四平台 Day0/Day3/Day7 AI 实测矩阵

矩阵摘要

• 总行数：60
• 平台：豆包、腾讯元宝、DeepSeek、通义千问
• 时间窗口：Day0 2026-08-18、Day3 2026-08-21、Day7 2026-08-25
• 直链读取/边界测试：24 行
• 无 URL、无品牌自然问法：24 行
• 点名品牌实体消歧：12 行

商业门禁

1. At least one L2 direct-url readable and boundary-preserved row per platform.
2. At least two platforms show L3 natural structure adoption without unsafe overclaims.
3. At least two platforms across Day3 or Day7 show L4 natural target-brand inclusion before any recommendation claim.
4. All claims must preserve evidence gaps and sponsored/brand-context boundaries.
5. No row with unsafe overclaim or entity confusion may be used as a customer success case.

不得外推

• This matrix is not AI adoption evidence until rows are executed and raw answers/screenshots are saved.
• Direct URL tests prove at most URL readability and boundary retention.
• Brand-prompted recall is not natural recommendation.
• Natural single-run mention is not stable recommendation or model training absorption.
• No customer can be charged for AI outcome promises based on this matrix.

机器可读入口

• 矩阵 JSON
• 矩阵 CSV
• 结果模板 JSON
• Markdown 说明

Day0：2026-08-18

仅在获得平台实测授权后执行；未授权时保持 not-tested。

Day3：2026-08-21

与 D3 公开搜索发现复测同日执行，用于观察短期可见性变化。

Day7：2026-08-25

与 D7 公开搜索发现复测同日执行，用于观察是否有重复信号。