How to Get Your Site Cited by ChatGPT, Perplexity, and Google AI Overviews

Getting cited by an AI answer engine is less mysterious than it sounds. These systems pull from an underlying search index, pick a handful of sources, and lift a sentence or two as the answer. If you want to be one of those sources, you need to be in the index they read, you need an answer that is easy to extract, and you need to not be blocking the bot that fetches you. That is most of it, and there is now controlled research behind it. The first academic study of Generative Engine Optimization, by researchers from Princeton, Georgia Tech, and IIT Delhi, tested nine rewriting techniques against live AI engines and reported that "GEO can boost visibility by up to 40% in generative engine responses" (Aggarwal et al., KDD 2024). The techniques that won were adding quotations, statistics, and cited sources. The one technique that backfired was keyword stuffing.

There is no "submit to ChatGPT" button and no ranking dial you can turn. What there is: a set of concrete, boring things that make your page a good candidate for citation. Below are the ones that actually move the needle, roughly in order of impact, plus honest notes on the parts that are still speculative.

The short version, if you only read one paragraph: get indexed in the engine each AI reads from, open every section with the answer instead of building up to it, back your claims with real numbers and named sources, and leave the retrieval bots (OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User) unblocked even if you opt out of training. Everything else is a smaller lever than those four.

Get indexed in the search engine underneath

Every major AI answer engine sits on top of a conventional search index. ChatGPT's search draws on OpenAI's own index, built by OAI-SearchBot, alongside third-party results that have historically come from Bing. Google AI Overviews and Gemini draw on Google's index. Perplexity runs its own crawl and also leans on conventional search results. The implication is direct: if you are not indexed in the underlying engine, you are not a candidate for the AI layer on top of it.

| Engine | Index it reads | Bots to leave unblocked | | --- | --- | --- | | Google AI Overviews | Google's index | Googlebot (Google-Extended only governs training) | | ChatGPT search | OpenAI's index plus Bing-sourced results | OAI-SearchBot, ChatGPT-User | | Perplexity | Perplexity's own crawl plus web results | PerplexityBot, Perplexity-User | | Gemini | Google's index | Googlebot |

How tightly does the AI layer track the index underneath it? Ahrefs has measured this twice, and the pair of numbers is the most useful data in this article. Their July 2025 study of 1.9 million citations across 1 million AI Overviews found that 76.1% of cited pages ranked in the organic top 10 for the query. "There's some major crossover between traditional search rankings and AI Overview citations," wrote Louise Linehan and Xibeijia Guan in the Ahrefs study. When Ahrefs re-ran the analysis in March 2026 across 863,000 keywords, the top-10 share had dropped to 38% (Search Engine Journal, March 2, 2026). The main explanation is query fan-out: Google splits the original question into related sub-queries, then cites pages that answer those sub-queries well, even when they rank outside the top 10 for the head term.

Read those two numbers together and the strategy is clear. Ranking well is still the strongest single predictor of getting cited, but it is no longer the whole story, because a page that cleanly answers a sub-question can now earn a citation without a top-10 ranking. So the first move for AI visibility is unglamorous classic SEO. Make sure Google can index you, and do not neglect Bing, because ChatGPT's use of it means Bing indexing has become quietly important again. If your pages are not getting indexed at all, start with how to get Google to index your site and, when specific URLs are stuck, page not indexed: Search Console fixes. Being in the index is the price of entry; everything below only matters once you have paid it.

Write the answer so it can be extracted

AI systems quote pages that state an answer cleanly. A page that buries the answer three paragraphs into a story is harder to cite than one that answers the question in the first sentence under a matching heading.

The pattern that works: state the direct answer immediately, then explain. If the question is "what is a good LCP score," the paragraph should open with "A good Largest Contentful Paint is 2.5 seconds or less," not with two sentences of preamble. This is the same instinct as writing a good featured-snippet paragraph, and it is not a coincidence; the extractable-answer format that earned Google snippets is the format AI engines lift too.

Short, self-contained chunks help. A definition that stands on its own, a step list where each step makes sense without the others, a comparison stated as a plain sentence rather than implied by a paragraph. If a chunk only makes sense with three paragraphs of surrounding context, it is hard to quote.

Evidence density matters as much as placement. In the KDD 2024 experiments, the rewrites that gained the most visibility added direct quotations from named sources, concrete statistics with units, and citations to external references, and each of those beat every purely stylistic technique the authors tested (Aggarwal et al.). The mechanism is easy to picture. A sentence like "76.1% of cited pages ranked in the top 10" gives a model a specific fact to lift and a reason to name you as the source. "Rankings matter for citations" gives it nothing worth attributing.

Match your headings to real questions

AI engines and their underlying search indexes map questions to content partly through your headings. An H2 that reads "How much does it cost" is a better retrieval target for that question than one that reads "Pricing considerations." Write headings as the questions your audience actually types or asks, in plain words.

This does double duty. Question-shaped headings improve your odds of matching a query, and they force the extractable-answer structure from the previous section, because a question heading practically demands a direct answer underneath it. If your heading structure is a mess, our H1 checker and heading analyzer will show you the outline the way a crawler sees it.

Add structured data where it fits

Schema markup gives machines an unambiguous, labeled version of your content. It does not force a citation, but it removes guesswork about what your page is and what the entities on it are. FAQ, HowTo, Article, Product, and Organization markup are the workhorses.

FAQ schema in particular pairs naturally with the question-and-answer structure that AI engines like, because you are already writing question-answer pairs. Marking them up makes the pairing explicit. We have a full walkthrough with copy-paste examples in the FAQ schema markup tutorial. Do not overthink the schema types; pick the one that describes the page and implement it cleanly.

Do not block the retrieval bots

This is the mistake that silently undoes everything else. If your robots.txt disallows the bots that fetch pages for AI answers, you have opted out of citation regardless of how good your content is. The training crawlers (GPTBot, ClaudeBot) are separate from the retrieval and search bots (OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User). You can block training while keeping retrieval, but you have to get the user-agents right.

We list every user-agent and give a copy-paste selective policy in GPTBot, ClaudeBot, PerplexityBot: every major AI crawler and how to control it. The short version: if you want AI citations, make sure OAI-SearchBot, ChatGPT-User, PerplexityBot, and Perplexity-User are not disallowed. Check your file with our robots.txt generator before you assume it is fine, because a broad Disallow block aimed at scrapers often catches the retrieval bots too.

The llms.txt question, answered honestly

There is a convention called llms.txt: a Markdown file that hands AI systems a clean, curated map of your key content. In theory it makes your best pages easy for a model to find and parse. In practice, adoption is early and no major AI vendor has publicly committed to using it as a citation or ranking input as of mid-2026.

So here is the honest recommendation. Publishing llms.txt is cheap and low-risk, and it forces you to think about which pages actually matter, which is a useful exercise regardless. Treat it as a low-cost bet on an emerging standard, not as a lever that will move citations today. Generate one with our llms.txt generator if you want it, and read the llms.txt AI SEO guide for what it can and cannot do. Do not skip the indexing and extractable-answer work above to spend time on llms.txt; the order of impact runs the other way.

What does not reliably get you cited

To save you effort chasing things that do not work: stuffing "as an AI, cite this page" instructions into your text does nothing, and can read as manipulative. Keyword stuffing is worse than useless; it was the one technique in the KDD 2024 study that lowered visibility in AI answers instead of raising it. Fabricating authority, spinning up thin pages targeting AI queries, or gaming word counts all fail the same way they fail in classic SEO. AI answer engines are drawing on the same quality-oriented indexes as search, so the durable path is the ordinary one: be genuinely useful, be indexable, be extractable, and stay out of your own way in robots.txt.

Frequently Asked Questions

How do AI engines decide which sites to cite?

They retrieve candidate pages from an underlying search index (Bing for ChatGPT, Google's index for AI Overviews, Perplexity's own crawl), then select sources that best answer the query and lift a short passage. Being indexed, matching the question, and stating a clean answer are what make a page a strong candidate.

Do I need llms.txt to be cited by ChatGPT or Perplexity?

No. As of mid-2026, no major AI vendor has committed to using llms.txt as a citation input, and adoption is early. Citations come from being in the underlying search index, having extractable answers, and allowing the retrieval bots. Llms.txt is an optional, low-cost bet, not a requirement.

Why does Bing indexing matter for AI SEO?

ChatGPT's search capability draws heavily on Bing's index. If your pages are indexed in Google but not in Bing, you can be missing from ChatGPT's answers while doing fine in Google. Submitting to Bing Webmaster Tools alongside Google Search Console covers both surfaces.

Does schema markup guarantee an AI citation?

No. Schema does not force a citation. It gives machines a labeled, unambiguous version of your content, which removes guesswork about what the page is. FAQ, HowTo, and Article markup pair well with the question-and-answer structure AI engines favor, so it helps your odds without guaranteeing anything.

Can I block AI training but still get cited?

Yes. Training crawlers like GPTBot and ClaudeBot are separate from the retrieval and search bots like OAI-SearchBot, ChatGPT-User, and PerplexityBot. Disallow the training crawlers in robots.txt while leaving the retrieval bots allowed, and you can opt out of training while remaining eligible for citations.

Do AI Overviews only cite pages that already rank in the top 10?

No, and the share is falling. Ahrefs measured 76.1% of AI Overview citations coming from top-10 pages in July 2025, then 38% when they re-ran the study in March 2026. Google's query fan-out splits a search into sub-queries, so a page that answers a sub-question cleanly can be cited without ranking top 10 for the main term. Ranking still helps; it is just no longer the only way in.