# What an answer says about you depends on whom it read.
> Which sources AI answers cite, what the available studies measured, what no one has documented, and how to audit the sources behind your own description.
- Page: https://mohammedteto.com/source-citation-authority/
- Language: en — French version: https://mohammedteto.com/fr/visibilite-ia/
- Author: Mohammed Teto (https://mohammedteto.com/#person)
- Published: 2026-10-07
- Updated: 2026-10-04
An AI answer is written from sources, and sometimes names them. This page separates what has been measured about those sources from what is assumed, and shows how to audit the ones behind your own description.
Source authority is the weight a source carries in what search engines and AI systems say about a subject. Citation authority is the narrower, observable part: which sources an answer names. Studies have measured which domains are cited, engine by engine. None of the sources cited on this site explains why a given source is chosen for a given answer. The work therefore starts from observation: recording which sources are cited about you, and classifying them.
## A source used and a source cited are not the same
An answer can draw on a source without naming it, and can name a source for a detail and not for its main point. A citation is what can be seen and counted. What a model learned in training usually leaves no citation.
This sets the limit of everything that follows. An audit of citations describes the visible part of the sources. It is the only part that can be recorded, and it is not the whole.
## What has been measured
| Finding | Source | What it does not show |
|---|---|---|
| Wikipedia accounts for 47.9% of the citations going to ChatGPT's ten most-cited domains, in an analysis of 680 million citations | [Status Labs, 2026](https://statuslabs.com/whitepapers/ai-and-the-future-of-reputation-management-2026) | The share of Wikipedia among all citations; the authors of the underlying analysis are not named in the white paper |
| Only 11% of domains appear in both ChatGPT and Perplexity answers to similar queries | Research reported in the same white paper | Which domains, and for which kinds of question |
| ChatGPT ran 2.1 web searches per question on average, in five million query fan-outs analyzed in April 2026 | [Peec AI, 2026](https://peec.ai/blog/patterns-we-see-in-chatgpt-query-fanouts) | Which results were then used |
| Adding citations, quotations and statistics to a source raised its visibility in generated answers by up to 40% in a controlled benchmark | [Aggarwal et al., KDD 2024](https://arxiv.org/abs/2311.09735) | Anything about sources other than the page being rewritten |
| A supporting link in AI Overviews or AI Mode must be a page indexed and eligible for a snippet in Google Search | [Google Search Central](https://developers.google.com/search/docs/appearance/ai-features) | How a link is chosen among eligible pages |
Three of these are reported by commercial providers and describe specific samples. They indicate a direction: Wikipedia is prominent among one engine's most-cited domains, and engines differ. They are not a universal measurement.
## What is not documented
The pages cited here do not say how an assistant ranks one source above another, whether it prefers a source because of its domain, its age or its content, or how much weight a model's training carries against what it retrieves. Statements about "trust scores" or a fixed hierarchy of sources go beyond what these sources support.
For Google's search systems, one stated principle exists: Google writes that of experience, expertise, authoritativeness and trustworthiness, "trust is most important" ([Google Search Central, creating helpful content](https://developers.google.com/search/docs/fundamentals/creating-helpful-content)). That is a description of what its systems aim to reward. It is not a citation rule.
## Kinds of source, by who controls them
| Kind | Examples | Who controls it | What it can establish |
|---|---|---|---|
| Owned | Site, official biography, company pages | The entity | The reference version of the facts |
| Managed | Professional and social profiles, directory listings | The entity, within the platform's format | Consistency with the reference version |
| Registers | Company registries, regulatory and professional registers | Public or professional bodies | Legal facts and their dates |
| Reference | Encyclopedias and open databases | Their editors, under their own rules | A summary that other systems may reuse |
| Editorial | Press, specialist publications, conference programs | Journalists and organizers | Independent description |
| Commercial | Paid placements, sponsored profiles, reposted press releases | Whoever pays | That a statement was made; little about whether it is confirmed |
| User-generated | Reviews, forums | Their authors | Experience reported by third parties |
A claim made by an owned source and confirmed by a register or an editorial source is corroborated. A claim made by owned and commercial sources only is repeated.
## How to audit your own sources
1. **Ask the questions that matter**, in each assistant separately, several times, and keep every answer with its cited sources and its date. The protocol is in [how AI visibility is tested](https://mohammedteto.com/ai-visibility/#testing).
2. **List the cited domains** and count how often each appears, per engine.
3. **Classify each one** with the table above.
4. **Check each source for accuracy.** A source that is cited and wrong usually matters more than one that is right and ignored.
5. **Note what is missing.** Which kinds of source are never cited about you? Which exist and are not attributed to you?
6. **Compare with what is cited for the others** named in the same answers. It indicates which kinds of source the engine cites in your field.
The result is a list of sources to correct, to align and to earn. How the readings are stored and compared over time is described in [AI search intelligence](https://mohammedteto.com/ai-search-intelligence/); for a brand in a category, cited domains are also the basis of the citation-based variant of [AI share of voice](https://mohammedteto.com/ai-share-of-voice/#methods).
## What can be done, by kind of source
**Owned and managed:** state each fact once, the same way everywhere, with markup that matches the page. See [structured data for entities](https://mohammedteto.com/structured-data-entity-schema/). Write so that a passage can be reused intact: see [answer engine optimization](https://mohammedteto.com/answer-engine-optimization/).
**Registers:** check them, and use the register's own procedure where an entry is wrong.
**Reference:** propose changes openly, under each source's rules.
**Editorial:** give journalists and organizers facts they can verify. Coverage is their decision.
**Commercial:** label it for what it is. Presenting a paid placement as independent coverage is excluded from this work.
**User-generated:** nothing beyond what each platform's rules allow.
No step in this list guarantees that a source will be cited.
## Where it sits in the work
This is the fourth layer of [The Authority Architecture System](https://mohammedteto.com/method/), [Evidence and Source Authority](https://mohammedteto.com/method/#evidence-and-source-authority). The situation in which only the entity describes itself has a diagnostic page, [weak independent sources](https://mohammedteto.com/problems/weak-source-authority/), and what counts as evidence is set out in [which evidence counts](https://mohammedteto.com/brand-authority/#evidence). Where source work sits relative to the three acronyms is shown in [SEO, AEO and GEO compared](https://mohammedteto.com/compare/seo-vs-aeo-vs-geo/). The wider subject is [AI visibility](https://mohammedteto.com/ai-visibility/); related analyses are collected in [Insights](https://mohammedteto.com/insights/).
## Frequently asked questions
### Which sources do AI assistants cite most?
It differs by engine. One analysis reported by Status Labs found that Wikipedia accounts for 47.9% of the citations going to ChatGPT's ten most-cited domains, and research in the same white paper found that only 11% of domains appear in both ChatGPT and Perplexity answers to similar queries. These figures describe their own samples ([Status Labs, 2026](https://statuslabs.com/whitepapers/ai-and-the-future-of-reputation-management-2026)).
### Why does an assistant cite one source and not another?
The sources cited on this site do not say. What can be observed is which sources are cited, how often, and in which engine. Explanations that go further are hypotheses.
### Is my own website a source?
Yes, and it is the reference version of your facts. An account supported only by its subject is thin, which is why the work also concerns sources you do not control.
### Can I pay to be cited?
A paid placement is a commercial source. It may be read like any other page, and nothing in the sources cited here shows that it counts as independent confirmation. Presenting one as editorial coverage is excluded from this work.
### How do I find out which sources are used about me?
Ask each assistant the same questions several times and keep the sources it cites, with the date. Counting and classifying them gives the audit described on this page.
---
Understand how search and AI currently interpret your authority. Request a confidential assessment of your visibility, entity signals, narrative consistency and source authority.
[Request a Private Assessment](https://mohammedteto.com/private-assessment/)