How to Get Cited by AI Search: What the Data Shows

Q: How do I know if AI crawlers can even reach my site?

Check robots.txt for accidental blocks on retrieval bots (OAI-SearchBot, PerplexityBot, Claude-SearchBot, ChatGPT-User, Claude-User), then cross-check your CDN/WAF logs for those user agents — a permissive robots.txt doesn't guarantee a permissive edge configuration.

# How to Get Cited by ChatGPT, Perplexity, and Gemini: What Actually Correlates With Citations

TL;DR: The only controlled study on this (Princeton's GEO-bench) found that adding quotations, statistics, and cited sources to a page lifted its visibility in generative answers by 28–41%; a large-scale 2026 Ahrefs study found that adding schema markup, despite correlating with 53% of AI-cited pages, produced no measurable citation lift on its own. Ranking well in Google still matters — but mostly for Google's own AI Overviews, not for ChatGPT or Perplexity, which pull from a mostly different set of sources.

Most "AI SEO" advice right now is recycled intuition — add schema, write FAQs, "optimize for E-E-A-T" — repeated without anyone checking whether it moves a citation count. A handful of actual studies exist. Here's what they found, where they disagree, and what's worth acting on.

The only controlled experiment: Princeton's GEO-bench

The most-cited research in this space is "GEO: Generative Engine Optimization" (Aggarwal et al., accepted to KDD 2024), which built a benchmark called GEO-bench: 10,000 real user queries across 25 domains, each paired with the actual source documents a generative answer engine would draw from. The team then applied nine content-modification strategies to those source documents and measured the change in a "Position-Adjusted Word Count" visibility score, simulating the generative engine with GPT-3.5-turbo and validating a subset against Perplexity.ai.

The result table, against a baseline score of 19.5:

Strategy	Visibility score	Change vs. baseline
Quotation Addition	27.8	+41%
Statistics Addition	25.9	+33%
Fluency Optimization	25.1	+29%
Cite Sources	24.9	+28%
Technical Terms	23.1	+19%
Easy-to-Understand	22.2	+14%
Authoritative Voice	21.8	+12%
Unique Words	20.7	+6%
Keyword Stuffing	17.8	−9%

Two things worth flagging before you treat this as gospel. First, it's a 2023–2024 study run against GPT-3.5-turbo — not the retrieval pipelines that ChatGPT, Perplexity, and Gemini run in late 2026, which lean much more heavily on live web retrieval and re-ranking than the base model's parametric knowledge. Second, effectiveness was domain-dependent: statistics worked best for law, government, debate, and opinion content; quotations worked best for history, society, and explanatory content; citing sources worked best for factual/statement-style queries. There's no single tactic that wins everywhere. But the direction is consistent and mechanistically sensible — generative engines are extracting and re-ranking passages, and passages with a concrete, attributable claim are easier to lift into an answer than a paragraph of generic prose. Keyword stuffing being the only strategy that hurt is also notable, since it's still common advice in a lot of SEO-era thinking carried over into GEO content.

What doesn't seem to work: schema markup

The most useful recent data point is a negative result. In May 2026, Ahrefs published a study tracking 1,885 pages that added JSON-LD schema between August 2025 and March 2026, matched against 4,000 control pages that didn't, measuring citation-rate changes across Google AI Overviews, AI Mode, and ChatGPT.

Google AI Overviews: −4.6% (a small but statistically significant decline)
Google AI Mode: +2.4% (statistically indistinguishable from zero)
ChatGPT: +2.2% (statistically indistinguishable from zero)

Yet 53% of AI-cited pages in their dataset carried schema, roughly three times the rate of non-cited pages. Ahrefs' own read on this — echoed by their content marketer Si Quan Ong — is that the correlation is real but not causal: sites that bother with structured data also tend to be the ones investing in technical maintenance, authoritative content, and link-worthy publishing, and those are the things actually driving citations. Schema is a marker of a well-run site, not a lever you can pull in isolation.

This matters for the FAQ-schema advice you'll see everywhere in GEO content, too. Google deprecated FAQ rich results in Search entirely as of May 7, 2026, removing the rich-result report and Search Console support by June and August. FAQPage markup isn't harmful to leave in place, but it's no longer earning you anything in Google Search, and the Ahrefs data suggests it wasn't earning much in AI citations either. What plausibly still helps is the format — a direct, self-contained, factually-anchored answer under each question — independent of whether it's wrapped in schema at all.

Ranking still correlates — but only for one engine

Ahrefs first put that overlap at 76.1% in July 2025, but its updated analysis of 863,000 SERPs revised the figure down to 38%. The gap is the story: Google increasingly assembles AI Overviews from the sub-queries it fans a search out into, rather than from page one of the original query. Ranking in the top 10 still improves your odds of being cited — it is just no longer where most citations come from.

That correlation is specific to Google's own AI Overviews, which are built on Google's own index. It does not transfer cleanly to the standalone answer engines. Profound's analysis of 11.84 billion citations across ChatGPT, Claude, Gemini, Copilot, Grok, and Perplexity between April and July 2026 found each platform's citation set diverging sharply — different engines are drawing from different retrieval indexes, licensing deals, and crawl histories, not a shared "who ranks #1" signal. A separate citation-source analysis put the domain overlap between ChatGPT's and Perplexity's citation sets at roughly 11%, and the overlap between AI citations generally and Google's top-10 organic results at around 12%. The two Ahrefs numbers (76% vs. this 12%) aren't contradictory — they're measuring different products: Google's AI Overview is closer to a rewritten SERP, while ChatGPT and Perplexity behave more like independent retrieval systems with their own source preferences.

Concrete platform differences worth knowing if you're prioritizing effort:

Perplexity cites an average of 21.9 sources per response, more than double ChatGPT's 10.4, per the ChatGPT/Perplexity citation source index.
Perplexity has the highest share of .edu (3.2%) and country-code TLD citations of the major engines, consistent with a bias toward institutional and government sources.
Wikipedia remains the single most-cited source inside ChatGPT, though its share reportedly dropped from roughly 55% of prompts to under 20% after a September 2025 retrieval change — a reminder that these baselines shift under you without warning.
Per 5W's 680-million-citation analysis, the top 15 domains account for roughly 68% of all AI citation volume across categories — concentration that exceeds anything organic search produced even at Google's peak.

The one thing none of this works without: crawl access

None of the above matters if the retrieval crawler can't reach your pages. Each AI platform runs a distinct set of user agents — a training crawler (feeds the base model) and a separate retrieval/search crawler (fetches pages live to ground an answer). Blocking the wrong one silently kills your citation eligibility on that platform. A reasonable, current robots.txt posture — blocking training bots while explicitly allowing the retrieval bots that ground live answers — looks like this:

# Block bots that scrape for model pretraining
User-agent: GPTBot
Disallow: /

User-agent: CCBot
Disallow: /

User-agent: Google-Extended
Disallow: /

# Allow bots that fetch pages to ground a live answer / citation
User-agent: OAI-SearchBot
Allow: /

User-agent: ChatGPT-User
Allow: /

User-agent: PerplexityBot
Allow: /

User-agent: Claude-SearchBot
Allow: /

User-agent: Claude-User
Allow: /

Sitemap: https://example.com/sitemap.xml

Check this against your actual server logs, not just robots.txt — CDN or WAF rules block these user agents independently and more often than people expect.

FAQ

Does adding FAQPage schema get you cited by ChatGPT?

The best available evidence (Ahrefs, 2026) says no — schema addition produced no statistically significant citation lift on ChatGPT or Google's AI Mode, and a small negative effect on AI Overviews. Google also deprecated FAQ rich results from Search entirely in May 2026. The Q&A format (a direct, self-contained answer per question) may still help; the markup wrapper around it likely doesn't.

What single content change has the best evidence behind it?

Adding specific, attributable statistics and direct quotations, per the Princeton GEO-bench study, which measured +33% and +41% visibility gains respectively against a controlled baseline. Vague, unsourced claims are the easiest thing for a generative engine to skip over when assembling an answer.

Does ranking #1 on Google guarantee an AI citation?

No. It correlates for Google's own AI Overviews (76% of cited pages also rank top 10 organically) but far less for independent engines like ChatGPT and Perplexity, which draw from largely different source sets — only around 11–12% domain overlap by some estimates.

How do I know if AI crawlers can even reach my site?

Check robots.txt for accidental blocks on retrieval bots (OAI-SearchBot, PerplexityBot, Claude-SearchBot, ChatGPT-User, Claude-User), then cross-check your CDN/WAF logs for those user agents — a permissive robots.txt doesn't guarantee a permissive edge configuration.

---

If you're auditing a site's technical readiness for AI crawlers as part of a broader engineering review, that's the kind of infrastructure work we do at Wise Hustlers.