AI SEARCH CITATIONS
Who Gets Cited in AI Answers?
Which brands and sources ChatGPT, Perplexity and Google AI Overviews actually cite, why they cite them, and how to measure your own share. This is the public evidence plus the method we use, not a benchmark you have to wait for.
The figures here are either cited public findings or clearly-labelled illustrative examples, not a VerityAI benchmark study.
IN SHORT
Who Gets Cited, and Why?
AI answers cite a small set of sources per query, and the pattern is fairly consistent across engines. Google AI Overviews leans on pages that already rank and on trusted reference sites. ChatGPT and Perplexity lean on their web index plus sources that are easy to quote cleanly: clear, well-structured, well-sourced pages. Across engines, three things keep coming up in the public research. Wikipedia, Reddit and established review and reference sites are cited heavily. Pages that include citations, direct quotations and concrete statistics get surfaced more often. And being cited is not the same as ranking first in classic search. This page sets out what the public evidence shows and gives you a repeatable method to measure your own AI-citation share. The figures below are either cited public findings or clearly-labelled illustrative examples, not a VerityAI benchmark.
3
Engines to test first
20-50
Buyer prompts to freeze
6
Signals that lift citation
The numbers above describe the method on this page. They aren't measured results for any site.
THE EVIDENCE
What Does the Public Research Actually Show?
Four findings that hold up across independent sources. Each row carries its source. We haven't invented a citation-share number of our own here.
Public evidence on AI citations
| Finding | What it means | Source |
|---|---|---|
| Citations, quotations and statistics lift visibility | Adding cited sources, direct quotations and concrete statistics to a page measurably raised how often generative engines surfaced it in the GEO study. | GEO study, Aggarwal et al., arXiv 2311.09735 |
| AI Overviews now sit on a large share of searches | Google AI Overviews appear on a meaningful and growing share of US searches, and users click through to source links less often when an AI answer is shown. | Pew Research Center, July 2025 |
| A handful of domains dominate AI citations | Independent citation trackers repeatedly find Wikipedia, Reddit and a short list of reference and review sites among the most-cited sources across engines. | Public AI-citation tracking (Profound, Peec AI, Ahrefs Brand Radar and similar) |
| Being cited is not the same as ranking first | The source an AI engine quotes is often not the top classic search result. Citation graphs and the ranked link list overlap, but they are not identical. | Observed across AI answer engines; consistent with the GEO study |
Cited public sources only. No VerityAI benchmark figure is presented here as a measured result.
THE SIGNALS
Which Signals Make a Page More Likely to Be Cited?
Six signals the public research keeps pointing at. These are grounded in named sources, not invented weightings. They're listed roughly in the order they tend to matter for a page trying to earn citations.
Already visible in classic search
Especially for Google AI Overviews, sources that already rank for the query are the pool the engine draws from. Classic SEO is table stakes for AI citation, not a replacement for it.
AI Overviews draw heavily on pages that already rank. Documented in Google Search guidance and observed in citation tracking.
Quotable, self-contained answers
A clear answer stated near the top, in a clean paragraph or list, is easy for an engine to lift as a self-contained span. Buried or rambling answers are harder to quote cleanly.
Retrieval-augmented engines select and quote passages. Answer-first structure matches how they pull content.
Citations, quotations and statistics on the page
Pages that cite their own sources, quote named people, and use concrete figures get surfaced more often. This is the strongest single finding in the public research.
GEO study, arXiv 2311.09735. Adding these elements measurably raised generative-engine visibility.
Structured data a machine can parse
Organization, Article and FAQPage schema tell a crawler what the page is, who wrote it and when, without guessing from the layout. Clean question-and-answer markup is among the most citable structures on a page.
Organization, Article and FAQPage structured data are documented by Google Search Central.
Named author and expertise signals
An identifiable human author with a real bio and credentials is a trust signal. Generative engines lean toward sources that show who is speaking and why they would know.
Author and expertise signals sit under Google E-E-A-T guidance and are echoed in how AI engines weight source trust.
Entity presence and third-party mentions
Being a recognised entity, with mentions on Wikipedia, Reddit, reviews and reputable third-party sites, feeds the pattern the trackers keep finding. Engines cite sources they already treat as established.
Consistent finding across public citation trackers: reference and community sites are cited disproportionately.
THE METHOD
How Do You Measure Your Own AI-Citation Share?
A repeatable, five-step method you can run yourself. It turns a vague worry (are we invisible to AI search?) into a number you can track and a work queue you can act on.
Build a prompt set
List 20 to 50 buyer questions your customers actually ask an AI engine. Mix informational, comparison and buying-intent phrasing. Write them in the spelling and English your buyers use. Freeze the list before you start so results stay comparable over time.
Run each prompt across the engines that matter
Run every prompt through ChatGPT, Perplexity and Google AI Overviews at a minimum, and add Gemini and Copilot if your buyers use them. Record the answer and every source the engine cites, one row per prompt per engine.
Log a fixed set of fields per answer
For each answer capture: whether your brand is cited or mentioned, which domains are cited, the content type of each cited page (reference, community, blog, tool, news), and whether a competitor is cited instead. Keep the fields identical every run.
Score citation share and the gap
Citation share is the percentage of prompts where you are cited. The gap is the set of prompts where a competitor or a reference site is cited and you are not. That gap list is the work queue: the pages and entities to build or fix.
Re-run on a schedule
AI answers drift as models and indexes update. Re-run the frozen prompt set monthly or quarterly and track citation share over time. A one-off snapshot tells you where you stand. The trend tells you whether your changes worked.
Tools such as Profound, Peec AI and Ahrefs Brand Radar automate parts of this. The method is the same whether you run it by hand or with a tool: freeze a prompt set, log who gets cited, measure the gap, re-run on a schedule.
WORKED EXAMPLE
What Does a Citation-Share Measurement Look Like?
A made-up example to show how the method produces a number and a work queue. This is illustrative. It isn't a real measurement of any company.
Illustrative citation-share snapshot for a fictional brand
| Metric | Value | Read |
|---|---|---|
| Prompts in the set | 30 | Buyer questions, frozen before the run |
| Engines tested | 3 | ChatGPT, Perplexity, Google AI Overviews |
| Answers logged | 90 | 30 prompts across 3 engines, one row each |
| Your brand cited | 6 of 90 | Citation share about 7%. Reference and community sites dominate the rest. |
| Biggest gap | Comparison prompts | A competitor is cited on most "X vs Y" prompts. Fastest lift: build the comparison content and its schema. |
Illustrative only. Invented figures shown to demonstrate the method, not a real measurement.
WHAT THIS IS BUILT ON
Where Does the Evidence Come From?
The findings draw on public research and platform documentation, not invented benchmarks. The core sources:
GEO: Generative Engine Optimization
Aggarwal et al., arXiv 2311.09735. Found that adding citations, quotations and statistics measurably raised how often generative engines surfaced a source. Underpins the page-level signals above.
Pew Research on AI Overviews
Pew Research Center, July 2025, on the reach of Google AI Overviews across US searches and the drop in click-through when an AI answer is shown. Grounds why citation share now matters, without importing a specific figure as our own claim.
Public AI-citation tracking
Independent trackers such as Profound, Peec AI and Ahrefs Brand Radar repeatedly find Wikipedia, Reddit and a short list of reference and review sites among the most-cited sources across engines. We cite the pattern, not a single vendor's headline number.
Google structured-data documentation
Google Search Central defines Organization, Article and FAQPage structured data and their required fields. Underpins the machine-readability signal above.
FREQUENTLY ASKED QUESTIONS
AI Search Citations, Answered
Who gets cited in AI answers?
A small, fairly consistent set of sources per query. Public citation trackers repeatedly find Wikipedia, Reddit and a short list of reference and review sites among the most-cited domains across ChatGPT, Perplexity and Google AI Overviews. Beyond those, engines tend to cite pages that already rank in classic search, that state a clear answer near the top, and that back their claims with citations, quotations and concrete figures. Being cited is not the same as ranking first, but the two overlap.
How do AI engines decide which sources to cite?
They combine several signals. For Google AI Overviews, the pool is largely pages that already rank for the query. Across engines, the public research points to the same levers: a clear, quotable answer near the top; citations, direct quotations and statistics on the page; structured data (Organization, Article and FAQPage schema) that a machine can parse; a named author with real credentials; and being a recognised entity with third-party mentions. The GEO study (arXiv 2311.09735) found that adding citations, quotations and statistics measurably raised how often generative engines surfaced a source.
How do I measure my own AI-citation share?
Build a frozen set of 20 to 50 buyer questions, run each one through ChatGPT, Perplexity and Google AI Overviews, and log whether your brand is cited plus which domains are cited instead. Citation share is the percentage of prompts where you are cited. The gap, the prompts where a competitor or a reference site is cited and you are not, is your work queue. Re-run the same prompt set monthly or quarterly to track the trend, since AI answers drift as models update. Tools such as Profound, Peec AI and Ahrefs Brand Radar automate parts of this.
Are the numbers on this page from a VerityAI study?
No. This page is a method plus the public evidence, not a published VerityAI benchmark. The cited findings come from named public sources: the GEO study (arXiv 2311.09735) on what raises generative-engine visibility, and Pew Research Center (July 2025) on the reach of Google AI Overviews and reduced click-through. The worked example is illustrative, invented to show how a citation-share measurement is built, not a real audit of any company.
PUT A NUMBER ON IT
Want Your AI-Citation Share Measured?
We'll build the prompt set, run it across the engines your buyers use, and hand you the citation-share number plus the gap list ranked by lift. No benchmark to wait for.
Measure My Citation ShareSTAY INFORMED
Get the Full Measurement Guide
Drop your email and we'll send you the full guide: the prompt-set template, the fields to log per answer, and the scorecard we use to track citation share every quarter.
Get Notified When Published
No spam. The measurement guide and template, once.