Module 1 · Lesson 1.2 · How an AI engine chooses its sources · 3 min

How ChatGPT and other AIs choose the sources they cite

Lesson 1.2 of the free AEO and GEO course: how ChatGPT, Perplexity, Gemini, Copilot, Claude and Google AI Overviews choose their sources. The model's memory or a web search, the indexes and the bots, how much ranking matters, and what makes a page worth citing.

The video for this lesson is coming soon to YouTube. Meanwhile, the full transcript is right here.

Transcript

0:00Introduction

Lesson 1.2: How an AI engine chooses its sources.

0:06Memory or search

When you ask an AI assistant a question, the answer can take one of two routes. From the model's memory: what it learned during training, up to a certain date. In this case, there are usually no links at all. Or from a web search done right then: the engine searches, reads a few pages, and writes the answer citing them. The two routes call for different work.

0:28Where the engines search

To search, every engine uses an index, and a bot that reads the pages. Google's AI Overviews and AI Mode use Google's own index. Copilot uses Bing's index. ChatGPT has its own search bot, OAI-SearchBot, and also relies on outside search engines. Perplexity and Claude have their own: PerplexityBot and Claude-SearchBot. If you block these bots, those engines can't read you.

0:57Does ranking well matter?

Does ranking well matter? Yes, it does. In a study published by Seer Interactive in February 2025, 87% of ChatGPT search citations matched Bing's top results. But on its own, it matters less and less. According to Ahrefs, in July 2025, 76% of the pages cited in Google's AI Overviews ranked in the top 10. In March 2026, when the study was repeated, that share had dropped to 38%.

1:30How a source gets chosen

No engine publishes the exact rules it uses to pick its sources. But if you look at the answers, three things keep coming back. The page can be found: bots can access it, and it's indexed. It answers: clearly, the exact question that was asked. And it's credible: it has sources, data, and other sites that confirm what it says.

1:53The model's memory

And memory? You can't edit it. You build it over time. The model learns from what the web says about you, up to the date of its training. That's why consistent mentions matter, on different sites, repeated over time: your name, your business and your location, always stated the same way. In this course we work on both routes: findable pages for search, and a consistent presence for memory.

2:19Your task for today

Your task for today: take one of your three questions and ask the same assistant twice. Once explicitly asking it to search the web, and once without. Then compare: do the names change? Are there sources? Which ones?

2:45Next lesson

In the next lesson: the engines one by one. How they search, whether they show their sources, and which bots they use to read pages. See you there.

Your task

Take one of your three questions and ask the same assistant twice. Once explicitly asking it to search the web, and once without. Then compare: do the names change? Are there sources? Which ones?

Track it in the course worksheet.

Sources

Checked 17 October 2026.