Cornell Found 'Low Sodium' Labels Hide Added Sugar. Our Scanner Can't Even Show You the Sodium.
A Cornell analysis of 145,609 packaged foods found health claims like low sodium and no trans fat often hide a trade-off elsewhere on the label. Checking whether Bite would let you catch that trade-off yourself turned up a real gap in our own barcode scanner.
Cornell researchers published an analysis this month that I think every food-tracking app, ours included, should sit with for a minute. Miguel Gomez, a professor of food marketing at Cornell, and Pei Zhou, a postdoctoral researcher in his lab, went through a decade of new-product data from the market research firm Mintel: 145,609 packaged foods and beverages launched in the United States between 2015 and 2024. For each one, they ran the full nutrient profile, not just whatever claim was printed on the front of the box, through the Nutrient Profiling Model, a scoring system built by the U.K. Food Standards Agency that weighs sugar, sodium, saturated fat, and calories against fiber, protein, and produce content all at once. Then they checked that score against whatever health claim the product was actually carrying.
The claims didn’t hold up. A product labeled low sodium was disproportionately likely to carry more added sugar than a comparable product with no claim at all, as if the sodium had to go somewhere and it went into the sugar column instead. A no trans fat claim tracked with higher saturated fat and higher sodium, the reformulation quietly getting worse in two places to get better in the one place that made it onto the front of the box. Beverages were the worst offenders: claims like low sodium, sugar free, and low saturated fat were the ones most likely to diverge from the product’s real nutritional quality. None of this means the claims are lies exactly. Removing trans fat is still removing trans fat. It means a single claim on the front of a package tells you about one ingredient the manufacturer chose to fix, and says nothing about what they did instead to the ones they didn’t.
I read that and did what I’ve been doing all year with these stories: went and checked what Bite actually shows you when you scan one of these products. The honest answer split into a good half and a bad half. The good half first. When you scan a barcode, Bite pulls the product from Open Food Facts and shows you calories, protein, carbs, fat, and sugar, straight off the nutrition panel. There’s no health score, no green badge, no “healthy” label layered on top of the raw numbers. We never built the thing this Cornell study is warning people about, a single front-of-package verdict standing in for the whole picture. That’s not a virtue we earned on purpose. It’s just that nobody ever built a scoring layer for scanned products, so there was never a claim to trust or distrust in the first place. But it means the app’s default behavior, for this specific failure mode, happens to be the right one: show the numbers, let the person add them up.
The bad half is that showing the numbers only works if you’re actually showing all of them, and we’re not. I went into the code that maps an Open Food Facts product onto what the app displays, and the field list we request from their API is calories, protein, carbs, fat, and sugar. Sodium isn’t in it. Our own nutrition label component knows exactly how to render a sodium row, it’s used for recipes pulled from Spoonacular, which does return sodium. But the barcode path, the one you’d actually use standing in a grocery aisle looking at a can that says “low sodium” on the front, never asks Open Food Facts for the one number that would tell you whether that claim is true. You’d see the sugar. You wouldn’t see the sodium it may or may not have traded away.
That’s a strange gap to have sitting in a nutrition app for this long, and I don’t have a good excuse for it beyond the ordinary one: nobody was looking for it until a study gave me a reason to look. The fix is small, add sodium to the same field request we already make for calories and fat, and I’d rather say plainly that it isn’t fixed yet than write a paragraph implying it already works the way it should. The Cornell paper’s practical advice for shoppers was to treat the front of the box as a hint and the nutrition panel on the back as the fuller story. Right now, if you’re using Bite to read that back panel for you, we’re leaving out a line.
Three Quick Bites

-
Governor Gavin Newsom signed AB 2244 this month, creating California’s first “Non-Ultraprocessed Certified” seal, a voluntary label that food makers can apply for once the state accredits certification agents. The bill passed the Senate 32-0 and the Assembly 72-0, and it builds on AB 1264, the 2025 law that gave California the country’s first legal definition of an ultra-processed food. It’s the opposite instinct from the Cornell findings above: a claim that has to be earned through third-party certification rather than picked by whichever marketing team wrote the box copy. Environmental Working Group
-
Pew Research found 34 percent of U.S. adults now use an AI chatbot for at least one health-related reason, from checking symptoms to decoding a diagnosis, according to a survey of 3,488 adults that’s been widely discussed again this month. Only 29 percent said they’re very comfortable actually sharing personal health information with the tool they’re asking. People trust the answers more than they trust the thing holding their data, which is a strange but very human way to adopt something. Cato Institute
-
Oura filed to go public on September 21st, targeting a fully diluted valuation of roughly $15.6 billion on revenue that grew 74 percent to $1.21 billion over the nine months ending in June. A ring that tells you your Readiness score is now a Nasdaq-bound business bigger than most people would have guessed, and it’s a reminder that the “just tell me one number” instinct in health tech isn’t a fad, it’s a real market. Yahoo Finance
Tool of the Week: NutriProfiler-style label reading, done by hand

Since this issue is about a front-of-package claim hiding what’s actually in the nutrition panel, this week’s pick isn’t an app at all: it’s the FDA’s own guide to reading a Nutrition Facts label, which walks through exactly which lines to check against each other, like sodium against sugar, when a package is making a claim about only one of them. It’s free, it’s two minutes long, and it works on literally any product, scanned or not. FDA: How to Understand and Use the Nutrition Facts Label
Thanks for reading. If you want a coach that tells you what it isn’t showing you yet, Mr BITE is free on iOS and Android.
Collins, building Mr BITE