{"id":33423,"date":"2026-11-14T20:52:20","date_gmt":"2026-11-14T19:52:20","guid":{"rendered":"https:\/\/wooacademy.sk\/?p=33423"},"modified":"2026-11-14T20:52:20","modified_gmt":"2026-11-14T19:52:20","slug":"woocommerce-fuzzy-search-typos","status":"publish","type":"post","link":"https:\/\/wooacademy.sk\/en\/woocommerce-fuzzy-search-typos\/","title":{"rendered":"Fuzzy search in WooCommerce: a store that finds products despite typos"},"content":{"rendered":"\n

Fuzzy search is search that finds results even when the query does not match the product text exactly.<\/strong> A shopper types “shmapoo”, “shampo” or “sampoo” and the store still shows shampoos. It works by counting how many edits (insertion, deletion, substitution or swapping letters) turn one word into another and allowing a small number of errors.<\/p>\n\n

Why it matters more than it seems<\/h2>\n

People on mobile type fast, with a thumb, and often without accents. Brand names and foreign words get spelled the way they sound. Baymard Institute found that 69% of e-commerce sites<\/strong> fail to show relevant autocomplete suggestions for slightly misspelled queries. In the same research 56% of sites<\/strong> overall failed to support what people actually type into search.<\/p>\n

The outcome is simple: the shopper sees “No products were found” and leaves, even though you have the product in stock.<\/p>\n\n

How fuzzy search works<\/h2>\n

Edit distance<\/h3>\n

The basis is Levenshtein distance: the number of letter insertions, deletions and substitutions. “jaket” and “jacket” have distance 1 (one missing letter), “jackte” and “jacket” distance 2 in plain Levenshtein.<\/p>\n

When typing on a keyboard, swapping two adjacent letters is one of the most common mistakes (“jackte”). Classic Levenshtein counts it as two errors. The variant with transpositions (Damerau-Levenshtein, in its restricted form called OSA, optimal string alignment) counts it as one. For a store this is a real difference, because short words usually allow only one error.<\/p>\n

How many errors to allow<\/h3>\n

Too much tolerance hurts. For the word “cap”, two errors would also match “map”, “cape” or “car”. A sensible rule is no error for very short words, one for medium words and two for long ones.<\/p>\n

Where the correct words come from<\/h3>\n

The search engine does not correct against a general dictionary but against words that actually occur in your catalogue. So it corrects “bosh” to “bosch” if you sell Bosch, and it will not invent words you do not have.<\/p>\n\n

Diacritics: the most common error that is not a typo<\/h2>\n

In Slovak, Czech, German, French and many other languages, diacritics are a topic of their own. A shopper who types “cerveny” instead of “\u010derven\u00fd” did not make a mistake, they just skipped the accents. Search should normalise both the product text and the query (remove accents) before comparing and only then look for typos. Otherwise a missing accent counts as an error and uses up the whole tolerance on a longer word.<\/p>\n\n

Did you mean: the correction the shopper sees<\/h2>\n

If the engine finds results only for the corrected word, it should tell the shopper. For example “Showing results for shampoo. Search instead for shmapoo?”. The shopper then knows why they see what they see, and for a genuinely unusual name (a code, a model) they can go back to the original query.<\/p>\n\n

When fuzzy search does not help<\/h2>\n