---
title: "Do 82% of AI Citations Come From Third-Party Sites?"
description: "Aleyda Solis measured the top 10 cited source domains per brand across 45 brands. That is a share of source slots, not a share of citations. Owned domains hold about 10% to 11% of slots and carry far more of the mention weight."
published: 2026-08-04
author: "Ben Tabas, Founder"
publisher: "Zion Labs"
canonical: https://zionlabs.io/research/ai-citations-82-percent-third-party
source: Zion Labs research
---

# Do 82% of AI Citations Come From Third-Party Sites?

**The figure as commonly quoted:** 82% of AI citations come from third-party sites, therefore your own content is a minor input into AI answers.

The measurement is careful and the author's own framing is more careful still. The metric is not the one the quoted sentence describes.

## What the source actually measured

Aleyda Solis published a cross-vertical analysis on [August 2, 2026](https://www.aleydasolis.com/en/ai-search/ai-search-citations/).

| | |
|---|---|
| Publisher | Aleyda Solis |
| Date | August 2, 2026 |
| Sample | 15 leading brands in each of SaaS, ecommerce and finance, 45 brands in total |
| Data source | Semrush Enterprise |
| Surfaces | Google AI Mode, Gemini, ChatGPT |

The owned versus external split she reports: SaaS 17.7% owned and 82.3% external, ecommerce 30.4% and 69.6%, finance 20.6% and 79.4%.

The method is the part that matters. She took the **top 10 cited source domains per brand**, roughly 150 top source slots per vertical, and hand-classified each domain. So the metric counts distinct domains occupying slots. It does not count citations.

## Where the quoted version and the source diverge

One owned domain cited five hundred times occupies one slot. Fifty forum threads cited once each occupy the other nine. Under a slot count those two situations look like a 10% owned share either way, and under a citation count they look nothing alike.

**A slot share is not a citation share.** The 69.6% to 82.3% figure measures source diversity. Anyone quoting it as "82% of AI citations are third-party" has changed the metric, and changed it in the direction that makes owned content look weaker than the same dataset says it is.

Solis says so herself, in the passage that almost never travels with the number. Brand-owned domains are mention-dense but slot-scarce. They hold only about **10% to 11% of the top source slots** across all three verticals, yet contribute a much larger share of the mention weight: **20.6% in finance from about 10% of slots**, with similar disproportion in SaaS and ecommerce. A domain type that produces double its slot share in mentions is a concentrated asset, not a marginal one.

There is a second reason the slot metric behaves this way, and it is structural rather than a flaw. Capping at the top 10 domains per brand puts a hard ceiling on how much of the picture any single domain can occupy. A brand can hold at most one of its own ten slots, so the maximum possible owned share under this method is 10%, and the measured 10% to 11% means owned domains are effectively at that ceiling almost everywhere. Read that way the finding is close to the opposite of how it circulates.

Her own framing is more careful than the industry's use of it. The title calls AI search a third-party citation problem *with an on-page corroboration base*, and the second half is load-bearing. Her stated caveats are among the best in this literature: domain-level citation data cannot establish sponsorship or promoted distribution, the data represents source-environment associations rather than campaign attribution, and you cannot responsibly guarantee or buy an organic AI citation.

## One unresolved thing

In fairness to anyone using her data, there is a loose end. A separate post of hers on SaaS brands reports external share of citation *weight* at 83.6% to 93.4% depending on subvertical, against 82.3% for SaaS in the cross-vertical post. Those are different denominators, and the definition of citation weight is not published in either post.

Do not quote an external-share figure from her work without naming which post it came from. Two numbers from the same author measuring two different things is not a contradiction, but treating them as interchangeable produces one.

## The honest version of the claim

Say "share of top cited source domains" rather than "share of citations", and carry the 10% to 11% slot share against the 20.6% mention weight in the same breath.

It changes the strategic reading, and changes it in a way that matters for where budget goes. The quoted version argues that owned content is a rounding error. The measured version says owned content is a concentrated one: fewer positions, far more weight per position. Both of those can be true at once, and only one of them is an argument for abandoning your own site.

---

This page is part of [Where the AI Search Statistics You Keep Quoting Actually Came From](/research/geo-statistics-source-audit), an audit of seven of the most repeated figures in AI search optimization, with the checks we run on each one.

## Frequently asked questions

### Do 82% of AI citations come from third-party sites?

Not as measured. Aleyda Solis's August 2, 2026 cross-vertical analysis reports 82.3% external for SaaS, but the metric counts distinct domains occupying the top 10 cited source slots per brand, not citations. Brand-owned domains hold about 10% to 11% of those slots while carrying a much larger share of mention weight, 20.6% in finance.

### What is the difference between slot share and citation share?

A slot share counts how many of the top cited source positions a domain type occupies. One owned domain cited five hundred times occupies one slot; fifty forum threads cited once each occupy the other nine. Slot share measures source diversity. Citation share measures volume. They can point in opposite directions.