# How AI decides what to cite

> When someone asks ChatGPT, Claude, Perplexity or Google's AI a question, a handful of sources get named and the rest are invisible. Here is the actual pipeline behind that choice — and the properties of a page that survive it.

Published 2026-09-19 · InfoRelay GEO Lab · Explainer

Ask an assistant "what's the best tool for transcribing interviews?" and you'll get an answer with two or three named products and a short list of sources. Those named sources won the query. Everyone else was not in the room. Understanding how that shortlist gets built is the whole game.

## The answer is not a search result

A classic search engine returns a ranked list of links and lets you choose. An answer engine writes a single response and decides, mid-composition, which sources to lean on and name. If your domain isn't among the handful the model grounds its answer in, you get zero visibility — not a lower ranking, and no signal that it happened.

## The pipeline, step by step

Whether it's ChatGPT with browsing, Perplexity, Google AI Overviews, or Claude with search, the shape is the same four stages:

1. **Retrieval — get a candidate set.** The engine turns the question into search queries and pulls candidate pages from a search index. If your page isn't in that index, or is too slow or blocked to fetch, the pipeline never sees it. Retrieval is a hard gate.
2. **Ranking — pick the few worth reading.** From dozens of candidates the engine keeps three to six to actually read, using source reputation, topical relevance, freshness, and how directly the page answers the exact question.
3. **Grounding — write the answer from those pages.** The model composes an answer grounded in the shortlist. Self-contained, factual sentences are easy to lift; claims buried in a sales narrative or a video are not.
4. **Attribution — decide whom to name.** The engine credits the sources that most directly supported specific claims, and prefers sources that corroborate each other. A fact confirmed across several reputable pages is safe to state and attribute; a lone claim is riskier to repeat.

**The short version:** to be cited you must be *retrievable* (in the index, fast, crawlable), *relevant* (you answer this exact question), *quotable* (a clean, self-contained statement), and *corroborated* (other trusted sources agree). Miss any one and you fall out of the shortlist.

## What actually makes a page citable

- **It's indexed and fast to fetch** — verified in the index, not just submitted; no login walls or JS-only content.
- **It answers one real question directly** — the answer is in the first paragraph, then supported.
- **Its claims are self-contained** — sentences a model can lift verbatim and still be true; concrete numbers and dates beat adjectives.
- **It's structured cleanly** — real headings, lists, tables, and a clean machine-readable rendering (this is why InfoRelay serves markdown to agents).
- **It's dated and maintained** — visible, honest freshness.
- **It agrees with the record** — if reputable sources contradict your page, the model trusts the corroborated version and may describe you incorrectly.

## What does not reliably work

- **Keyword stuffing** — engines rank on relevance and clarity, not density; stuffing makes a page less quotable.
- **Thin, AI-spun content at volume** — a hundred weak, uncorroborated candidates do not manufacture trust.
- **Schema markup alone** — it helps a machine parse a page; it does not make claims true, relevant, or corroborated.
- **Buying "mentions"** — low-trust pages add noise the model discounts and can distort your described facts.

## How to find out where you stand

The number that matters is not traffic — it's **Share of Citation**: across the buyer questions in your category, how often do the engines name your domain as a source? See the companion field report, *We measured our own AI citation share for 90 days* (https://inforelay.ai/lab/we-measured-our-own-ai-citations/). Measure where you stand with a free check at https://citedaudit.ai/check/.

None of this is a trick. The engines are trying to give a good answer from trustworthy, quotable sources. Getting cited by AI is mostly the discipline of being that source, on the exact questions your buyers ask, and proving it moved with a repeatable measurement.

---
Source: https://inforelay.ai/lab/how-ai-decides-what-to-cite/
Index: https://inforelay.ai/llms.txt