How Ai engines choose which sources to cite

The pipeline behind a cited answer

1. Query understanding. The engine rewrites the question into searches, often several (Google calls this query fan-out).
2. Retrieval. It pulls candidate pages from a search index.
3. Selection. It reads the candidates and keeps passages that directly answer a part of the question, preferring sources that look reliable and current.
4. Synthesis. It writes the answer, combining what several sources agree on.
5. Citation. It links to some of the sources it used. Not every source used is shown, and not every page shown is equally weighted.

Which index each engine searches

Crawler access: search vs training

Ai companies typically run separate crawlers for training models, for building search indexes and for fetching pages when a user asks. They can be allowed or blocked independently in robots.txt. For example, ChatGPT’s documentation separates GPTBot (training) from OAI-SearchBot (search), and says sites that block OAI-SearchBot won’t appear in ChatGPT search answers. Google’s Google-Extended token governs some Gemini uses, but Ai Overviews in Search follow Googlebot.

The practical rule: whatever you decide about training, keep the search crawlers of the engines your buyers use allowed, at the robots.txt, firewall and CDN level.

What makes a source get selected

• Relevance to the sub-question. The passage answers one specific part of the question, using the same terms the searcher used.
• Extractability. A clear heading followed by a direct answer, a list or a table is easy to lift. Answers buried in long paragraphs, images or PDFs are not.
• Specificity and evidence. Numbers, named sources, dates and first-hand detail beat general statements.
• Corroboration. Claims that other trusted sites also make are safer to repeat. This is why brand mentions, reviews and listings matter so much.
• Authority. Sites with strong backlink profiles, real reviews and a history of being cited are more likely to be chosen again.
• Freshness. For time-sensitive questions, recently updated pages have an edge.

The types of sources engines lean on

Look at the citations in answers for your own buyer questions and you will usually see a mix of: review platforms and directories, “best of” and comparison articles, forums and community threads, reference sites, news and trade publications, and brand websites. Each industry has its own favorites. Finding out which ones the engines cite for your questions tells you where to earn reviews, listings and mentions.

A worked example

Imagine a buyer asks: “Which accountant in Austin is best for a small restaurant?” A generative engine might fan that out into searches like “restaurant accountant Austin”, “best small business CPA Austin reviews” and “what to look for in a restaurant accountant”. It then reads the top results for each.

The accountant who gets named is usually the one who appears in several of those result sets at once: a service page that says “accounting for Austin restaurants” in plain words, a strong profile on the review sites that rank for the second search, and a mention in a local business publication or association directory. A firm with a better website but no reviews or mentions often isn’t in enough of the candidate pages to be chosen.

How to become a cited source

1. Open the door. Allow the search crawlers you care about and make sure key content renders as plain text.
2. Map the questions. List the questions buyers ask at each stage, then the sub-questions an engine would fan out to.
3. Write extractable answers. One question per heading, the answer in the first two or three sentences, details after.
4. Add evidence. Prices, timelines, data, named sources and real examples.
5. Earn corroboration. Reviews on the platforms engines cite, brand mentions in lists and publications, and backlinks from relevant sites.
6. Measure and iterate. Track which sources get cited for your questions and close the gaps.