04Case studyKristin Ess
Kristin Ess. When AI gets your name wrong.
The brand is Kristin Ess Hair. Perplexity calls it "Kristen Ess." The Curl Shampoo, the Curl Repair Mask, the Curl Co-Wash. All of them, sometimes both spellings in the same answer. Shopthru.ai caught every instance.
- Industry
- Beauty
- Headline outcome
- 12 brand swaps
Measured 443 executions on a Perplexity-only audit. The brand-confusion class lit up.
The team ran 443 LLM queries across the 14-SKU Curl line. 12 returned a spelling error in the brand name itself. 219 of 430 ground-truth comparisons came back as mismatch (a 51% mismatch rate) concentrated on product name, ingredients, and the "for whom" claim.
- 12confirmed name swaps in one audit
- 51%mismatch rate on Curl-line ground truth
- 14SKUs across the Curl sub-brand
Simulated the disambiguation: every alias seeded into the next query set.
The team expanded the audit query set to include every observed alias ("Kristen Ess", "Kristin Ess Hair", "Kristen Ess Hair") not to validate the wrong name, but to surface the queries where it gets returned. That made the gap measurable instead of anecdotal.
Shipped 12 "Correct Brand Confusion" tickets + 34 SOURCE_OUTREACH items to authority pubs.
Each brand-confusion ticket carries the offending Perplexity response, the disambiguation copy to feed back, and the source URLs that need updating (where the wrong spelling originated). 34 SOURCE_OUTREACH items target beauty publications that propagate the misspelling.
Compounded: the brand-alias seeding pass is now standard on every audit.
Every new merchant audit now seeds two-character-edit variants of the brand name into the query generator. We catch the confusion before the merchant does. (Honest aside: even Shopthru.ai’s own marketing site has spelled it "Kristin" / "Kristen" inconsistently. The pattern is industry-wide.)