What 33,698 AI Citations Reveal About Why Brands Get Recommended
A study of 33,698 AI citation events explains why specific brands earn recommendations over competitors based on third-party source authority and information retrieval patterns.
By Aayan · Thu Aug 13 2026 · 4 min read
I've now analysed more than 50 brands through In-Answer AI, across more than 6,800+ AI responses. In the newer citation-enabled analyses, I've also recorded 33,600+ AI citation events across 16 brand analyses, covering 11,412 unique cited URLs across 3,932 domains.
I'm not collecting those citations just to count links. I'm trying to understand why one brand keeps getting recommended while another doesn't, and what information keeps sitting underneath those answers.
AI citations make recommendations easier to investigate
A recommendation tells you what happened. The citations can help show what the AI used while getting there.
If a competitor keeps appearing ahead of a brand, I want to know which sources keep showing up around those answers. Is the AI relying on the competitor's own website, independent publications, comparison pages or reviews? Is the brand I'm analysing missing from places that repeatedly surface around the category?
In-Answer AI looks at those patterns rather than handing over a list of links. That can show which publications keep appearing and where competitors have stronger coverage.
Retrieval and citation are different
A source being found does not mean it will be cited. In the current In-Answer AI dataset, 20,084 source selections were retrieved but not cited in the final answer.
That is why I separate retrieval from citation. I want to know what was found, what was actually used and what happened in the answer the user finally saw.
Your own website is only part of the picture
A strong website still matters, but recommendation questions can pull from a wider information environment. An industry publication might matter for a category question. A review platform might matter for trust.
That means two companies with equally good websites can still have very different AI visibility. One may have much stronger third-party coverage around the questions that matter commercially.
This is where competitor citations become especially useful. If the same publications keep appearing around a competitor, I can investigate whether that competitor has stronger coverage there and whether the brand I'm analysing is missing from an information environment that AI systems repeatedly use.
That does not prove one publication caused the recommendation. It gives me evidence about what keeps appearing around the answers where a competitor is winning.
AI citations are not just backlinks with a new name
I don't treat citation volume as a ranking score. A source can support one factual statement without being the reason a brand was recommended. It can matter for one question and disappear for another.
Zikang Liu and Peilan Xu's ACL 2026 research on generative citation visibility found that document-level content properties mattered more than isolated wording changes in their experiments. Research by Nelson F. Liu, Tianyi Zhang and Percy Liang also showed why citations should be examined rather than automatically treated as proof that every surrounding claim is supported.
What I'm researching inside In-Answer AI
I don't want In-Answer AI to stop at telling a company that its visibility is high or low. I look at which brands appear, which competitors take the recommendation, how strongly companies are positioned, whether the information is accurate and what source patterns keep appearing around those outcomes.
The citation layer can then show who has influence around a category, which publications repeatedly appear, where competitors have stronger coverage and where a brand may be missing.
Across 33,698 citation events, I care much less about one source appearing once than I do about what keeps happening across the answers. Knowing that AI recommended your competitor is useful. Understanding what sits underneath that recommendation is more useful.
About the In-Answer AI citation research
The figures in this article come from first-party research conducted through In-Answer AI. The wider dataset covers more than 50 unique brands and more than 6,800 AI responses. The newer citation-specific dataset covers 16 brand analyses, 33,698 citation events, 11,412 unique cited URLs and 3,932 unique cited domains as of 13 August 2026.
A citation event is an instance where a source is recorded as cited within an analysed AI answer. The same URL can contribute multiple citation events across different answers.
Research referenced
Zikang Liu and Peilan Xu (2026), Think Before Writing: Feature-Level Multi-Objective Optimization for Generative Citation Visibility. ACL Anthology.
Nelson F. Liu, Tianyi Zhang and Percy Liang (2023), Evaluating Verifiability in Generative Search Engines. ACL Anthology.
Frequently asked questions
What is an AI citation?
An AI citation is a source surfaced within an AI-generated answer. In this research, each recorded use of a cited source is counted as a citation event.
Why are AI citations useful for brand visibility?
In-Answer AI uses citation data to show which sources keep shaping answers, which publications repeatedly appear, where competitors have stronger coverage and where a brand is missing.
Does having more AI citations mean a brand will be recommended more often?
No. Citation volume alone does not prove stronger visibility or cause a recommendation. Context matters.
What is the difference between a retrieved source and a cited source?
A retrieved source was found during information gathering. A cited source was surfaced in the final answer. In the current dataset, 20,084 source selections were retrieved but not cited.
What does In-Answer AI analyse beyond citations?
I also look at brand visibility, competitor appearances, recommendation strength, positioning, accuracy and the patterns that repeat across relevant AI questions.