这里展示的是有项目范围、有日期、有样本数的可回查记录。不同品类、不同问法不能拼成一条“增长曲线”；未完成同口径复测，就只展示基线和执行路线。These are auditable records with scope, date and sample size. Different categories and queries are not merged into a growth curve; without like-for-like re-testing, we show only the baseline and execution route.
每次测试必须同时保留平台、问法、轮次、地区 / 语言与日期。单张截图不算结果，被提及也不等于采购认可。Every test retains platform, query, run, region / language and date. One screenshot is not a result, and a mention is not procurement approval.
在指定问题库与测试范围内，企业名称被正确提到的比例。它是基线指标之一，不等于采购短名单或成交。The share of scoped answers that correctly mention the company. It is one baseline measure, not a procurement shortlist or sale.
提到企业时，是否给出真实官网，还是缺少链接、指向第三方或生成错误域名。Whether a mention includes the real official site rather than no link, a third party or a fabricated domain.
同一批问法里，AI 反复提到哪些企业；这是一组答案内竞品，不直接等于现实采购竞争格局。Which companies recur across the same queries; these are answer-level rivals, not the complete real-world competitive set.
答案实际引用或明显依赖了哪些官网、目录、媒体与社区页面，用于定位证据缺口。Which sites, directories, editorial and community pages are cited or clearly relied on, revealing evidence gaps.
记录答案如何描述能力、规格、市场与限制，判断是抓取、理解、匹配还是核验层出了问题。Record how answers describe capabilities, specifications, markets and limits to identify crawl, understanding, match or verification gaps.
同一企业不同品类可以得到完全不同的答案。下面只陈述测试文件支持的数字，并把“已验证”与“待执行 / 待复测”分开。Different categories for one company can produce very different answers. The numbers below are supported by archived test files, with verified facts separated from planned work and pending re-tests.
8 类买家问题 × 3 平台 × 5 轮 = 120 条
已验证：120 条答案中企业出现率 0%，测试覆盖 ChatGPT、Perplexity、Gemini。
执行路线：完善香水玻璃瓶直答内容、结构化数据与实体一致性，再补独立信源。
状态：尚无同口径优化后复测，因此不宣称“已进入推荐池”。Verified: 0% across 120 answers on ChatGPT, Perplexity and Gemini. Route: improve direct-answer content, structured data, entity consistency and independent sources. Status: no like-for-like post-build re-test, so no recommendation claim.
吸油纸专项 · 2026-07-21 · 40 条 · 4 平台
吸油纸：Gemini 100%、Perplexity 70%、ChatGPT 30%、Claude 20%，官网地址准确性 100%。
痘痘贴：另有 2026-06-29 的 20 条独立基线，出现率 0%。
边界：两组样本的品类、问法和日期不同，不能相减为“提升 55 个点”。Blotting paper: Gemini 100%, Perplexity 70%, ChatGPT 30%, Claude 20%, with 100% official-domain accuracy. Acne patch: a separate 20-record baseline dated 29 Jun 2026 was 0%. Boundary: different scopes cannot be subtracted into a growth claim.
2026-07-01 · 4 条基线测试 · 0/4 出现
已验证：4 条基线均未出现企业，且早期定位把“软胶囊代工”和“软胶囊包装设备”混在了一起。
纠偏：“整线 turnkey”已有强势既有企业，并非无人竞争；更合理路线是护肤 / 美妆软胶囊、小批量与 OEM 到整线的分层承接。
状态：执行路线已建立，优化后复测待完成。Verified: the company was absent from all four baseline tests, while early positioning mixed softgel manufacturing with packaging equipment. Correction: turnkey has established incumbents; the route is skincare / beauty softgels, small batches and a staged OEM-to-line offer. Status: route defined, post-build re-test pending.
数据边界：一次问法 × 一个平台 × 一轮计 1 条。出现率仅对该项目范围有效；AI 答案会随平台、地区、语言与日期变化，不承诺固定排名或推荐。Data boundary: one query × one platform × one run = one record. Mention rate applies only to the stated scope; answers vary by platform, region, language and date, with no fixed-rank or recommendation promise.