← Knowledge Notes
Global Growth / Knowledge note · Chinese

GSC Blog: AI 搜索演化与 Google 对 AI 内容的立场(2015-2026)

Google Search Central blog 18 篇 AI 内容立场 / RankBrain / BERT / MUM / Knowledge Graph / AI Overviews / Succeeding in AI Search 的主题化提炼。

Source collection:Google 搜索与 SEO · Published here:2026-09-26

SEO搜索引擎自然增长

关于候选语料的说明 / Caveat 本批次 18 篇候选里,真正命中 "AI / ML / Knowledge Graph / AI Overviews" 主题的只有 8 篇(2007 auto-generated / 2015 Knowledge Graph events / 2017 user-generated spam / 2021 page titles / 2023 google-search-and-ai-content / 2023 definitely-not-ai-generated-post / 2025 succeeding-in-ai-search / 2025 ai-powered-configuration)。 其余 10 篇(2005 sitemap generator v1.3 / 2005 OAI-PMH / 2005 sitemap v1.4 / 2008 retire OAI-PMH / 2009 sitemap generator / 2009 caffeine sandbox / 2012 crawl errors / 2017 Vietnamese 公告 / 2018 German 公告 / 2026 SCL Shanghai)与 AI 主题无关,本 wiki 不展开,仅在源列表里登记以保留 ingest 完备性证据。 2026-05-18 更新:RankBrain (2015) / BERT (2019) / MUM (2021) / HCU (2022) / AI Overviews (2024) 5 大里程碑官方原文已通过 blog.google (The Keyword) 抓回,见 Google Search 5 大 AI 算法里程碑(RankBrain/BERT/MUM/HCU/AIO)。这些公告不在 Google Search Central blog,在 blog.google 主站,所以原 GSC scraper 没抓到 —— 现已独立归档到 sources/reference/google-the-keyword/blog/。Part 2 / Part 3 的骨架内容现在可以用 Google Search 5 大 AI 算法里程碑(RankBrain/BERT/MUM/HCU/AIO) 的官方原文校验/扩展。


Part 1 — Google 对 AI 生成内容的官方立场("how, not who" 框架)

1.1 核心命题:质量先于"是谁/什么写的"

来源:2023/02 Google Search's guidance about AI-generated content(Danny Sullivan + Chris Nelson)

"Our focus on the quality of content, rather than how content is produced, is a useful guide that has helped us deliver reliable, high quality results to users for years."

历史类比(Google 原话):

"About 10 years ago, there were understandable concerns about a rise in mass-produced yet human-generated content. No one would have thought it reasonable for us to declare a ban on all human-generated content in response. Instead, it made more sense to improve our systems to reward quality content."

逻辑:2011 Panda 时代 Google 没有 ban "人写但批量低质",而是升级系统识别质量;面对 LLM 时代 Google 重申同样路径,不 ban AI 内容,而是评估质量。

1.2 边界:scaled content abuse(规模化内容滥用)

"Using automation—including AI—to generate content with the primary purpose of manipulating ranking in search results is a violation of our spam policies."

判定的是意图不是工具:

  • ✅ AI 生成天气、体育、转录 → 长期合规
  • ✅ AI 辅助创造性表达 → 合规
  • ❌ AI 以"骗排名"为首要目的批量产出 → spam policy 违规

执法系统:SpamBrain(2022 webspam report 提到,本 wiki 引用自 2023/02 那篇内嵌链接)。

1.3 给创作者的官方建议(FAQ 提炼)

来源:2023/02 FAQ 段(8 问 8 答)。

问题 Google 立场(原话精简)
AI 内容违规吗? 不违规——除非用来操纵排名
为什么不直接 ban AI 内容? 自动化在出版业一直有合法用途
怎么防止低质 AI 内容刷屏? 用已有的 helpful content / original news / spam 系统应对,这不是新问题
AI 内容会优待吗? "Using AI doesn't give content any special gains. It's just content."
我该用 AI 写吗? 用 AI 帮你产出真正有用的原创 = 可以;把 AI 当便宜的刷排名工具 = 不要
要署名作者吗? 当读者会问 "Who wrote this?" 时应该署
要披露 AI 参与吗? 当读者会问 "How was this created?" 时应该披露
能把 AI 列为作者吗? "probably not the best way"——按"如何创建"维度披露更合适,而不是把 AI 当作者

1.4 Who / How / Why 三问框架

Google 在 Creating helpful, reliable, people-first content 帮助页里加了新框架:Who 写的、How 写的(含是否用 AI 工具)、Why 写的(为人还是为搜索引擎)。

"Evaluating your content in this way, whether you're using AI-generated content or not, will help you stay on course."

这意味着 Who/How/Why 是通用质量标尺,不是"AI 特别审计"。

1.5 早期"AI 立场前史"——auto-generated 链接 / 内容

来源:2007/08 Quality websites: avoid auto-generated links(Jianfei Zhu)+ 2017/01 Protect your site from user-generated spam(Anouar Bendahou)

2007 那篇是 Google 第一次系统讲 "auto-generated" 的反对态度——当时还是 spamware 自动生成 guestbook 链接这种"前 LLM 自动化"。原话:

"Google will not give credit for auto-generated links. When we detect that a site has unnaturally generated their own backlinks, the site's ranking can be negatively affected and can be removed from our index."

2017 那篇延伸到 forum / comment 的自动化垃圾:CAPTCHA / 限频 / nofollow / Akismet 等 8 条防御手段。

主题脉络的连续性:

2007 自动生成链接 → 2017 用户区自动化垃圾 → 2022 helpful content update → 2023 generative AI scaled abuse

Google 反复强调:"automation"本身从来不是问题,"用 automation 操纵排名"才是。LLM 时代的政策只是这条 15+ 年老线的延伸,不是新规则。


Part 2 — 机器学习信号引入史:RankBrain → BERT → MUM → 后续

2.1 本批语料里能锚定的 ML 印记

注意:本批 18 篇候选里 没有 RankBrain (2015-10) / BERT (2019-10) / MUM (2021-05) / Helpful Content Update (2022-08) 的官方发布博文。能找到的 ML 痕迹只有以下两处二手引用:

  1. 2021/08 Update to how we generate page titles(Danny Sullivan)——Google 用"超出 HTML title tag"的系统(实际上是 BERT 之后的语言理解模型)重写标题。原话:

    "We are making use of text that humans can visually see when they arrive at a web page. We consider the main visual title or headline shown on a page, content that site owners often place within <H1> tags ... Other text contained in the page might be considered, as might be text within links that point at pages."

    关键数据:HTML <title> 标签仍然占用 80%+ 的最终展示标题——即使 ML 接管了重写决策,显式语义仍是主信号。

  2. 2023/02 google-search-and-ai-content 提到 SpamBrain——Google 的反垃圾 ML 系统,从 2018 部署、2022 公开 webspam report。Google 立场:

    "Our spam-fighting efforts—including our SpamBrain system—will continue, however spam is produced."

2.2 ML 引入时间线骨架(待下一轮 ingest 补全官方原文)

年 系统 性质 本批是否有源?
2009 Caffeine 索引基建预览 索引架构,非 ranking ML ✅(2009/08 sandbox 预览)
2012 Knowledge Graph 首发 实体图谱 ❌(本批只有 2015 events 衍生帖)
2015 RankBrain 第一个深度学习 ranking 信号 ❌
2018 SpamBrain ML 反垃圾 仅 2023 引用
2019 BERT 双向 transformer 理解查询 ❌
2021 MUM 多任务统一模型,多模态/多语言 ❌
2021 标题重写 ML 系统 基于页面可见文本的标题生成 ✅(2021/08)
2022 Helpful Content System 站点级 helpfulness 信号 仅 2023 引用
2024 AI Overviews 生成式答案 仅 2025 引用
2025 AI Mode 复杂多轮生成式搜索 ✅(2025/05 引用)

TODO 给下一轮 ingest:

  • 抓 GSC blog 里 rankbrain / bert / mum / helpful-content-update / ai-overviews 关键词的官方原文
  • 抓 2015/10 RankBrain 官方公告(Bloomberg 首发,Google 后期补 blog)
  • 抓 2019/10 "Understanding searches better than ever before"(BERT 发布)
  • 抓 2021 Google I/O 围绕 MUM 的 search blog 转述

Part 3 — Knowledge Graph 到 AI Overviews 的演化

3.1 本批仅有的 Knowledge Graph 锚点

来源:2015/01 Upcoming Events In The Knowledge Graph(Justin Boyan)

这篇本身不解释 Knowledge Graph 是什么——它假设读者已经知道。它讲的是把 schema.org Event / MusicGroup / ComedyEvent markup 写到官网,以实体卡片形式出现在搜索结果里。

关键机制:

  • 直接标记:官网内嵌 schema.org markup
  • 委托标记(delegation markup):"我的事件不在官网"时,用 MusicGroup JSON-LD 指向第三方(Bandsintown / Songkick / Ticketmaster)
  • 场地实体:Event markup 让 venue 进入 Knowledge Graph
  • 基础设施借力:Ticketmaster 等大站已经标好,接他们家票务的 venue 不用做任何事就被覆盖

意义:Knowledge Graph 在 2012 上线后,逐步从"百科实体"扩展到"事件 / 商品 / 组织 / 商家"——这条路径直接喂给后来的 AI Overviews:AI Overviews 在生成答案时引用的"事实块",大量来自 Knowledge Graph + structured data 抓取出的实体。

3.2 演化逻辑(基于 2025/05 succeeding-in-ai-search 反推)

2025/05 那篇明确把 structured data 放进 "AI 搜索 success" 7 条之一:

"Structured data is useful for sharing information about your content in a machine-readable way that our systems consider and makes pages eligible for certain search features and rich results. If you're using structured data, be sure to follow our guidelines, such as making sure that all the content in your markup is also visible on your web page."

这等于把 2012 起的 Knowledge Graph / structured data 路径,正式接入 AI Overviews 时代——markup 不变,消费端从"知识卡片"扩展到"AI 生成答案的事实锚"。

3.3 AI Overviews 的官方表述(2025/05)

"With AI Overviews and AI Mode, people are using Search more often, asking new and more complex questions, and are more satisfied with their results. AI Overviews display links in a range of ways, and show a wider range of sources on the results page so it's easy for people to click out and explore content on the web."

Google 给的"AI Overviews 不抢流量"的辩护:

"When people click to a website from search results pages with AI Overviews, these clicks are higher quality, where users are more likely to spend more time on the site. ... Our AI results may give people more context about a topic overall, and display more relevant supporting links."

(注:这是 Google 自家的说法,SEO 社区有大量反证数据,但本 wiki 只忠实记录 Google 原话。)


Part 4 — Succeeding in AI Search 指南(2025+)

来源:2025/05 Top ways to ensure your content performs well in Google's AI experiences on Search(John Mueller)

4.1 7 条建议清单(完整提炼)

# 建议 关键执行点
1 Focus on unique, valuable content for people 非商品化内容;承接"更长、更具体、追问式"的 AI Mode 查询;沿用 creating helpful, reliable, people-first content 标尺
2 Provide a great page experience 跨设备、低延迟、主内容易辨识
3 Ensure we can access your content 满足 technical requirements for Google Search;Googlebot 不被屏蔽;200 状态;内容可索引——"Meeting the technical requirements covers you for search generally, including AI formats."
4 Manage visibility with preview controls nosnippet / data-nosnippet / max-snippet / noindex 控制在 AI 格式里的露出;越严格 → AI 展示越少(站点用 noindex 等于退出 AI Overviews 引用)
5 Make sure structured data matches the visible content markup 内容必须页面可见;遵循 sd-policies;通过 Rich Results Test
6 Go beyond text for multimodal success 高质量 image / video;Merchant Center / Business Profile 保持新鲜——对接 AI Mode multimodal search
7 Understand the full value of your visits 别只看点击;看 session 时长、转化、注册、咨询——Google 自称 AI Overviews 来的流量"更高质量"

4.2 不变的底层 ethos

"The underpinnings of what Google has long advised carries across to these new experiences. Focus on your visitors and provide them with unique, satisfying content."

Google 没在 2025 给出"新 AI SEO 公式";它给出的是 "老方法在新格式里继续有效" 这个反诉求——劝创作者别 chase tactic,继续做 helpful content。

4.3 "Search 一直在演化"的修辞

"The only thing predictable in Search is that it always evolves because people's needs are always evolving. The classic 'ten blue links' format changed to handle the needs of those seeking visual, video, news, and other types of content. Desktop displays evolved to mobile-friendly ones. Search evolved to handle voice queries, or 'multimodal' queries..."

把 AI Overviews 框定为 "ten blue links → mobile → voice → multimodal → AI" 这条自然演化曲线的最新一节,而不是"Google 被 OpenAI 逼着革自己的命"——这是定性,不是定量。


Part 5 — Search Console / Crawler 的 AI 化(AI Powered Configuration)

来源:2025/12 Streamline your Search Console analysis with the new AI-powered configuration(Nati Yosephian)

5.1 功能定位

Search Console Performance Report 加了 AI-powered configuration(实验性,灰度推出):用自然语言描述要看的数据,AI 帮你把 filter / comparison / metric 一键配置。

"Powered by AI, this feature lets you describe the analysis you want to see in natural language. Your inputs are then transformed into the appropriate filters and settings, instantly configuring the report for you."

5.2 AI 能做的三件事

  1. Applying filters — query / page / country / device / search appearance / date range
  2. Configuring comparisons — 复杂自定义日期对比
  3. Selecting metrics — Clicks / Impressions / Average CTR / Average Position 4 选

5.3 示例自然语言查询

  • "Show me queries on phone searches that contain the word 'sports' in the last 6 months."
  • "Compare traffic for my pages that contain '/blog' in this quarter to the same quarter last year"
  • "Show me the Average CTR and Average Position of my queries in Spain in the last 28 days."
  • "Show clicks for pages that include the word 'google'."

5.4 已声明的边界

维度 限制
Scope 仅 Performance report for Search results;Discover / News 不支持
Accuracy "AI can sometimes misinterpret requests. Always review the suggested filters."
不会做的事 不会排序表格、不会导出数据——只配置不执行

5.5 模式判断

这是 Google 内部工具的 AI 化样本,不是 ranking 信号变化——它影响 SEO 从业者的工作流(数据查询效率),不影响搜索结果本身。本 wiki 记录它是为了完整捕捉 2025 这条"AI 渗透到 GSC 每一层"的产品线。

5.6 Crawler 维度的 AI 化(暂缺源)

本批语料没有覆盖 Google-Extended / GoogleOther 等"AI 训练 / Gemini 抓取"专用 crawler 的官方公告(这些通常在 crawlers 总览 页和单独的 blog 帖里)。下一轮 ingest 补 googleother / google-extended / vertex 关键词。


Part 6 — 多模态搜索(Image / Lens / Multisearch)

本批语料没有单独的 Image / Lens / Multisearch 发布博文,但 2025/05 succeeding-in-ai-search 把多模态归到 AI Mode:

"Through the power of our AI, people can perform multimodal searches where they snap a photo or upload an image, ask a question about it and get a rich, comprehensive response with links to dive deeper."

对应站长建议:

  • 配高质量 Google Images (本地参考资料) 内容
  • 配高质量 Video (本地参考资料) 内容
  • Merchant Center 商品信息保持新鲜
  • Business Profile 信息保持新鲜

判读:Google 把 multimodal 从单独产品(Google Lens / Multisearch)合并进 AI Mode 这个统一入口——不是"再发一个新功能",而是"AI Mode = 文本 + 图像 + 视频 + 语音的统一前台"。

下一轮 ingest 应该抓的源:

  • Google Lens 官方 blog 历史
  • Multisearch (2022/04) 发布原文
  • "AI Mode multimodal search"(blog.google 上 2025/04 的发布)

Part 7 — Google 的核心信念:质量先于"是否 AI 写的"

把 2007 / 2017 / 2021 / 2023 / 2025 五个时间点的官方原话叠在一起,Google 的立场一脉相承:

年份 来源 一句话立场
2007 quality-websites-avoid-auto-generated "Google will not give credit for auto-generated links."(反操纵性自动化,不反自动化本身)
2017 protect-your-site-from-user-generated 给站长 8 条防 spam 工具(技术防御 ≠ 价值判断)
2021 update-to-generating-page-titles "HTML title 标签仍占 80%+ 最终展示"——ML 接管的同时,显式信号仍是主信号
2023/02 google-search-and-ai-content "It's just content. Using AI doesn't give content any special gains."
2023/12 definitely-not-ai-generated-post Gary Illyes 在年终总结里自嘲一句 "all this text (that we definitely didn't generate with AI)"——Google 内部对"AI 写作"持轻松、不焦虑态度
2025/05 succeeding-in-ai-search "The underpinnings of what Google has long advised carries across to these new experiences."

7.1 三句"原话压缩"

  1. "Our focus on the quality of content, rather than how content is produced, is a useful guide." — 2023/02
  2. "It's just content." — 2023/02 FAQ
  3. "The underpinnings of what Google has long advised carries across to these new experiences." — 2025/05

7.2 立场背后的策略

Google 选择不分类裁判 AI 内容是有产品考量的:

  • 可执行性:无法可靠判定一段文字"是不是 AI 写的"——任何 detector 都会有假阳/假阴
  • 激励对齐:与其惩罚工具,不如奖励结果(helpful content),让 AI 工具反过来服务质量
  • 历史一致:2011 Panda 没禁人写垃圾内容,2023 不禁 AI 写内容,逻辑同构

7.3 对站长的可操作 takeaway

场景 行动
用 AI 帮写草稿,人编辑润色 OK,不需要披露(除非读者会问"How was this created")
用 AI 大规模生成关键词页 scaled content abuse,会被打
AI 写但加了真实经验 / 案例 / 数据 OK,满足 E-E-A-T 里 "Experience" 维度
把 AI 列为作者 不推荐——按"如何创建"披露,不按"谁写的"
给所有 AI 内容打 disclosure 不必;仅在"读者会问 How"时披露
退出 AI Overviews 引用 用 noindex 或 nosnippet/max-snippet,但同时退出搜索可见性

附录 A:本批 18 篇候选的实际归类

文件 AI/ML 相关? 归入 Part
2005/09 announcing-sitemap-generator-version ❌ sitemap 编码 —
2005/09 using-oai-pmh-with-google-sitemaps ❌ OAI-PMH 协议 —
2005/12 new-version-of-sitemap-generator ❌ GZip bugfix —
2007/08 quality-websites-avoid-auto-generated ✅ 早期 auto-generated 立场 Part 1.5 / Part 7
2008/04 retiring-support-for-oai-pmh-in ❌ OAI-PMH 下线 —
2009/01 new-google-sitemap-generator-for-your ❌ sitemap 工具 —
2009/08 help-test-some-next-generation ⚠️ Caffeine 索引基建预览(非 AI/ML ranking) 仅时间线骨架 Part 2.2
2012/03 crawl-errors-next-generation ❌ Webmaster Tools 报错优化 —
2015/01 upcoming-events-in-knowledge-graph ✅ Knowledge Graph 实体扩展路径 Part 3.1
2017/01 protect-your-site-from-user-generated ✅ 自动化垃圾防御(前 LLM) Part 1.5 / Part 7
2017/12 nhac-nho-ve-lien-ket-tai-tro ❌ 越南语公告 stub —
2018/07 ein-meilenstein-fur-die-sicherheit-in ❌ 德语公告 stub —
2021/08 update-to-generating-page-titles ✅ ML 重写标题 + HTML title 80% 锚 Part 2.1 / Part 7
2023/02 google-search-and-ai-content ✅✅✅ 核心立场原文 Part 1 全段 / Part 7
2023/12 definitely-not-ai-generated-post ✅ 年终总结里 AI 自嘲一句 Part 7
2025/05 succeeding-in-ai-search ✅✅✅ AI 时代 SEO 7 条 Part 4 全段 / Part 3.3 / Part 6
2025/12 ai-powered-configuration ✅ Search Console AI 助手 Part 5
2026/04 scl-shanghai-2026 ❌ 活动公告 —

命中率:18 篇里 8 篇命中(44%),其中 3 篇(2023/02 / 2025/05 / 2025/12)是核心原文。


附录 B:下一轮 ingest TODO(补齐 RankBrain → AI Overviews 的官方原文)

  1. developers.google.com/search/blog 全量抓取,过滤关键词:
    • rankbrain / bert / mum / helpful-content / ai-overviews / ai-mode / google-extended / googleother
    • knowledge-graph(2012 首发)
    • multisearch / lens
  2. 抓 webspam-report 系列(SpamBrain 公开成绩单)
  3. 抓 google-search-and-ai-content 之前 / 之后的姐妹帖(往前找 helpful content system 2022/08 / 2022/12 发布原文;往后找 2024/05 AI Overviews 发布原文)
  4. 与本 wiki 合并升级——不要新建第二个 AI 主题 wiki,直接在本文件追加 Part 8 / Part 9

来源与关联资料