Guide

How to get cited by ChatGPT, Perplexity and AI engines

By Fabien Cavanna, Going for Growth · June 29, 2026 · 8 min read

In shortNobody can guarantee a citation in ChatGPT or Perplexity: no publisher controls what a model chooses. You can, however, honestly improve your chances with established levers: answer the question clearly and early, structure the content so it is easy to extract, stay crawlable and indexable, and build authority (mentions and citations elsewhere on the web). Structured data from schema.org helps comprehension and rich results in classic search, but its causal effect on AI citations is not demonstrated. The llms.txt file is a proposed convention, not adopted by the major engines to date.

What can, and cannot, be promised about AI citations?

This topic attracts a lot of dubious promises: add this file and you will be cited, 3x more citations thanks to schema, GEO guarantee. The reality is more sober.

You improve your chances of being cited by ChatGPT or Perplexity by answering the question clearly at the top of the page, structuring the content for extraction, staying indexable and building authority. No publisher can guarantee a citation: source selection depends on proprietary models, on the user's query and on timing. What you can do is reduce the friction that prevents an engine from finding, understanding and reusing your content.

This guide systematically separates what is established (documented by primary sources) from what is emerging or uncertain. If a claim circulates widely on blogs but has no solid source, we say so.

How these engines find and cite a source

Generative answer engines do not all work the same way, but a common principle emerges: they rely heavily on a search index and on document retrieval, not on pure internal model knowledge.

Google AI Overviews and AI Mode

Google states in its official documentation that AI answers are grounded in its search index. For a page to be able to appear as a supporting link in AI Overviews or AI Mode, it must be indexed and eligible to appear in Google Search with a snippet. Practical consequence: if you are not indexable, you are not a candidate. A nuance documented by serious analyses: the selection does not follow the exact order of the web ranking, and pages far below the first position are frequently cited.

Perplexity

Perplexity builds its cited answers through a RAG-type mechanism (document retrieval then generation), combining index and crawl, favoring factual, structured and authoritative sources. This general principle is plausible and widely described, but the specific figures that circulate (index size, number of pipeline stages, model names) come from third-party, unofficial blogs: treat them as unverified.

ChatGPT

When ChatGPT cites sources, it does so in web search mode, again via retrieval of pages on the indexable web. The detail of the internal ranking is not public.

The common thread: being findable in an index and easy to extract is the shared prerequisite. For SEO/AI fundamentals, see our guide to SEO for AI and GEO.

The actions that improve your chances

None of these actions is a guarantee. Together, they remove the most common obstacles between your content and a citation. The order follows an implementation logic.

Identify the target questions. List the real questions your prospects ask an AI (natural phrasings, not raw keywords). This is the foundation: a generative engine answers a question, and your content must match a specific question.
Answer clearly and right away. Give the direct answer at the top of the page, in a few self-contained sentences, before the details. A short paragraph that answers the question on its own is easier to extract and cite than an answer buried in the middle of the text.
Structure for machine reading. Explicit headings, lists, citable definitions, tables, an FAQ. A clear structure helps extraction. The foundational academic paper on GEO (Princeton et al., arXiv 2311.09735) shows that tactics such as citing sources, adding attributed quotations and statistics improve visibility in generative answers, with a reported gain of up to 40 percent. An honest caveat: that experiment was run on a system imitating Bing Chat and Perplexity in 2023-2024, not on Google AI Overviews, so do not over-generalize.
Add structured data (schema.org). Its role is established for content comprehension and rich results in classic search. Google does point out, however, that no special schema is required for its AI features, and the causal link between schema and AI citations is not demonstrated. Do it as a sound underlying best practice, not as a magic lever. Details in our guide to llms.txt and structured data.
Strengthen authority and entity. Mentions, citations and links from other recognized sites, a consistent entity page (who you are, your expertise, your evidence). Engines favor sources that carry authority elsewhere on the web. This is the slowest lever but often the most decisive.
Make sure you are crawlable and indexable. No unintended robots.txt blocking, no noindex, content accessible without heavy JavaScript, fast pages. If an engine cannot read the page, nothing else matters. Also check that your pages are properly indexed in Google.
Track and iterate. Regularly test your target questions in ChatGPT, Perplexity and Google, note whether you are cited, and adjust the pages that are not. Measurement is imperfect (see below), but iteration beats optimizing blindly.

A concrete and honest example: this site (goingforgrowth.net) applies structured data itself and publishes an llms.txt. That guarantees nothing on its own, but it is consistent with the best practices described here.

What does not work (the false good ideas)

So much wasted effort, sometimes even counterproductive.

Keyword stuffing. Inherited from old SEO, useless for engines that reason about meaning. It degrades readability, and therefore extractability.
Mass-generated content with no value. Producing hundreds of empty pages to cast a wide net does not build authority and risks being ignored or penalized. Better to have a few pages that genuinely answer.
GEO guarantee promises. No provider controls what a model chooses. A citation guarantee is a red flag, not a credible selling point.
Relying on llms.txt as a citation lever. It is a convention proposed by Jeremy Howard (Answer.AI) on September 3, 2024: a Markdown file at the root listing your key pages. Its adoption by the major engines is very limited, even unconfirmed. Google explicitly states it does not use it (John Mueller compares it to the old keywords meta tag, a self-declared and therefore manipulable signal). An Ahrefs study of 137,000 sites shows that nearly all llms.txt files are never read by crawlers. Publishing it can make sense for agents or internal documentation, but do not present it as a guarantee of being cited by ChatGPT or Google.
Unsourced miracle figures. 3.2x more citations with schema, 82 percent of Perplexity in under 30 days: these statistics circulate without a verifiable primary source. Do not build a strategy on them.

How do you measure whether you are cited by the AIs?

You measure your AI citations by trend, by testing your target questions in ChatGPT, Perplexity and Google at regular intervals, because no official, exhaustive tool exists. Measurement is genuinely difficult, and it is better to own that than to pretend otherwise.

Why it is difficult

Engine answers are non-deterministic: the same question can return different sources depending on the moment, the user, the conversation context.
There is no official, exhaustive measurement tool provided by the engines.
Referral traffic from AIs is often poorly attributed in analytics.

What you can do

Test manually, at regular intervals, your target questions in ChatGPT, Perplexity and Google, noting whether and how you are cited.
Monitor server logs to spot AI crawlers passing through (they identify themselves by their user-agent).
Track referral traffic attributable to the AI engines' domains, knowing that it is partial.
Iterate page by page: when a page is not cited on its target question, review the clarity of the answer at the top, the structure and the authority.

Treat the results as trends, not as exact metrics. If you want to industrialize this tracking (regular tests, citation collection), an AI automation can save time, without turning a probabilistic approach into a certainty.

Frequently asked questions

Can you guarantee being cited by ChatGPT?

No. Source selection depends on proprietary models, on the query and on the context, beyond the control of publishers. You can improve your chances by answering the question clearly, structuring the content, staying crawlable and building authority, but any guarantee promise is a red flag.

Do you have to pay to appear in AI answers?

No, there is no official paid placement to appear as a cited source in the organic answers of ChatGPT, Perplexity or Google's AI Overviews. These engines select sources through retrieval on the indexable web. This may change, and it is distinct from any advertising formats the platforms develop separately.

How long does it take?

There is no reliable timeframe to announce, and be wary of figure-based promises (such as a few weeks). It depends on the indexing of your pages, on the competition for the question and above all on building authority, which is slow. Treat it as ongoing background work, not as a one-off action with a dated result.

Does structured data help with getting cited?

Its role is established for content comprehension and rich results in classic search, so it is a good practice. However, Google indicates that no special schema is required for its AI features, and no proven causal link between schema.org and AI citations is demonstrated. Do it for the right reasons, without expecting a magic effect on citations.