What Should Food Businesses Do When AI Makes False Allergen or Gluten-Free Claims?

What should you do first?

1. Record the answer: the question, the assistant, the date and the cited sources.
2. Check what is actually true for the dish or product today, including recent recipe or supplier changes.
3. Correct your own information on your website, menus and profiles.
4. Fix the cited sources: delivery platform menus, retailer listings, product databases and old menus online.
5. Report the answer through the assistant's feedback option, describing the safety risk.
6. Brief staff, so customers who mention the AI answer get the right information in person or by phone.

How to preserve evidence properly is covered in how to document a false statement generated by an AI assistant.

What should my restaurant fix first when AI confuses gluten-free options with celiac-safe dining?

Fix the wording that lets the two be confused. "Gluten-free options" describes the ingredients of some dishes. "Safe for coeliacs" also requires controls against cross-contact in preparation, cooking and serving. Say which applies, in plain words:

• If you have controls: "Our gluten-free dishes are prepared in a separate area with dedicated utensils and fryer, and staff are trained in gluten-free handling."
• If you do not: "Some dishes are made without gluten-containing ingredients, but our kitchen is shared and we cannot guarantee they are free from cross-contact. They are not suitable for people with coeliac disease."

Then put that statement on your menu page, in your Business Profile description, and on delivery platform listings. Mark gluten-free dishes on the menu in text, not only with an icon. If you hold a recognised accreditation, such as Coeliac UK's gluten-free accreditation for venues, say so and link to the listing.

In the US, the FDA's gluten-free rule for packaged food sets a limit of less than 20 parts per million of gluten. The FDA expects restaurants that use "gluten-free" claims to be consistent with that definition. In the UK and EU, a similar limit of 20 mg per kg applies to "gluten-free" claims.

Which wording means what?

Assistants copy the words you use, so pick terms whose meaning cannot stretch:

In the UK and EU, "no gluten-containing ingredients" statements are not permitted on prepacked food, and the FSA advises caterers not to use them; if you do, state that the food is not suitable for people with coeliac disease. In the US the FDA allows "made with no gluten-containing ingredients" if it is truthful and not misleading.

Avoid "low gluten", "gluten-friendly" and "GF options" as the only description. They are the phrases most easily turned into "safe for coeliacs" in a summary, because they sound like a claim without saying what it covers.

What is the best way to publish allergen and ingredient data for AI shopping?

For packaged food and drink brands:

• An ingredient list in text on every product page, matching the label, with allergens highlighted in words, not only in bold or colour.
• An allergen statement, such as "Contains: milk, soy", and any precautionary statement, such as "May contain nuts", exactly as on the pack.
• Dietary claims only where valid, such as gluten-free, vegan or nut-free, with the basis, such as certification or testing. Certification marks such as GFCO, a programme of the Gluten Intolerance Group that requires 10 ppm of gluten or less, are specific and checkable, so name them in text.
• Per-variant data: flavours and sizes often differ in allergens. Give each its own page or clearly separate section.
• Product feeds that repeat the ingredient and allergen text in descriptions or product detail attributes, and keep images of the current label.
• Retailer data, sent through the channels retailers use, updated whenever recipes change.

In the US, food labels must declare the nine major allergens, with sesame added from 1 January 2023. In the UK, prepacked-for-direct-sale food must carry the name of the food and a full ingredient list with allergens emphasised, under the rules known as Natasha's Law. Your website and feeds should match your labels exactly. How recipe changes affect AI answers is covered in whether AI assistants can accurately tell shoppers a product was reformulated.

What should delivery platforms and third-party listings say?

Delivery apps and booking sites often carry their own dietary tags, set once when the menu was uploaded. Log in to each platform, check every dish tagged gluten-free, vegan or nut-free, and remove tags you cannot support. Copy your cross-contact statement into the restaurant description field. Where a platform will not let you edit a tag, contact its partner support and keep a record of the request. Do the same for any retailer carrying your packaged products: send the current ingredient list and allergen statement, and check the live listing after the next data refresh.

Where do false claims usually come from?

• Delivery platform menus with outdated dish descriptions or dietary tags.
• Old PDF menus still online after recipes changed.
• Reviews that say "great gluten-free options" without mentioning cross-contact.
• Product databases and retailer listings that copied an old label.
• Generic assumptions, such as "rice dishes are gluten-free", applied to your menu.

How do you test what assistants say?

Ask ChatGPT, Gemini, Perplexity and Google's AI features, several times each:

• "Is [restaurant] safe for someone with coeliac disease?"
• "Does [product] contain [allergen]?"
• "Gluten-free restaurants in [area]"
• "Is [product] nut-free?"
• "What allergens are in [dish] at [restaurant]?"

Record whether answers are accurate, whether they mention cross-contact where relevant, and which sources they cite. Retest after every menu or recipe change, and monthly otherwise.

What might an illustrative case look like?

Consider an illustrative pizzeria that offered a gluten-free base but used a shared oven and prep area. Assistants described it as "a great option for coeliacs", citing a review and a delivery app listing tagged "gluten-free". A coeliac customer became ill and complained. The pizzeria added a clear statement to its website, menu, Business Profile and delivery listing. It said the gluten-free base was made without gluten ingredients but prepared in a shared kitchen, and not suitable for coeliacs. It removed the delivery app's gluten-free tag and reported the AI answer. A month later, answers described the gluten-free base with the cross-contact warning.

What do people ask about AI and allergen claims?

Are we responsible for what an AI assistant says?

The assistant's answer is not your statement, but your published information is. Make it accurate and clear, and correct sources you control.

Should we remove gluten-free claims to be safe?

Only if you cannot support them. Clear, accurate claims with cross-contact information are more helpful than silence.

Does a certification stop AI getting it wrong?

Not on its own, but a named certification in text gives assistants a specific fact to repeat instead of guessing from reviews.

How often should allergen information be reviewed?

Whenever a recipe, supplier or preparation method changes, and on a regular schedule, such as monthly.