How to Get Cited by ChatGPT and Perplexity

To get cited by ChatGPT and Perplexity, be the single clearest, most specific source on a narrow question, and write the answer as a passage a model can lift whole. AI answer engines retrieve and quote text that directly answers the user, so a self-contained factual paragraph backed by primary sources beats a long, hedged page every time. You cannot guarantee a citation, but you can raise your odds a lot. Here is how, and why measuring the result is harder than earning it.

Quick answer

How do AI answer engines actually pick sources?

Retrieve

candidate passages

Clarity wins

clearest specific source

Quote

a self-contained answer

Cite

you get named

Retrieve

candidate passages

Clarity wins

clearest specific source

Quote

a self-contained answer

Cite

you get named

ChatGPT (with search), Perplexity, Google AI Overviews, Gemini, and Copilot all work in roughly the same two steps: retrieve, then synthesize. They pull a set of candidate passages that seem to answer the query, write an answer from them, and attach citations to the sources they leaned on. That means you are not competing for a ranked list of ten links. You are competing to be one of a handful of passages the model trusts enough to quote and name.

Three things decide whether your passage makes the cut:

• Relevance to a specific question. Not a topic, a question. The tighter the match between the user's phrasing and your answer, the better.
• Extractability. The model wants a clean, self-contained chunk it can lift without dragging in context from three paragraphs up.
• Trust signals. Consistent identity, primary sources, dates, and structure all tell the model your passage is safe to repeat.

None of this is exotic. It rewards being genuinely clear and genuinely expert, which is exactly why cheap tricks fade fast.

Why does being "the clearest specific source" beat being comprehensive?

Big generic guides lose to small precise ones in AI answers, and this surprises people. If someone asks Perplexity "what value should I send with offline conversions," the model does not want your 4,000-word pillar on measurement. It wants the two sentences that answer that exact question. A page that nails one narrow query, and says so plainly in its first line, is a cleaner retrieval target than a sprawling page where the answer is buried on paragraph nine.

So narrow your targets. One question, one confident answer, near the top. I would rather own the definitive 300-word answer to a specific query than place tenth on a broad one. Broad pages still matter for humans and for topical authority, but the citation itself is usually won by the most specific source in the room.

What makes a passage quotable by an AI?

A quotable passage answers the question in its first sentence, uses concrete nouns instead of vague ones, and does not depend on the surrounding text. Compare these two:

Weak (not extractable):

As we discussed above, there are several factors here, and it really depends on your situation, but generally it can be a good idea to consider this approach.

Strong (extractable):

Send the true business value of the conversion, not the lead value. For a medical practice, that means the revenue from the patient who actually showed up and paid, passed back to Google Ads as an offline conversion keyed to the click id.

The second one names the thing, answers the question, and stands alone. A model can quote it and attribute it without guessing what "this approach" meant. That is the entire game in one example.

Practical rules for quotable passages:

• Lead with the answer, then explain.
• Keep one idea per paragraph.
• Use specific numbers, names, and steps where you honestly have them.
• Prefer short tables and lists for anything comparative or sequential.
• Add a date so freshness is legible.

Which technical signals raise your citation odds?

These are the levers that genuinely move the needle. Treat this as a checklist:

• Answer-first intro on every page, one or two sentences that resolve the title.
• Question-style H2s that match how people and models phrase queries.
• Self-contained sections that each answer their heading in full.
• Schema.org structured data: Article, FAQPage, HowTo, and Organization or Person for identity.
• Entity clarity: a consistent name, bio, and sameAs links to your profiles so the model knows who is talking.
• Primary sources cited inline, so claims are traceable.
• Visible dates and periodic freshness updates.
• Concise tables and lists for comparisons and steps.
• Server-rendered content, not answers injected by client-side JavaScript a crawler may never run.
• Internal links and topical clusters that build authority around a subject.

Entity clarity deserves a note. AI engines are more confident citing a source they can identify. If your name, your role, and your work are described consistently across your site and your profiles, and you mark that up with Person or Organization schema plus sameAs, you become a known entity instead of an anonymous page. Known entities get cited more, because the model can attribute the claim to a known author with more confidence.

Does an llms.txt file help you get cited?

An llms.txt file is a low-cost, optional signal, not a requirement. It is a plain-text file at your site root (/llms.txt, sometimes paired with a fuller /llms-full.txt) that points AI crawlers to your cleanest, most important content in Markdown. Some tools read it today, many ignore it, and none of them will cite a site whose actual pages are thin or hidden. Add it because it is cheap and forward-looking, not because it will carry a weak site. Your crawlable, well-structured HTML is still doing the heavy lifting.

A concrete example: turning one page into a citable source

Take a page targeting "how to send offline conversions to Google Ads." Here is the before and after.

Before: a 2,500-word essay that opens with the history of conversion tracking, defines terms for 600 words, and finally explains the actual steps around the midpoint. No schema. Author is "admin." No dates. The steps are rendered by a JavaScript widget.

After:

1. First line answers it: "To send offline conversions to Google Ads, capture the gclid at form submission, store it with the lead, and upload the click id, value, and timestamp back to Google Ads once the deal closes."
2. Question-style H2s: "How do you capture the gclid?", "What value should you send?"
3. HowTo and FAQPage schema, plus Person schema naming the author with sameAs.
4. The steps live in server-rendered HTML, not a widget.
5. A visible "updated" date and a link to Google's own documentation as the primary source.

Same expertise, restructured so a model can retrieve, quote, and attribute it. This is the work I do on my own site and for measurement clients: make the page findable, quotable, and, crucially, measurable. I will not promise a specific citation count, because GEO is probabilistic, but the structural difference is real and repeatable.

Why measurement is the part everyone skips

Here is the trap. Much AI-referral traffic arrives with no clean referrer, so your analytics files it as direct or unattributed. You earn citations, real high-intent visitors click through, and your reports show nothing changed, because the channel is invisible by default. GEO's traffic volume is small but the intent is high, which makes the undercount especially costly: you conclude the work does nothing and stop, right when it was starting to pay.

To isolate it, you need to catch the referrers that do identify themselves (Perplexity and ChatGPT increasingly pass one), segment those sessions, and tie them to what actually happened downstream: the form filled, the call booked, the order paid. That is the same offline-truth problem I solve for ad platforms, applied to a new source. If you want that channel wired up properly, my patient conversion tracking work does exactly this for practices, and my broader conversion tracking and attribution service does it for any business that needs to connect a click to a real outcome.

The honest limits

GEO complements traditional SEO, it does not replace it, and it rewards genuine expertise over tricks. You cannot guarantee a citation, the engines change their behavior often, and the same query can cite different sources from one week to the next. What you can do is stack the odds: be the clearest specific source, write quotable self-contained passages, cite primary sources, sharpen your entity signals, and then measure the channel so you actually know it is working.

If you are new to the topic, start with what generative engine optimization is for the foundations, then come back and apply the levers above.

Want your site restructured to be both citable by AI and correctly measured, so the traffic stops hiding in "direct"? That second half is where most people lose the plot, and it is the half I am built for. Tell me the query you want to own, and I will tell you what stands between you and owning it.