Retrieval Mechanisms of Local Businesses in the Generative Search Era: Third-Party Footprints, First-Party Anchors, and the MCAC Empirical Evaluation Framework

English Abstract This exploratory study examines six purposively selected beef noodle merchants in Taipei using the Perplexity Sonar API on 16 September 2026. Of 24 retained observation envelopes, 21 were included: three discovery, six fact, six action, and six order-counterbalanced comparison responses. Three merchants had verified official domains and three had no domain confirmed in the study. Discovery responses contained 9/9 and 6/9 brand-run mentions, respectively; these counts come from three responses with incomplete endings and are not independent trials. Address and telephone agreement with the browser-compiled reference table was 6/6 per group (12/12 total). The source lists contained 408 entries, 206 unique literal URLs, and 112 normalized hostnames. Six unique URLs and ten entries belonged to the sample official domains (2.91% and 2.45%, respectively); these shares do not measure answer contribution or trust. All six action responses supplied basic contact information, while official URLs appeared in 3/3 website-group responses and 0/3 unverified-domain responses. Comparison completeness was 6/6 for the first merchant and 0/6 for the second, including after order reversal. We organize these observations as Mention, Citation, Accuracy, and Conversion-readiness (MCAC). The pilot identifies measurement distinctions and limitations; it does not estimate the causal effect or financial return of website ownership, isolate third-party source contributions, or establish behavior of other AI products. 中文摘要 本研究以 2026 年 9 月 16 日的 Perplexity Sonar API 留存紀錄,探索台北市六家目的性選取的牛肉麵商家。三家確認具備官方網域,三家於本次查核未確認自有網域。24 筆觀測封裝中納入 21 筆,包括推薦 3 筆、事實 6 筆、行動 6 筆及正反序比較 6 筆。推薦回答的品牌×執行提及數分別為 9/9 與 6/9;三次回答均有尾端未完成情形,不能視為獨立試驗。地址與電話對照瀏覽器查核參照表,各組為 6/6,合計 12/12。來源清單含 408 個項目、206 個依原始字串去重的網址及 112 個正規化主機名稱;其中樣本官方網域占 6/206(2.91%)與 10/408(2.45%),不代表答案貢獻或信任權重。兩組各 3/3 的行動回答均提供基本聯絡資訊,官方網址則為 3/3 對 0/3。成對比較中,第一順位商家的地址與電話完整度為 6/6,第二順位為 0/6,反序後亦然。本文以提及、引用、準確性與行動就緒度(MCAC)區分量測對象;本試驗未估計建站的因果效果或投資報酬,也未分離第三方來源的個別貢獻或測試其他 AI 產品。