Allergen errors often enter the system at the quietest moment: someone types a new special, pastes a supplier description, or updates a sauce recipe in a hurry. AI ingredient checks are useful here — not as a magic compliance stamp, but as a second pair of eyes on text humans skim.
Where mistakes sneak in
- Recipe descriptions that mention "soy sauce" without flagging cereals containing gluten or soy
- "Creamy" language that implies milk without a milk allergen flag
- Garnishes listed in the blurb but missing from the structured ingredients
- Copy-pasted dishes from old menus with outdated flags
Humans miss patterns when busy. Software is good at pattern hints — if you treat them as hints to verify.
What AI checks are good for
Quiteful's AI ingredient checks help surface likely allergens from recipe wording before a dish is published. That is valuable when:
- You onboard a large legacy menu
- Chefs write free-text methods
- Marketing edits descriptions without kitchen review
- You need a consistent first pass across many SKUs
What AI checks are not
- Not a substitute for supplier specifications
- Not a guarantee of legal compliance on their own
- Not a cross-contact risk assessment
- Not permission to skip human sign-off on high-risk dishes
If the model suggests sesame and the chef knows the bun is plain, reconcile the data — do not blindly accept or blindly ignore.
A sensible workflow
- Enter or import the recipe and description
- Run ingredient checks for likely allergens
- Confirm or correct structured allergen flags
- Publish to the live menu only after review
- Re-run when the recipe text changes materially
Pair this with goods-in checks on compound ingredients. AI reads your words; specs confirm the tub.
Governance for teams
Decide who can publish dishes with unresolved AI suggestions. A simple rule: no live publish while critical allergen prompts are still open. That prevents "we'll fix it later" from becoming the guest-facing menu.
Multi-site menus
Central teams can use AI checks when rolling out new group recipes, then require site managers to confirm local overrides. The combination of multi-location control and early text checks reduces estate-wide blind spots.
Keep guests in mind
AI is an internal quality tool. Guests should see clean, verified allergen data and filters — not experimental labels. The win is fewer incorrect dishes reaching the QR code in the first place.
Bottom line
Use AI to catch likely misses early; use chefs and specs to decide the truth; use a live menu system to publish only what you have verified. That sequence turns allergen management from reactive firefighting into quieter routine.