A “gluten-free” label and a “may contain” warning are not the same risk

Health Canada bought 657 products off ordinary shelves, all of them made from grains and ingredients that contain no gluten of their own, and measured what was actually in them. The label turned out to predict the result — just not in the direction people usually assume.

Two lines on a package look like they are doing similar work. One says gluten-free. The other says may contain wheat, or produced in a facility that also processes wheat. Both are voluntary. Both are printed by the manufacturer rather than verified by a shopper. A reasonable person could conclude that both are marketing, that neither means much, and that the sensible move is to distrust the pack either way.

A survey published in the Journal of Food Protection in 2026 gives that assumption a test. The team, from Health Canada's Food Research Division, sampled Canadian retail products built on naturally gluten-free ingredients — rice, corn, buckwheat, pulses, oats and similar — and ran them through a gliadin ELISA, the RIDASCREEN R7001 assay. The threshold of interest was the one the rest of this site keeps returning to: 20 mg of gluten per kilogram, which is 20 ppm, the ceiling Codex Alimentarius sets for a gluten-free claim.

What the shelf actually held

Across all 657 samples, 30 were above 20 mg/kg. That is 4.6% — a headline number that sounds bad until you ask where those 30 were sitting.

Among the products carrying an explicit gluten-free claim, a single sample exceeded the limit, which the authors report as 0.4% of that group. Among the products carrying a precautionary allergen statement — the may contain family — 28 samples exceeded it, or 18.9% of that group. Nearly all of the contamination in the survey was concentrated in the products that had already warned you about it.

The precautionary statement is information, not a hedge

The usual complaint about may contain labelling is that it is defensive boilerplate: no regulator defines a number behind it, no one audits it, and manufacturers are free to print it on anything to limit their exposure. All of that remains true. What this survey adds is that, whatever the motive, the wording appears to track a real difference in the product.

The plainest reading is that companies tend to add the statement when they know a genuine cross-contact route exists — a shared mill, a shared line, a shared truck — and that route sometimes deposits gluten. The warning is vague about how much and how often, which is its real weakness. It is not, on this evidence, empty.

The mirror finding is the one worth carrying into a supermarket: an explicit gluten-free claim on a product made from naturally gluten-free ingredients held up almost every time it was tested. Not perfectly, and one failure in a group is still one person's bad week. But the claim was doing work.

Oats, again

Of the 30 samples over the limit, 19 listed oats among their ingredients. That is the same story oats keep telling. The grain itself carries no gluten in the wheat, barley and rye sense, and most people with coeliac disease tolerate genuinely uncontaminated oats. The problem has never been the botany. It is that oats are grown in rotation with wheat, harvested with the same equipment, moved in the same trucks and milled in the same mills, and each of those steps is an opportunity.

This is why purity protocol or certified gluten-free oats exist as a separate category, and why an oat product without one of those designations is a different object from an oat product with one, even when the ingredient list reads identically. An ingredient list cannot describe a supply chain.

What this changes about reading a pack

It suggests the two claims deserve to be read as separate signals rather than as noise. A gluten-free claim is a statement someone is accountable for. A precautionary statement is an admission that a risk route exists, and in this data it was the single strongest predictor of a product being over the limit. Treating them as interchangeable throws away the most useful thing on the package.

It also sets a boundary on what any label-reading tool, this one included, can do. GlutenScanning reads what a product declares: the ingredient list, the allergen line, the precautionary wording, and it names which of those produced its verdict. It cannot measure parts per million, and no app on a phone can. What the Health Canada numbers offer is a way to weigh the declaration you do have — the difference between a product that claims gluten-free and one that flags a shared facility is not a matter of tone.

Limits worth naming

This is one country's retail market at one point in time, and national supply chains differ; a Chinese survey of online gluten-free products published around the same period found a much higher failure rate. The sample was deliberately weighted towards products where cross-contact is plausible, so the 4.6% is not the failure rate of the gluten-free aisle as a whole. ELISA measures gliadin and converts to gluten, which carries its own assumptions, particularly in fermented and heavily processed foods. And a single positive in the gluten-free group is a small number resting on a small denominator.

Read the full paper: Koerner TB, Cleroux C, Poirier C, Dai S, Cowell S. “Canadian Retail Survey of Gluten-Free Products and Products That Contain Inherently Gluten-Free Ingredients.” Journal of Food Protection 2026;89(6):100784, or see the PubMed record.

Common questions

Keep reading