Grocery Dietary Search: Enforce Shopper Constraints
Design grocery search with verified dietary attributes, explicit exclusions and careful missing-data handling. Separate semantic relevance from eligibility.
Dietary and attribute search in grocery: enforcing shopper constraints
Dietary language contains different kinds of requirements. A shopper’s preference, a product’s ingredients and a verified free-from claim are not interchangeable. Build search around reliable catalogue evidence, explicit rules and clear uncertainty, then rank products only within the assortment that satisfies the agreed constraints.
Preserve what the shopper asked for
Separate the requested product type from attributes and exclusions. “Vegan biscuits” asks for a dietary property. “Milk-free biscuits” requires a different evidence check. A broad semantic association between those phrases is not permission to treat their eligible product sets as identical.
Official UK guidance explains that vegan labelling does not establish suitability for a person with an allergy to an animal-derived ingredient. Products can carry cross-contamination warnings. Vegan food labelling guidance.
Give the catalogue an evidence model
For each relevant attribute, record its value, source, scope and update time. Preserve authoritative ingredient and precautionary statements rather than replacing them with a generated summary. Distinguish verified yes, verified no, unknown and conflicting evidence. An absent field must not become a positive suitability assertion.
Keep the product variant and market context attached to the evidence. Similar packaging or a shared brand is not enough to transfer a claim from one formulation to another. Give the catalogue owner a process for resolving changes and contradictions.
Apply explicit eligibility before relevance ranking
The proposed design is to interpret the query, resolve the constraints against approved attributes, select eligible products and then rank for relevance. Do not let a popularity boost or sponsored placement override an essential exclusion. PA documents text and numeric search filters; the exact dietary mapping and its failure behaviour must be agreed and tested for a deployment. Search filter documentation.
In a synthetic example, Biscuit A is labelled vegan but has a “may contain milk” statement. Biscuit B has retailer-approved evidence for the milk-free attribute. Biscuit C has no verified dietary data. Under an explicitly approved milk-free filter, B is eligible; A and C must not be silently included as equivalents. This is a test fixture, not advice about a real product.
Explain limits without weakening the condition
If the eligible set is empty, keep the condition visible and explain the result. Offer a way to review the query or inspect product information. Do not automatically broaden an allergy-related requirement to avoid an empty page. The distinction between free-from and precautionary information should remain available in the shopping journey. Allergen labelling guidance.
Test the failure cases explicitly
Build fixtures for missing attributes, conflicting evidence, changed formulations, alternative wording and combined conditions. Have the retailer's catalogue and food-information owners approve expected results. Repeat the checks after feed or mapping changes. A high average relevance score does not excuse including an item that violates a required constraint.
Illustrative example
Synthetic policy fixture: A {vegan:verified_yes,milk_free:verified_no,may_contain_milk:true}; B {vegan:verified_yes,milk_free:verified_yes,may_contain_milk:false}; C {vegan:unknown,milk_free:unknown,may_contain_milk:unknown}; D {vegan:verified_yes,milk_free:verified_yes,may_contain_milk:true,conflict:true}. For the fixture’s approved vegan policy A/B qualify, D requires conflict review. For its milk-free policy only B qualifies; C unknown and D conflict do not pass. These are invented records and a proposed test policy, not clinical advice or real product assertions.
Downloads
- dietary-search-fixtures.json (illustrative template)
Sources
- GOV.UK: Vegan food labelling, checked 2026-10-25
- GOV.UK: Allergen labelling, checked 2026-10-25
- Facets & Filters, checked 2026-10-25
How Particular Audience does it
Frequently asked questions
Can semantic similarity establish dietary suitability?
No. Language understanding can help interpret the query, but suitability requires the relevant verified product information and approved rules. Keep unknown or conflicting evidence explicit. Do not infer an allergen-related claim from a title, image or embedding similarity alone.
Should vegan and milk-free use the same filter?
No. They describe different things. Official UK guidance explains that vegan products may carry allergen cross-contamination warnings. Define each filter with the retailer’s food-information owner and preserve the product’s relevant ingredient and precautionary statements.
What should happen when dietary data is missing?
Do not treat missing evidence as a positive match. Use an explicit unknown state, withhold unsupported suitability claims and direct the data issue to its owner. The chosen display behaviour should preserve the shopper’s constraint and be tested before release.
Related guides
- Product-data requirements for AI ecommerce search: Prepare product data for AI search with a field checklist covering identity, variants, attributes, prices, stock, update behaviour and measurement joins.