不是"我们做了很多"，而是一条可以问责的曲线Not "we did a lot", but a curve you can hold us to
同一套问题集、同一个抓取节奏、同一种判定规则，定时自动跑一次。优化有没有效果，看图就知道。One question set, one cadence, one set of rules — run automatically on a fixed schedule. Whether the work is paying off is visible at a glance.
怎么保证数据可信How we keep it trustworthy
生成式回答天然带随机性，单次提问说明不了任何问题。每个问题在每个平台重复取样多次取中位数；能通过官方接口联网提问的平台走接口，没有开放接口的按真实网页端采集，并记录每一条答案的原文与时间戳，随时可回溯复核。Generative answers are inherently stochastic — a single query proves nothing. Each question is sampled multiple times per platform and reduced to a median. Platforms with an official web-enabled API are queried through it; the rest are collected from the real web client. Every raw answer is stored with a timestamp, so any number can be traced back.
口径一旦确定就不改The ruler does not change
问题集、取样次数、判定规则在项目启动时确定并写进文档。中途换尺子，前后就不可比。The question set, sample count and matching rules are fixed at kickoff and written down. Change the ruler mid-way and nothing is comparable.
五个指标，每个都写明怎么算Five metrics, each with its arithmetic written down
下面是我们实际统计的五个量，以及它们各自的分母。The five quantities we actually count — and what each one is divided by.
- 提及率Mention rateMention rate
- 问题集中，主动提到该品牌的答案条数 ÷ 该平台上的有效答案总数。不区分正面负面，只看有没有被想起来。Answers that bring the brand up, divided by all valid answers collected on that platform. Sentiment is not counted here — only whether you came to mind at all.
- 引用率Citation rateCitation rate
- 把客户自有站点列为来源的答案条数 ÷ 有效答案总数。被提到不等于被引用——引用意味着模型认可你的站点本身是可信信源。Answers citing the client's own site as a source, divided by all valid answers. Being mentioned is not being cited: a citation means the model treats your site itself as a credible source.
- 答案份额Share of answerShare of answer
- 同一问题的答案中，描述该品牌的字数 ÷ 描述所有同类主体的总字数。用来看你和竞品在同一个答案里各占多少位置。Characters describing the brand, divided by characters describing all comparable entities in the same answer. It shows how much room you occupy next to competitors inside one answer.
- 来源构成Source mixSource mix
- 全部答案所引用域名的频次分布。这一项不作为考核指标，而是决定下一阶段内容发到哪里的依据。The frequency distribution of domains cited across all answers. Not a KPI — it is the input that decides where the next round of content goes.
- 表述准确性Factual accuracyFactual accuracy
- 按统一基准文档逐条比对答案中关于该品牌的陈述，标记过时、错误、张冠李戴三类问题，输出问题清单而不是一个分数。Every statement about the brand is checked line by line against the canonical fact sheet and flagged as outdated, wrong, or attributed to the wrong entity. The output is a list of problems, not a score.
监测哪些平台，按什么节奏Which platforms, on what cadence
默认覆盖中国大陆主流的四个生成式平台，全部使用联网模式提问——关掉联网测的是模型的记忆，打开联网测的才是今天的客户会看到什么。Four mainstream platforms in mainland China by default, all queried with web access on. Querying without it measures the model's memory; querying with it measures what a customer would see today.
豆包 · DeepSeek · 通义千问 · 腾讯元宝Doubao · DeepSeek · Qwen · Yuanbao
四个平台同一套问题集、同一联网设置。有官方接口的走接口，其余按真实网页端采集，两种方式在报告里分别标注。One question set and one web-access setting across all four. Platforms with an official API are queried through it; the rest come from the real web client, and the report labels which is which.
定时采集，固定时段On a schedule, at a fixed hour
答案随时间波动，固定时段采集才能把"平台当天状态"这类噪声压下去。具体周期按项目约定，采得越密成本相应上升。Answers drift over time; collecting at a fixed hour suppresses day-of-week and time-of-day noise. The cadence is agreed per engagement; denser collection costs more.
答案原文全量留档Every raw answer archived
每条答案连同时间戳、平台、问题编号一起存档。看板上任何一个数字都能点回它背后的原始答案。Each answer is stored with its timestamp, platform and question ID. Any number on the dashboard can be traced back to the raw answer behind it.
看板演示Dashboard demo
一份公开的看板演示：数据模型、指标卡片、来源构成与采集器状态。里面的数字全部是示意数据，页面上已标注，不代表任何真实客户。A public demo of the dashboard: data model, metric cards, source mix and collector status. Every figure in it is illustrative and labelled as such on the page, and represents no real client.
常见问题Common questions
- 多久能看到变化？How soon does it move? 信息一致性类问题通常 2–4 周见效；引用率的变化取决于内容被收录的速度，一般 6–12 周。Consistency fixes usually show in 2–4 weeks; citation rate depends on how fast new content gets indexed, typically 6–12.
- 能监测竞品吗？Can competitors be tracked? 可以。竞品同现与答案份额本来就在同一套采集里，不额外增加成本。Yes. Competitor co-mentions and share of answer come from the same collection run, at no extra cost.
- 数据归谁？Who owns the data? 归客户。合作结束时答案档案与看板数据可完整导出。The client. The full answer archive and dashboard data can be exported when the engagement ends.
先看看 AI 现在怎么说你First, see what AI says about you today
免费出一份基线报告：四个平台、一套问题集、真实答案原文。A free baseline report: four platforms, one question set, real raw answers.