How to get cited by ChatGPT

Getting cited by ChatGPT comes down to two separate battles that most advice collapses into one. There is what the model already believes about your category from training, and there is what it retrieves and quotes at the moment someone asks. They have different mechanics, different timelines, and different tactics. Confuse them and you will spend months optimizing the wrong thing.

First, understand which ChatGPT you are optimizing for

ChatGPT answers in two modes. In its default mode it draws on parametric memory, everything absorbed during training. That memory is broad, frozen at a cutoff date, and impossible to edit. If it thinks your competitor is the category leader, no amount of publishing this week changes that answer directly.

In search mode it retrieves live pages and grounds the answer in them, citing sources. This is the mode you can influence quickly, and it is where practical citation work happens.

A detail worth knowing: ChatGPT's search runs on a Bing-based index, not Google's. Google's AI Overviews and Gemini use Google's. This is a concrete reason a brand shows up strongly in one engine and vanishes in another, and why judging your AI visibility from one engine gives you a quarter of the picture.

The citation funnel is brutal

Before your page can be quoted, it has to survive a sequence of filters. The engine expands the user's question into multiple back-end searches. Each search returns candidates. A reranking model scores those candidates and discards the large majority. Only a handful survive into the context the model actually reads, and ChatGPT cites notably few of them, single digits per answer in most studies, against twenty or more for Perplexity.

Fewer citation slots means the bar per slot is higher. Being retrievable is not enough. You have to be the most quotable thing in the pile.

What the research says actually works

This is where most advice becomes guesswork, so it is worth anchoring to the one serious study on the question. Princeton-led research on generative engine optimization tested which content changes increase citation rates inside AI answers. The results were specific:

• Adding authoritative quotations: roughly +41 percent visibility.
• Adding statistics: roughly +32 percent.
• Adding citations to credible sources: roughly +30 percent.
• Improving fluency: roughly +24 to 28 percent.
• Keyword stuffing: essentially nothing.

Notably, the gains were largest for sources that started lower in the ranking. Generative engines are, in this narrow sense, more meritocratic than classic search: a well-constructed page from a smaller site can win a citation slot that its domain authority would never win in blue links.

The practical checklist

1. Answer in the first two sentences. Models lift opening lines. Bury the answer under three paragraphs of preamble and you forfeit the slot.
2. Use question-shaped headings. Match how a person actually phrases the question, then answer immediately underneath.
3. Make every claim self-contained. A sentence that needs the previous paragraph to make sense cannot be quoted in isolation.
4. Lead with specifics. Numbers, dates, prices, named sources. Vague superlatives are unquotable.
5. Add FAQ and product structured data. It will not rescue a weak page, but it helps a strong one get parsed and reused.
6. Show up where the engines already look. Reddit, YouTube and established industry press are cited far more than most brands expect. A strong Reddit thread can outrank your own site as a source.
7. Keep facts current. Stale prices on your own pages and in old third-party listicles get quoted back at buyers as truth.
8. Ignore llms.txt. The proposed file has little evidence behind it and Google has said it does not use it. It is the current snake oil of the category.

How to tell if it is working

You cannot see this in analytics. Most AI answers end without a click, so there is no referrer and no impression to count. The only way to know is to ask the engines the questions your buyers ask and record what comes back, repeatedly, because a single check is a coin flip rather than a measurement. We wrote about why single-shot AI visibility scores are noise if you want the statistics.

And when you check, read the whole answer, not just whether your name appears. Being cited with a wrong price is not a win. In our own testing, roughly one in eight brand facts came back wrong or outdated.

Common questions

For the full picture across every engine, read the guide to Answer Engine Optimization, or run a free scan to see where you stand today.