Large language models (LLMs) select sources in a retrieval stage, before they make an answer. This page explains where the sources come from, the five factors that control source selection, and why source selection is important for brands. An answer engine such as ChatGPT, Claude, Gemini, or Perplexity usually uses information from many sources on the web. These sources give the factual material for the answer. Thus, they explain why some websites, brands, or claims occur in AI-generated answers.

​
Retrieval before generation: how answer engines get sources

Most modern AI search systems use this pattern: the answer engine gets sources and uses them to build its answer. In this process:
  1. A user asks a question.
  2. The system gets the relevant documents or passages.
  3. The answer engine analyzes the information that it got.
  4. The model makes an AI-generated answer.
The model uses the information that it got as context. Thus, the documents that the system selects in the retrieval stage have a strong effect on the final answer.

​
Where sources come from

Usually, AI systems get information from a combination of sources:
  • Websites that search engines index
  • Editorial articles
  • Product reviews
  • Comparison pages
  • Documentation sites
  • Forums and communities
  • Training data and web sources
Some systems use external search engines. Other systems use their own internal indexes or knowledge sources.

​
Five factors that control source selection

Different factors have an effect on which sources the system gets when it answers a question.

​
Relevance to the query

The most important factor is the relevance of the content to the question, or to the related queries that the system makes. If a page clearly answers a specific question, the system has a higher probability to get it.

​
Authority and trust

The system can give priority to sources that many people know as authoritative. These sources can include:
  • Well-known publications
  • Established websites
  • Industry experts
  • Resources that many sources cite
Authority signals can come from links, citations, reputation, and reliability over time.

​
Content clarity

AI systems frequently find it easier to understand and use content that clearly explains a topic or gives structured information. For example, AI systems can more easily use pages that include these items:
  • Clear headings
  • Lists
  • Comparisons
  • Short explanations

​
Coverage of the topic

The system has a higher probability to get sources that fully cover a topic. A page can answer many related questions or give deep explanations. The system can select this page more frequently than a page that mentions the topic only briefly.

​
Consistency across sources

Many sources can repeat similar claims or mention the same brands. The AI system can then use those signals as stronger evidence when it makes its answer.

​
How sources change AI answers

After the system gets the relevant documents, the model uses them as context to make its answer. In this stage, the model can do these tasks:
  • Summarize information from many sources
  • Compare products or services
  • Merge explanations from different documents
  • Extract key facts or recommendations
The answer is AI-generated. Thus, the final answer can be different from all the individual pages. It shows a combination of the information that the system collected in the retrieval stage.

​
Citations in AI answers

Some AI systems include citations or links. These show which sources had an effect on the answer. These citations can point to:
  • Articles
  • Documentation pages
  • Product pages
  • Review sites
Citations help users verify where the information came from. They also show clearly which sources the system used to make the answer. For more, read How AI citations work.

​
Why source selection is important for brands

AI systems build answers from the sources that they get. Thus, brands that occur frequently in those sources have a higher probability to occur in AI-generated answers. For this reason, visibility in AI answers depends on these items:
  • Mentions by relevant sources
  • Content that answers usual questions
  • Presence in comparisons and reviews
  • A connection to the topic across the web
If a brand almost never occurs in the sources for a topic, it can have a small visibility in AI-generated answers. To see the sources that answer engines cite for your brand, read Citations.

​
In the API

The public API lists the domains that answer engines cite: Cited domains.