15.0 g
nf_protein     : 13.0 g     nf_sodium: 336 mg

# grocery record keyed by UPC - the enrichment join key
upc            : 041631000564
food_name      : Wheat Bread
brand_name     : Nature's Own
serving_qty    : 1          serving_unit: slice
serving_weight_grams : 43
nf_calories    : 110        nf_total_carbohydrate: 20.0 g
nf_dietary_fiber: 3.0 g     nf_sugars: 2.0 g

# restaurant row from a monitored chain menu
food_name      : Grilled Chicken Sandwich
brand_name     : Chick-fil-A
serving_qty    : 1          serving_unit: sandwich
nf_calories    : 320        nf_sodium: 970 mg

Read the anatomy rather than the digits. The parse block shows the grain decision the platform makes for you: one sentence, two foods, two separate rows, each carrying its own serving definition. The grocery block is the workhorse for retail enrichment - UPC as the join key, a named brand, and a per-serving panel anchored by serving_weight_grams so household measures convert cleanly. The restaurant block shows the second shelf living in the identical schema, which is why one pipeline serves both. Your sample ships cut against live records in these shapes before anything is committed.

Sixteen fields form the item spine, grouped four ways. Identity: food_name and brand_name, the latter splitting the 48,317 grocery brands from the 860 monitored chains. Serving block: serving_qty, serving_unit and serving_weight_grams - the trio that makes panels comparable, because 110 calories per slice and per loaf are different claims. Nutrient panel: the eight nf_ values from nf_calories through nf_potassium, reported per stated serving in grams and milligrams. Identifiers: nix_item_id, the internal key that survives renames and reformulations and rides alongside UPC for exact-item resolution.

Everything beyond the spine arrives through the request note rather than being invented into the schema: the full nutrient vector behind each nf_ summary value with its derivation attributes, the brand taxonomy with grocery categorization and derived wellness claims, multilingual item names across 15+ languages, and the common-foods phrase table mapping 38,828 expressions to 10,477 dietitian tags. Name the extras your workflow touches when you request a sample and they land as populated columns, not promises.

Geography - United States and Canada carry the depth that matters for barcoded goods: the stated above-92% UPC scan match rate applies to North American shelves. International extension comes through common foods and the multi-language add-on, not through foreign grocery UPC depth, so a European private-label barcode should be tested in your sample rather than assumed.

Temporal - the corpus is continuously maintained while items stay current. More than 3,000 grocery items are added or updated monthly, and the 860 restaurant chains are watched for menu changes, so reformulations and new SKUs arrive as updates rather than waiting for an annual re-release. There is no deep historical archive here; this is a current-state shelf, not a longitudinal study series.

Granularity - strictly item-level, one record per food with its per-serving nutrient panel, resolvable three ways: by UPC barcode, by internal item identifier, or by free-text phrase through the parser and autocomplete. Against the wider Datadory catalog - 1,744 datasets averaging 7.81 - this record scores 6/10, with breadth and verification pulling up and inferred response documentation holding it down.

API, files, or your warehouse. Daily, weekly, or hourly.

Channel and cadence are settings, not projects. Files suit the team loading the whole grocery corpus once and joining it to their own product tables offline. A direct pipe suits warehouses running assortment screens in SQL next to sales data. Record-level feeds suit anything interactive - a scanner page, a logging assistant, a chatbot answering 'how much sodium is in this?' mid-conversation. Every delivery carries the field dictionary above unchanged plus validation rows, units stay native to the source (calories, grams and milligrams per stated serving) unless you ask for conversions, and column naming locks at sample time.

• Product enrichment at barcode speed - e-commerce and retail-media teams attach complete nutrient panels to listings by UPC lookup instead of transcribing labels by hand.
• Conversational food logging - health apps turn plain sentences into structured nutrient rows via natural-language parsing, the mechanic behind roughly 250 million queries a month.
• Assortment and claim screening - category managers filter the grocery shelf on sugar, sodium, fiber and protein thresholds per serving to assemble better-for-you sets.
• Menu-change intelligence - analysts track nutrition drift across 860 monitored restaurant chains without re-collecting a single menu board.
• Model training - ML teams fit parsers and classifiers on a million-plus labeled, consistently keyed food records.
• Reference citation - researchers and journalists anchor claims about branded and restaurant nutrition on a dietitian-verified corpus rather than crowd-sourced guesses.

• Edamam Food Database API - the nearest competitor: close to a million items, about 790,000 of them UPC-barcoded. This record wins on verified depth and restaurant coverage; Edamam wins on diet-and-recipe add-ons.
• World Food Facts (Open Food Facts mirror on Kaggle) - roughly 356,000 volunteer-contributed packaged foods with 150+ columns; crowd-sourced breadth beside verified panels, useful as a cross-check layer.
• Food-Tagged Collections on Data.gov - the government-statistics complement: 208 food-keyed federal catalog records for demand and policy context rather than item-level labels.
• NRF Research & Insights Hub - industry-level retail benchmarks to sit above the item-level shelf this record provides.
• The Other Specialty Retail hub holds all ten records in the slice, including USDA FoodData Central's branded-foods files on the adjacent packaged-foods shelf for a government-run alternative corpus.

Three things. First, the response shape is documented here from observed structure - the nf_ convention is consistent across client libraries and live responses, but the full attribute list is pinned against live records when your sample is cut, not before. Second, geography has a center of gravity: North American barcodes resolve best, so test any non-US or non-Canadian private-label items in your sample rather than assuming parity. Third, nutrient completeness varies by item - potassium appears only where labels carry it, and restaurant items report per menu serving rather than per 100 grams. Name the brands, chains and cuts you care about, and the sample returns in exactly the schema shown above.