Original research · Open data
Two Rulebooks, One Protein: How Gluten Disclosure Differs Between the US Food Supply and US Medications
US food labels must declare wheat; US drug labels need declare nothing. Measuring 800 US food products against 1,360 EU drug leaflets and 86 paired molecules finds the outcome inverts the rules — 21.1% of food products with a source-ambiguous ingredient name no source, against 11.6% of US drug labels.
Data and code
- off-results-2026.jsonMeasured disclosure rates across 800 US food products
- off_sample.pyOpen Food Facts sampler — reproduces every food figure
- ema-starch-census-2026.csvEU drug starch census used for comparison (CC0)
Released CC0 — no attribution required, though it is appreciated and makes corrections easier to route.
A measured comparison of what labels actually tell you, and what the law actually requires
Abstract
A person with celiac disease navigating a grocery aisle and a pharmacy counter is operating under two completely different disclosure regimes, and almost nobody states the difference precisely.
In food, wheat must be declared by name — it is one of nine major allergens under FALCPA. In medication, nothing must be declared. The FDA's guidance on gluten in drug products has sat in draft since December 2017.
We measured what that produces. Sampling 800 US food products with transcribed ingredient lists from Open Food Facts, and comparing against our earlier censuses of 1,360 EU drug leaflets and 86 paired US/EU molecules, we find:
- 21.1% of US food products containing a source-ambiguous ingredient (starch, maltodextrin, natural flavoring, hydrolyzed protein) name no botanical source anywhere on the label.
- 11.6% of US drug labels containing a starch excipient fail to name its source.
- 50.0% of EU drug leaflets containing a starch excipient fail to name its source.
US drug labels are, on this measure, clearer about ingredient origin than US food labels — despite food having a mandate and drugs having none. The mechanism is the same one we identified in EU-versus-US drug labelling: the FDA's substance registry assigns controlled names that embed botanical source (STARCH, CORN), while food ingredient lists are free text where "modified food starch" is a complete and legal declaration.
Both regimes also share a specific hole: FALCPA's mandate covers wheat, but not barley or rye.
1. Why this comparison
This piece exists because of a link. Celiac Sift — a celiac-safe recipe, ingredient-checking and wellbeing app built by a mother and daughter who both live with the condition — lists Gluten Screen among the resources it trusts for medication checking, alongside the Celiac Disease Foundation, DailyMed and Beyond Celiac.
Their framing of the medication problem is exactly right and admirably blunt:
Pills can contain gluten through starch-based fillers, and the FDA does not require gluten labeling on drugs.
They also do something most tools do not: their barcode scanner covers food and personal care, and they say plainly that medications and supplements are out of scope, routing users to specialist resources and their pharmacist instead of guessing. That is the correct call, and it is the kind of stated limit that makes a tool trustworthy.
It also raises the obvious question. If the FDA does not require gluten labelling on drugs, what exactly does it require on food — and does the difference in rules produce a difference in outcomes you can measure?
We measured it.
2. What the law actually requires
2.1 Food: one mandate, one voluntary claim, one gap
Two separate mechanisms operate, and conflating them is the most common error in this area.
FALCPA (2004) — mandatory. The Food Allergen Labeling and Consumer Protection Act amended the FD&C Act to require that foods containing a "major food allergen" declare it by source name. Eight allergens were named in 2004: milk, egg, fish, crustacean shellfish, tree nuts, peanuts, wheat and soybeans. The FASTER Act (2021) added sesame as the ninth, effective 1 January 2023.
Wheat is on that list. If a product contains wheat, the label must say "wheat" — in the ingredient list or in a "Contains: Wheat" statement.
21 CFR 101.91 (2013) — voluntary but regulated. The gluten-free labelling rule, issued under a FALCPA directive, became effective 4 September 2013 with compliance required by 5 August 2014. It defines "gluten-free" to mean the food contains no gluten-containing grain, no ingredient derived from one that hasn't been processed to remove gluten, and under 20 ppm gluten. A product bearing the claim and failing the standard is misbranded.
Note the shape: making the claim is optional; meeting the standard once you make it is not.
The gap. Barley and rye are gluten-containing grains. Neither is a FALCPA major allergen. So for any product that does not carry a voluntary gluten-free claim, there is no federal requirement to disclose barley or rye at all.
Gluten Free Watchdog has pressed this point publicly, and the FDA's stated position in draft guidance is that it will consider adding allergens only where there is robust evidence of an IgE-mediated reaction — which celiac disease, being autoimmune rather than IgE-mediated, does not produce. The practical effect is that malt (barley-derived) can appear as "natural flavoring" or "malt flavoring" with no allergen callout.
2.2 Medication: a draft, nine years old
There is no mandatory gluten or excipient-source disclosure for US drugs.
The FDA issued Gluten in Drug Products and Associated Labeling Recommendations as draft guidance in December 2017. It remains draft. Draft guidance is non-binding on its face and states the agency's current thinking; it does not require anything of anyone.
The contrast with the EU is instructive but not in the direction most people assume. EMA guidance (CHMP/704219/2013) does require that wheat starch, when used as an excipient, be declared on the product information leaflet. That is a real mandate where the US has none — and as §3 shows, it produces worse outcomes.
3. What the labels actually say
Rules are one thing. We measured output.
3.1 Method
Food. We sampled US products from the Open Food Facts API, restricted to those with transcribed ingredient text, and read the text directly rather than trusting derived allergen tags. Final sample: 800 products, all with ingredient lists.
The restriction matters. Open Food Facts is crowd-sourced, and a missing allergen field can mean "the label says nothing" or "nobody has typed it in yet." Those are different facts, and conflating them would destroy the comparison. Reading the ingredient text directly asks a cleaner question: when we can see the label, does the label name the grain?
Drugs. Figures come from our earlier work: a census of 1,360 EMA centrally-authorised product leaflets, and a within-molecule comparison of 86 medicines authorised in both the US and EU.
The comparison metric. For both domains we ask the same question: among products containing an ingredient whose botanical source is ambiguous, how often is the source named?
For food, source-ambiguous means starch, modified food starch, maltodextrin, dextrin, natural or artificial flavouring, caramel colour, hydrolyzed vegetable protein, yeast extract, or vinegar. For drugs it means starch, dextrin or dextrate excipients.
3.2 Results: US food
| Measure | Products | Share |
|---|---|---|
| Sampled, with ingredient text | 800 | 100% |
| Name a gluten-containing grain | 201 | 25.1% |
| Carry an explicit "Contains" statement | 42 | 5.2% |
| Carry a gluten-free claim | 86 | 10.8% |
| Contain a source-ambiguous ingredient | 298 | 37.2% |
| Ambiguous ingredient, no source named anywhere | 63 | 7.9% |
Of the 298 products containing a source-ambiguous ingredient, 63 — 21.1% — name no botanical source anywhere on the label.
3.3 The three-way comparison
| Corpus | Products with ambiguous ingredient | Source not named | Rate |
|---|---|---|---|
| US drug labels (paired molecules) | 86 | 10 | 11.6% |
| US food labels (this study) | 298 | 63 | 21.1% |
| EU drug leaflets (1,360-product census) | 218 | 109 | 50.0% |
US drug labels come out clearest. US food labels are roughly twice as ambiguous. EU drug leaflets — the only one of the three with a wheat-declaration mandate — are worst by a wide margin.
3.4 Why the mandate loses
This is the same mechanism we found comparing US and EU drug labels, showing up again across a different boundary.
US drug labels encode excipients through the FDA's Substance Registration System, which assigns each substance a UNII code and a controlled name. For starch derivatives, that controlled name carries the botanical source: STARCH, CORN, SODIUM STARCH GLYCOLATE TYPE A POTATO.
Food ingredient lists are free text. "Modified food starch" is a complete, legal, compliant declaration. So is "natural flavoring." Neither tells you the grain.
A mandate answers exactly one question — is it wheat? — and answers it by absence. A controlled vocabulary answers a better question — what is it? — from which the wheat answer and every other source question follow for free.
The food regime is not badly designed. It was designed to protect people with IgE-mediated wheat allergy, and for that purpose a wheat-specific mandate is exactly right. Celiac disease has a different shape: it responds to barley and rye as well, and it is triggered by quantities that a "contains wheat" callout was never calibrated to catch. The regime is doing its job; its job is just not quite the celiac job.
4. What this means in practice
4.1 The pharmacy is not obviously worse than the grocery store
The received wisdom — including, reasonably, on most celiac resource pages — is that medication is the harder problem because the FDA requires nothing.
The requirement gap is real. The information gap is not what you would predict from it. On the specific question of ingredient origin, a US drug label is more likely to tell you than a US food label, because of a substance registry adopted for unrelated administrative reasons.
This does not make medications safe or make pharmacist verification unnecessary. Matson et al. (2025), reviewing 689 NDCs across the top 100 paediatric medications, found 25.7% could not be confirmed gluten-free even after contacting the manufacturer directly, and 1.5% were confirmed to contain gluten. Labels are only part of the picture: they say nothing about cross-contamination in manufacturing, and formulations change without notice.
What it does mean is that the reason medication is hard is not "labels tell you less." It is that nobody is required to answer, so there is no backstop when the label is silent.
4.2 The advice that follows
For food: a gluten-free claim is regulated and meaningful — under 20 ppm, enforceable, misbranding if false. Absence of a claim tells you much less. "Contains wheat" is reliable. Silence about barley and rye is not evidence of absence, because nothing requires their disclosure.
For medication: read the inactive ingredients on DailyMed, where the controlled substance name usually names the starch source. Where it does not, that is a genuine unknown, and the manufacturer is the next step — with the caveat that a quarter of the time they cannot answer either. Never stop or switch a prescribed medication over a gluten concern without talking to your care team, which is exactly what Celiac Sift tells its users.
4.3 One change that would fix most of it
Requiring the botanical source of starch-derived and hydrolysed ingredients — in both food and drug labelling — would close most of this gap in both domains simultaneously.
The field already exists. In food, 78.9% of products with an ambiguous ingredient already name a source somewhere. In US drugs, 88.4% do. Manufacturers are largely already doing it; they are just not required to, so the remainder is silent and you cannot tell which kind of silence you are looking at.
5. Limitations
- Crowd-sourced food data. Open Food Facts is contributor-maintained. Restricting to transcribed ingredient text mitigates but does not eliminate transcription error and selection effects — contributors may over-represent products of interest to allergen-avoidant users, which would bias toward more disclosure, making our 21.1% conservative.
- Sample size. 800 products, drawn from an API paginating over 946,758 US entries, and terminated by rate limiting rather than by a sampling plan. This is a first-pass measurement, not a stratified survey.
- Pattern matching, not adjudication. "Source named" is determined by whether a botanical source word appears in the ingredient text. It does not verify that the word refers to the ambiguous ingredient specifically.
- Ambiguous-term list is a judgement call. Including vinegar and yeast extract is defensible and debatable. A narrower list would lower the food ambiguity rate.
- Different denominators across corpora. Food and drug ambiguity are measured on analogous but not identical ingredient classes.
- Label-derived throughout. This measures disclosure, not gluten content. No claim is made about any product's actual safety.
- Not clinically reviewed. The author is an independent developer, not a clinician, dietitian or attorney. This is data enumeration, not medical or legal advice.
6. Reproducibility
python3 off_sample.py 30 > off-results-2026.json
All figures regenerate from the published script. Drug-side figures come from the EU excipient census and the paired US/EU study. Released CC0.
Corrections: https://glutenscreen.org/about.html#corrections
Acknowledgement
Thanks to Celiac Sift for the link that prompted this comparison, and for being unusually clear about what their tool does not cover. A scanner that says "medications are out of scope, ask your pharmacist" is more useful than one that guesses — and it is rarer than it should be.
References
Food Allergen Labeling and Consumer Protection Act of 2004, Title II of Pub. L. 108–282.
FASTER Act of 2021 — sesame added as the ninth major food allergen, effective 1 January 2023.
21 CFR 101.91 — Gluten-Free Labeling of Foods. Final rule 78 FR 47154 (2013); fermented and hydrolysed foods rule 85 FR 49240 (2020).
US Food and Drug Administration. Gluten in Drug Products and Associated Labeling Recommendations. Draft guidance, December 2017.
European Medicines Agency. Wheat starch containing gluten as an excipient in medicinal products for human use. CHMP/704219/2013.
Matson KL, et al. Gluten-Free Options for the Top 100 Pediatric Medications. Journal of Pediatric Pharmacology and Therapeutics 2025;30(3):367–371.
Open Food Facts. Collaborative food products database. openfoodfacts.org.
Walsch N. The Source-Disclosure Gap: Starch Excipients in 1,360 EU Centrally-Authorised Medicines. Gluten Screen, August 2026.
Walsch N. The Vocabulary Beats the Mandate: Starch Source Disclosure in Paired US and EU Drug Labels. Gluten Screen, August 2026.
Gluten Screen is a free, no-signup medication screening tool. No advertising, affiliate links, or manufacturer funding. Editorial policy: https://glutenscreen.org/about.html#editorial-policy