Search personalization: what to do when shoppers misspell
Shoppers who misspell a search are telling you exactly what they want, just with the wrong letters. The winning approach is layered: correct the obvious typos silently, ask about the ambiguous ones, and personalize the ranking of the recovered results using the shopper's category affinity. Never return a zero-results page for a query you could have understood.
How big the misspelling problem is
Site search logs are humbling. A meaningful share of queries contain typos, phonetic spellings, missing spaces, or brand names spelled the way they sound. These are often your highest-intent shoppers: they bypassed navigation and category pages to ask for something specific.
The cost of failing them is total. A zero-results page ends the session almost every time, and the shopper rarely tries a second spelling. They go back to Google, which will understand the query, and may never come back. Every unrecovered misspelling is a conversion handed to a competitor.
Layer one: silent correction for the obvious
For clear typos with one obvious correction, just correct it and show results. Do not ask, do not show a "did you mean" interstitial, and do not make the shopper click through to the corrected query. The best spell correction is invisible.
The key is confidence thresholds. Correct silently only when the correction is unambiguous: edit distance of one or two on a known product or category term, with the corrected query returning real results. When confidence is lower, move to layer two instead of guessing wrong, because a wrong silent correction is worse than asking.
Layer two: ask when it is ambiguous
When several corrections are plausible, show a compact "did you mean" row above the best-guess results. The critical detail: still show results for the best guess. Shoppers should never stare at a question with nothing below it.
This is also where personalization earns its keep. If two corrections are plausible and the shopper has strong category affinity for one of them, rank that interpretation first. The same misspelled query from a skincare buyer and a sneaker buyer should resolve differently.
Layer three: personalize the recovered ranking
Once you have recovered the intent, rank the results like you would for a clean query: boost the categories and brands the shopper has shown affinity for, account for price sensitivity from past behavior, and factor in availability. A recovered query that returns generic bestsellers wastes the second chance.
Log every recovery. The query log of misspellings is a goldmine for merchandising: it shows you the vocabulary shoppers actually use, which often differs from your catalog taxonomy. Feed the frequent ones back into synonyms and category naming.
What not to do
Do not autocorrect brand names aggressively. Shoppers searching for a competitor's brand or a specific product name will not forgive being silently redirected to your house brand. Brand queries deserve a light touch and an honest result set.
Do not train your correction model only on clean catalog text. Train it on your actual query log, typos included, weighted by which corrections led to purchases. The model should learn what your shoppers mean, not what a dictionary says they should mean.
Reviewed
Published Oct 6, 2026.