返回方法论Back to method
洞察 · 平台分诊Insight · Platform Split

同一企业,在不同 AI 的出现率
可以从 0% 到 57.1%
One company can range from
0% to 57.1% across AIs

这不是“一个平台排第几”的固定排名,而是指定日期、问法与轮次下的答案差异。必须逐平台记录,才能知道缺口发生在哪里。This is not a fixed cross-platform rank; it is an answer difference under a stated date, query set and run rule. Per-platform records show where the gap occurs.

57.1%
Gemini · 4/7Gemini · 4/7
0%
ChatGPT / PerplexityChatGPT / Perplexity
4
本组实测平台platforms in this sample
28
可回查答案 · 2026-06-29auditable answers · 29 Jun 2026
← 六大 AI 平台逐平台操盘手册
问题The problem

「我们 AI 上排第几?」——其实没有单一答案"Where do we rank in AI?" has no single answer

生成式答案不是一个长期固定榜单。平台的搜索模式、索引、引用能力、地区、语言与当日答案都可能不同。所以专业测试不回答“全网第几”,而是回答:在哪个平台、哪类问法、哪次测试被提到,官网是否准确,竞品和信源是谁,以及为什么被提到或忽略。Generative answers are not a permanent league table. Search mode, index, citation support, region, language and date can all differ. An expert audit therefore reports where and when you appeared, official-domain accuracy, rivals, sources, and the context behind inclusion or omission.

真实样例(已脱敏)A real example (anonymized)

一家中国美妆 OEM 的四平台基线A four-platform baseline for one Chinese beauty OEM

2026-06-29,共 7 类问法 × 4 平台 = 28 条答案。下列结论只对这组样本有效。Dated 29 Jun 2026: seven query groups × four platforms = 28 answers. Findings apply only to this scope.

Gemini 4/7 · Claude 2/7Gemini 4/7 · Claude 2/7

Gemini 出现率 57.1%,Claude 为 28.6%。这证明企业实体与部分能力在某些答案环境里可被找到,但不能据此推断“行业第一”或所有问法都已覆盖。Gemini mention rate was 57.1% and Claude 28.6%. This shows some entity and capability visibility, but does not prove an industry-leading position or complete query coverage.

⚠️

ChatGPT 0/7 · Perplexity 0/7ChatGPT 0/7 · Perplexity 0/7

同一组 7 类问法里,两平台均未提到企业。基线只能确认“当次没有出现”,不能仅凭结果反推某个平台固定只看哪类来源。Neither platform mentioned the company across the seven query groups. The baseline confirms absence in this run, but does not by itself prove a platform always uses one source type.

💡 可支持的诊断:不同平台存在明显差异,下一步要回到具体答案逐条检查官网地址、竞品、引用信源和描述上下文,再判断优先补官网事实、实体一致性还是独立证据。不能只凭 0% / 57.1% 直接归因。💡 Supported diagnosis: platforms differ. Next, inspect official domains, rivals, cited sources and answer context before choosing whether to fix site facts, entity consistency or independent evidence. Rates alone do not prove causation.

COSGEO 怎么做How COSGEO works

4 步把「玄学」变成可执行的分诊单4 steps from guesswork to a diagnosis

🧬

① 裂变问法Fission queries

把「找你这类工厂」的真实采购问题,按市场 × 买家 × 场景裂变成几十道 AI 问法。Split real sourcing questions into dozens of queries by market × buyer × scenario.

🛰️

② 约定平台与轮次Agree platforms and runs

按企业目标选择平台,固定问法、地区 / 语言、日期与重复轮次。Choose platforms for the business and fix queries, region / language, date and repeat runs.

📊

③ 5 类审计结果5 auditable outputs

出现率 / 官网地址准确性 / 竞品 / 信源 / 被提到或忽略的上下文原因。Mention rate / official-domain accuracy / rivals / sources / inclusion or omission context.

🧭

④ 平台分诊Platform split

拆开每个 AI 单独看,找出「哪个平台赢、哪个平台盲」,把钱花在刀刃上。Read each AI on its own — find which platform wins and which is blind, and spend where it counts.

价值Why it matters

有了分诊单,行动不再靠拍脑袋A diagnosis ends the guesswork

问「AI 上排第几」"Where do we rank in AI?"
说不清unclear
逐平台一目了然clear per AI
预算怎么花Where to spend
盲目做内容/SEOblind spend
只补真正缺口gap only
对手是谁Who are rivals
不知道unknown
AI 正在推荐的每一家all named
改了有没有用Did it work
无从验证no proof
按约定周期同口径复测like-for-like re-test

说明:样例来自 2026-06-29 的真实测试记录,企业身份与具体问法已脱敏。答案会变化;该数据不是当前实时排名,也不保证未来推荐。Note: this anonymized example comes from real records dated 29 Jun 2026. Answers change; this is not a current live rank or a future recommendation guarantee.

想知道你在哪些问法被提到,
又在哪些平台缺席?
Where are you mentioned,
and where are you missing?

申请专业基线测试 →Request an expert baseline audit →

首选加微信 13925122895(微信同名)WeChat 13925122895