谷歌的「人工智慧总览」如今常常取代经典连结清单出现，虽被包装成快速摘要，却可能带出有害的错误资讯，包括诈骗电话号码。连线杂志在 2026年2月15日 的一篇报导指出一种模式：使用者搜寻公司联络方式、信任人工智慧提供的号码，最后打到冒名者那里，对方试图索取付款或个人资料。核心风险不只是一般的人工智慧失误，而是把电话号码这类具体事实以高度确信的方式呈现，哪怕只错 1 个数字，也可能把来电导向诈骗。
该报导至少引用了 2 项外部媒体调查（华盛顿邮报与数位趋势），并汇整了至少 2 个社群平台（脸书与红迪）的使用者回报，另外还有信用合作社与银行等金融机构的警示。较可能的机制是大规模资料投毒：诈骗者在多个低能见度页面发布假号码，而人工智慧系统在缺乏足够验证下将其吸收并重新包装。谷歌表示反垃圾机制正在更新，并声称成效良好；但研究人员也展示了电子邮件摘要中的类提示注入操弄，以及其他人工智慧搜寻产品中的相似问题。
实务上的重点是养成 2-step 验证习惯：先找到公司的官方网站，再在拨打前至少用 1 次额外搜寻确认任何号码，特别是涉及付款或帐户细节时。文章强调目前无法完全关闭人工智慧总览，因此即使平台防御持续改进，使用者端的谨慎仍是必要条件。从趋势看，随著搜寻越来越由人工智慧中介，便利性提高，但信任门槛也同步升高；对银行、客服与旅游预订等高风险查询，应优先依赖第一手来源，而非综合生成的答案。
Google’s AI Overviews, now often shown instead of the classic link list, are framed as fast summaries but can surface harmful misinformation, including scam phone numbers. In a February 15, 2026 article, WIRED describes a pattern where users search for a company contact, trust the AI-provided number, and reach impostors seeking payment or personal data. The core risk is not just ordinary AI errors but high-confidence presentation of specific facts like phone numbers, where even 1 wrong digit can redirect a call to fraud.
The reporting cites at least 2 external media investigations (The Washington Post and Digital Trends), with user reports on at least 2 social platforms (Facebook and Reddit), plus warnings from financial institutions such as credit unions and banks. A likely mechanism is data poisoning at scale: scammers publish fake numbers across multiple low-visibility pages, and AI systems ingest and repackage them without sufficient verification. Google says anti-spam systems are being updated and claims strong effectiveness, but researchers also show similar prompt-injection style manipulation in email summarization and comparable issues in other AI search products.
The practical takeaway is a 2-step verification habit: find the company’s official site and confirm any number with at least 1 additional search before calling, especially for payments or account details. The article stresses that AI Overviews currently cannot be fully turned off, so user-side caution remains mandatory even as platform defenses improve. Trend-wise, as search becomes more AI-mediated, convenience rises but trust requirements also rise, and high-stakes queries (banking, support, travel bookings) should prioritize primary sources over synthesized answers.