AI engines cite pages they can retrieve and verify. When you ask a question, the engine runs several background searches, pulls indexed pages from the results, reads them in passages, and cites the passages that answer a sub-question directly. So citation starts with being indexed, is decided at the level of a single passage, and is won by content with facts worth quoting. Below is what Google, OpenAI, and Perplexity each document about their own process, and what the research measured about content that gets cited.
Every AI answer that names a business had to choose that business from somewhere. The choosing follows a process, parts of that process are publicly documented, and one part has been measured in a peer-reviewed study. This post walks through what is actually known, engine by engine, with a source for every claim.
One caution before the detail. These engines change their behaviour without notice, so treat every per-engine claim here as accurate at the time of writing and verify by observation for your own business. The mechanics below are the documented parts, which move slower than the day-to-day behaviour.
Where do AI answers come from?
Retrieval. An AI search engine does not answer from its training memory alone. It searches the web, retrieves pages, and generates an answer grounded in what it retrieved, then cites those pages. The research field calls this retrieval-augmented generation, described by Lewis et al. in 2020 as combining a language model with an external index it can pull from at answer time. Google says the same thing about its own product: AI Overviews are "built to only show information that is backed up by top web results."
This is the single most useful fact in the whole topic. If your page cannot be retrieved, you cannot be cited, no matter how good the content is. Retrieval runs on boring plumbing: crawling, indexing, and rendering. That is why an unindexed page is a dead page in AI search.
How does Google choose citations for AI Overviews and AI Mode?
Google documents three things about its selection process. First, eligibility: a page must be indexed and eligible to appear in Google Search with a snippet. Standard technical requirements apply, and meeting them still does not guarantee anything. Second, retrieval: AI features use what Google calls query fan-out, "issuing multiple related searches across subtopics and data sources," which lets them "display a wider and more diverse set of helpful links" than the classic ten blue links. Third, there is no separate optimization track: "There are no additional requirements to appear in AI Overviews or AI Mode, nor other special optimizations necessary."
Google also documents a passage-level ranking system, which it describes as identifying "individual sections or 'passages' of a web page to better understand how relevant a page is to a search." Your page is competing one passage at a time, and a strong self-contained answer buried in an otherwise unrelated page can still be selected.
What fan-out means for a business that does not rank first
Fan-out changes who gets cited. Because the engine runs several background searches on sub-questions, it pulls sources from those sub-query results, and many of them never ranked for the visible query at all. Two large studies show both sides of this. seoClarity, across 362,000 US queries, found that 94% of AI Overviews cite at least one URL from the top 20 organic results, so ranking still matters. Ahrefs, across 863,000 SERPs and 4 million AI Overview citations (March 2026), found that only about 38% of citations ranked in the top 10 for the triggering query. The two numbers measure different things and both are true: most AI answers lean on well-ranked pages somewhere in the mix, and most individual citations come from outside the visible top 10.
For a business that is not ranking first, that second number is the opening. You do not need to beat the market leader on their head term. You need to be the best retrievable answer to one of the sub-questions the engine asks on the way.
How does ChatGPT choose what to cite?
OpenAI documents the plumbing rather than the ranking. Three separate bots matter, and they do different jobs. OAI-SearchBot surfaces websites in ChatGPT's search features, and OpenAI states plainly that "sites that are opted out of OAI-SearchBot will not be shown in ChatGPT search answers." GPTBot collects content for model training, and blocking it does not affect search visibility. ChatGPT-User fetches pages when a user asks about them directly.
The practical consequence: a robots.txt rule written in 2023 to block "AI scrapers" may be silently removing you from ChatGPT search answers in 2026. Check which bot you blocked. Blocking GPTBot is a training opt-out. Blocking OAI-SearchBot is a visibility opt-out.
How does Perplexity choose what to cite?
Perplexity documents its crawlers the same way. PerplexityBot "is designed to surface and link websites in search results" and is not used for training foundation models. A second agent, Perplexity-User, visits pages when a user's question requires it. Perplexity has included citations in every answer since launch and runs a publisher revenue-share program on top of them.
What Perplexity has never published is a ranking formula. You will find blog posts claiming exact signal weights for its citation choices. We looked for a primary source for those numbers and found none, so this post does not repeat them.
| Engine | Search crawler | What is documented | Your first check |
|---|---|---|---|
| Google AI Overviews / AI Mode | Googlebot | Indexed + snippet-eligible pages; query fan-out; no special AI requirements | Is the page indexed in Search Console? |
| ChatGPT search | OAI-SearchBot | Opted-out sites are not shown in search answers; training (GPTBot) is separate | Does robots.txt allow OAI-SearchBot? |
| Perplexity | PerplexityBot | Crawler surfaces and links sites; citations in every answer; no published formula | Does robots.txt allow PerplexityBot? |
What content actually wins citations?
The best evidence is the GEO paper by Aggarwal et al., presented at KDD 2024. The researchers took source pages, applied one content change at a time, and measured how much visibility each change earned in generative engine answers across a 10,000-query benchmark. The winners were changes that made pages more citable, and the losers were classic SEO tricks.
Fluency editing (rewriting for clarity without changing the facts) lifted visibility about 29%. Keyword stuffing, the oldest trick in SEO, produced little to no improvement. The pattern across every winning change is the same: generative engines favour passages that carry concrete, attributed information they can hand to the user with confidence.
The paper's most encouraging finding is about who benefits. In the benchmark, pages sitting fifth in the underlying search results gained 115% visibility from adding source citations, while top-ranked pages actually lost share. The tactics help lower-ranked sites most. For a Lagos SME competing against bigger brands with older domains, this is the rare game where the smaller player has the structural advantage, because most incumbents have not made their content citable either.
Which sites get cited, and why you should not chase the list
Profound analysed 680 million AI citations collected between August 2024 and June 2025. Wikipedia was ChatGPT's most-cited source at 7.8% of all its citations. Reddit was Perplexity's at 6.6%. Lists like this tempt businesses into building their whole strategy around one platform.
The volatility data argues against that. Semrush tracked 230,000 prompts weekly from July to October 2025 and watched Reddit fall from roughly 60% of ChatGPT responses to about 10% within weeks in September, while Forbes citations doubled over the same window. One platform decision rewrote the source mix overnight. The durable strategy is consensus across several surfaces: your own site as the canonical answer, directory profiles that agree with it, and community and media mentions that confirm it. Any single channel can be turned off by someone who is not you.
What should a business do with all this?
Five checks, in the order the engines apply them:
- Be indexed. Google's documented eligibility rule for AI features starts here. Check every important URL in Search Console. An unindexed page cannot be cited.
- Allow the search crawlers. Confirm robots.txt does not block OAI-SearchBot or PerplexityBot. Blocking training bots is a separate decision with a separate line.
- Answer in self-contained passages. Selection is passage-level. Put the question in the header and a complete answer in the first paragraph under it, readable without the rest of the page.
- Give the engine something to carry. A named statistic, a sourced claim, a quotable sentence. This is the 27% to 41% lift the GEO paper measured, and it is also just honest writing.
- Build agreement off-site. Consistent profiles and third-party mentions give engines confirmation that your site's claims are real. Our guide to schema and the entity graph covers the on-site half of this identity work.
One thing you do not need: a special AI file. Google states that no new machine-readable files or markup are required for its AI features. We covered the honest status of llms.txt separately, and it has not changed.
Want to know if AI engines can cite you?
Our free AI Visibility Audit covers 12 checks, including indexing, crawler access, and answer structure, each with a 5-minute test and the exact fix. Straight to your inbox.
✓ Check your inbox, your checklist is on the way!
We respect your privacy. Unsubscribe at any time.
AI engines cite the page a background search retrieved, whose passage answered a sub-question on its own, and whose facts were worth carrying into the answer. Get indexed, leave the search crawlers open, answer in self-contained blocks, attach real numbers to real sources, and let several surfaces confirm who you are. Every step is checkable, and none of it requires a trick.
Frequently asked questions
How do AI engines choose which sources to cite?
AI search engines answer by retrieval. They run background searches on your question, pull pages from a search index, read them in passages, and generate an answer grounded in those passages, citing the pages they used. Google documents this as query fan-out and states that AI Overviews only show information backed up by top web results. So citation starts with being indexed and retrievable, and it is won by the passage that answers a sub-question on its own.
Do I need a special file or markup to appear in AI answers?
No, according to Google. Its documentation states there are no additional requirements to appear in AI Overviews or AI Mode, and no new machine-readable files or markup are needed. Standard indexing and quality requirements apply. Structured data still helps machines understand your pages, and each engine has a search crawler (OAI-SearchBot for ChatGPT, PerplexityBot for Perplexity) that your robots.txt must not block.
Does ranking number one on Google guarantee AI citations?
No. seoClarity found that 94% of AI Overviews cite at least one top-20 organic URL, so ranking clearly helps. But Ahrefs measured that only around 38% of AI Overview citations ranked in the top 10 for the triggering query, because query fan-out pulls sources from related background searches. A first-place ranking improves your odds without guaranteeing a citation, and pages far outside the top 10 still get cited.
What kind of content gets cited most?
The GEO research paper (KDD 2024) tested content changes across 10,000 queries. Adding quotations lifted visibility by about 41%, adding statistics by about 33%, and citing sources by about 27%, measured against the unmodified baseline. Keyword stuffing produced little to no improvement. Concrete, attributed facts win citations. Keyword repetition adds nothing.
Do ChatGPT and Perplexity cite the same sources?
No. Profound's analysis of 680 million citations found Wikipedia was ChatGPT's most-cited source while Reddit was Perplexity's. Semrush's tracking also showed the mix changes fast: Reddit appeared in roughly 60% of ChatGPT responses before September 2025, then fell to about 10% within weeks. Each engine has its own source mix, and that mix can change overnight.
How do I check whether my business can be cited?
Start with three checks. Is the page indexed? Google Search Console shows this per URL, and an unindexed page cannot be cited by AI Overviews. Does your robots.txt allow OAI-SearchBot and PerplexityBot? And does the first paragraph under each heading answer that heading's question on its own? Our free AI Visibility Audit walks through 12 checks like these, each with a five-minute test and a fix.
References & sources
- Google Search Central, AI features and your website (query fan-out, eligibility, no special requirements)
- Google, AI Mode in Google Search: Updates from Google I/O 2025 (May 2025)
- Elizabeth Reid, Google, AI Overviews: About last week (May 2024)
- Google Search Central, A guide to Google Search ranking systems (passage ranking)
- OpenAI, OpenAI bots documentation (OAI-SearchBot, GPTBot, ChatGPT-User)
- Perplexity, Perplexity crawlers (PerplexityBot, Perplexity-User)
- Aggarwal et al., GEO: Generative Engine Optimization (KDD 2024)
- Lewis et al., Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks (NeurIPS 2020)
- Ahrefs, AI Overview citations study (March 2026)
- seoClarity, AI Overview and organic rankings overlap (updated October 2025)
- Profound, AI platform citation patterns (updated August 2025)
- Semrush, The most cited domains in AI search (November 2025)