记录的问题
“How does Google AI Mode choose the sources it cites?”
- 产品
- ChatGPT
- 界面
- ChatGPT 网页版,启用网络搜索
- 模型
- ChatGPT 自动选择模型(未公开模型版本)
- 语言环境
- en-US(英文回答;中文界面)
- 观察到的回答
- 打开会话 ↗
- 观察日期
- 2026年11月15日 21:30 CST
- 全新对话
- 是
回答原文
Google AI Mode chooses cited sources through a combination of Google Search ranking, query expansion, relevance assessment, and AI-generated synthesis. The important distinction is that it doesn't simply find one webpage, read it, and cite it. It can search across multiple related queries, gather information from different websites, and use those pages to construct a single answer. Here's how the process works. 1. Query fan-out: it searches beyond your exact question This is one of the most important mechanisms. Google's AI Mode uses a technique called query fan-out. For example, if you ask: “What are the best ways for small businesses to improve customer retention?” It might break that into related searches covering: Customer retention strategies. Customer loyalty programmes. Customer experience research. Customer churn statistics. Examples of successful retention campaigns. These searches happen concurrently, allowing AI Mode to find relevant information that a single search might miss. Google explicitly documents this process. Google Search Help +1 2. Google Search ranking: it finds candidate sources AI Mode builds on Google's existing Search infrastructure. Its source discovery process draws on: Google's search index. Existing relevance and quality-ranking systems. Information freshness. Relevant information from Google's Knowledge Graph and other data sources. Google describes this as retrieval-augmented generation (RAG), also known as grounding. In simple terms: Google retrieves potentially relevant webpages. The AI examines information from those pages. It uses relevant information to construct an answer. It provides links to supporting webpages. Google for Developers +1 3. Relevance to individual statements Not every source is necessarily used to support the entire answer. For example, an answer about electric vehicles might contain: Answer section Relevant source type Battery technology Technical research Vehicle prices Manufacturer or pricing pages Environmental impact Scientific studies Government incentives Official government websites The aim is to connect relevant sources to the information they help support. Google doesn't publicly disclose the precise algorithm or weighting used to decide which retrieved pages receive citations. 4. Quality and usefulness matter Google says its AI Search experiences rely on high-quality web content. Factors worth understanding include: Relevance: Does the page answer the question? Information quality: Is the information useful and reliable? Originality: Does it offer firsthand information or unique insights? Freshness: Is the information current where that matters? Specificity: Does it address the particular subtopic? These are useful ways to understand Google's approach, rather than a published formula for citation selection. Google for Developers +2 5. Preferred Sources can influence what users see Google introduced Preferred Sources into AI Mode in 2026. Users can select websites they particularly value, and Google can highlight those websites when they appear in AI responses. This is worth distinguishing from ordinary source selection: a preferred source isn't automatically guaranteed a citation. blog.google +1 The complete process 1. User question AI interprets the intent and context 2. Query fan-out Generates multiple related searches 3. Search and retrieval Finds relevant webpages and other information 4. AI synthesis and grounding Combines retrieved information into an answer 5. Answer and citations Displays the response with links to relevant sources What does this mean for website owners? The key takeaway is that ranking in traditional Google Search and being cited in AI Mode are related, but not identical. A website can rank well organically without necessarily being cited in a particular AI answer. Equally, a page that provides particularly useful information for one of the subtopics generated by query fan-out may be relevant to an AI response. Google's official guidance is that traditional SEO fundamentals remain important for AI Search. Google for Developers One important limitation: Google has not published a complete, deterministic explanation of how it selects one specific URL over another for an individual AI Mode citation. For further reading: Google's official AI Search optimisation guide.
记录了 6 个来源
这次观察不能证明什么
本次使用已登录的免费账户在全新 ChatGPT 对话中观察。产品没有公开完整检索配置;来源面板显示 5 个来源,并记录了 5 个唯一来源。结果可能因会话、语言环境、账户状态和日期而变化。