How AI engines choose what to cite (and how to be chosen)
AI engines choose citations in stages: they retrieve candidate pages, parse what those pages actually say, select the passages that answer the question most directly, and prefer sources they can verify. You can influence every stage. Here's what happens at each one, and what you control.
Stage 1: Retrieval. Can the engine find you?
When ChatGPT, Perplexity or Google's AI Overviews answer with current information, they start by searching. The engine turns the user's question into queries, pulls a set of candidate pages, and works only within that set. If you're not retrieved, nothing downstream can save you.
What you control: allow the AI crawlers (GPTBot and OAI-SearchBot, PerplexityBot, ClaudeBot, Google-Extended) in robots.txt, keep a clean sitemap on one canonical host, publish an llms.txt file, and cover the questions your customers ask so your pages match the queries engines generate.
Stage 2: Parsing. Can the engine read you?
Retrieved pages get parsed, and parsing is less forgiving than a browser. Many AI crawlers read raw HTML without executing JavaScript. Content that only exists after scripts run, text trapped in images, and pages with no semantic structure all parse badly or not at all.
What you control: ship real content in the initial HTML, use semantic headings and landmarks, add alt text, and keep your important claims in text. We scored a failing 40 on our own audit for exactly this reason; the fix is documented in our case study.
Stage 3: Selection. Are you the easiest correct answer?
From the parsed candidates, the engine assembles an answer, and it favors passages it can use with minimal rework: a direct answer under a question-shaped heading, a table that compares options, a list of steps, a stated price. This is the heart of Answer Engine Optimization. The citation goes to whoever makes the answer easiest to assemble, not to whoever wrote the most words.
What you control: open every section with a one-to-three sentence direct answer, use lists and tables for anything comparative, and keep facts specific and current. Test any page's answer-readiness with the free AEO checker.
Stage 4: Trust. Can the engine defend citing you?
Between two passages that both answer the question, engines prefer the source they can verify: a named organization with structured data, a named author, visible dates, a consistent identity across the web, and third-party corroboration. These are the E-E-A-T signals, and they act as the tiebreaker in every close call. Google's own documentation on optimizing for AI features points the same direction: clear provenance, structured data, and content written for people.
What you control: Organization and FAQPage schema, author attribution with Person schema, machine-readable published and updated dates, and complete public profiles. Check yours with the free E-E-A-T checker.
The four stages at a glance
| Stage | The engine asks | You control |
|---|---|---|
| Retrieval | Can I find this page for this question? | Crawler access, sitemap, llms.txt, question coverage |
| Parsing | Can I read what it says? | Content in HTML, semantic structure, text over images |
| Selection | Does it answer directly? | Answer-first paragraphs, lists, tables, specific facts |
| Trust | Can I defend this source? | Schema, authorship, dates, corroboration |
What this means for your site
Work the stages in order. A trust signal can't rescue a page the engine never retrieved, and a retrieved page still loses if a competitor's passage is easier to quote. The sites that win AI citations aren't gaming anything: they're the ones that made every stage effortless. That's also why this work compounds; each fix helps every future question in your category.
See where your site fails the four stages
The free iCited audit runs 29 deterministic checks across retrieval, parsing, selection and trust, and hands you a prioritized fix list in about two minutes.
Run your free audit →Frequently asked questions
How does ChatGPT choose its sources?
When answering with web data, it runs searches, retrieves candidate pages, parses their content, and assembles an answer from the passages that most directly and credibly answer the question. Pages that are easy to retrieve, parse and quote win the citation.
Do AI engines use Google rankings?
Each engine has its own retrieval layer, and several draw on established search indexes. Ranking well in traditional search helps you enter the candidate pool, but the final citation depends on how quotable and verifiable your page is once retrieved.
Can I pay to be cited by AI engines?
No. Citations in AI answers are organic: engines select the sources that best answer the question. That's why citability work compounds while ad spend just meters.