AI Search does not replace exact matching.
Lexical search compares words, word parts and controlled variants. For article codes, brand names, models and specific product terms, that is often the most reliable route. Modern AI search becomes stronger when this precision is maintained.
What exactly lexical search does
Lexical search searches for similarity between the entered terms and indexed product fields. Word forms, typos, compositions and controlled synonyms can be included. The system explicitly knows which terms match and in which field they were found.
This makes lexical search transparent and strong for concrete searches. A model code in the SKU field can outweigh the same character pattern in a long description. A brand match can take precedence over an accidental mention elsewhere.
Exact identifiers first
SKUs, EANs, part numbers and model codes have little natural significance, but are commercially important. A customer who enters such a code usually expects one exact product or a small group of suitable variants.
These values should be protected against aggressive spelling correction and semantic widening. A character difference may refer to another section. Exact matches should therefore be given a separate, high priority and a clear regression test.
Brands and product-specific terms
Brands, series and fixed product names also benefit from lexical precision. When a customer explicitly mentions a brand, a semantically similar alternative should not displace that signal without clear reason.
Synonyms and aliases can help with well-known writing variants. Manage that targeted: an abbreviation can be correct in one category and have a different meaning in another. Category context prevents a broad synonym rule from touching the entire range.
Typos and word variants
Lexical search can be tolerant without letting go of all boundaries. Edit distance, keyboard errors, split or merged words and common misspellings can be verified. The degree of tolerance should depend on query length and term type.
With a long natural term, a slight correction is often safe. With a short brand name or code, the same correction can yield too many candidates. Confidence and field knowledge therefore determine whether a correction is applied automatically, if suggestion appears or is not used at all.
Ranking with fields and weights
- An exact SKU match is usually given priority.
- A match in title or brand field outweighs a separate description match.
- Multiple matching key terms reinforce each other.
- Rare distinctive terms can bear more meaning than general words.
- Business rules and stock can adjust the final order.
The weights must remain understandable and testable. A small change to analyzers or field boosts can affect many queries; fixed golden queries protect the existing quality.
Lexical and semantic complement each other
Lexical search delivers precision, semantic matching increases the recall in meaningful descriptions. Hybrid retrieval combines candidates from both sources and applies query-dependent priorities.
At “AB-1234”, lexical dominates. With “lamp for reading without bright ceiling light” semantic can contribute more. At “Bosch silent dishwasher 60 cm” brand and size are exact, while “silent” can be interpreted semantically or via attributes.
What you should measure
Analyze exact match rate, zero-result searches, corrections, reformulations and the position of known products. Segment on codes, brands, categories and natural language. This way you can see whether a broad improvement elsewhere is at the expense of concrete searches.
Findoviq maintains lexical search as a verifiable base and combines it only where necessary with semantic signals. This keeps the search function strong for customers who know exactly what they are looking for and can be used for customers who describe their needs.
Discuss your search questions
Do you want to know how this approach fits your assortment, product data and customer behavior? Together, we look at which query types have priority and where exact, semantic and business signals need to complement each other.