Mention detection is more precise. We fixed cases where partially overlapping names, or responses where the AI mentioned the analysis tool itself, were miscounted as brand mentions. Aliases (English/Korean spellings and similar variants) are also recognized more consistently.
Competitor comparison billing was corrected. Certain input combinations could misalign charges and refunds — credits are now consumed only for measurements that actually completed.
Korean text on English screens is fixed. Some measurement-validation notices displayed in Korean on the English UI; they are now properly localized, with the structure changed so this class of issue cannot recur.
As detection criteria got stricter, mention counts in some measurements may differ from before — the new numbers are the more accurate ones.