A common view in AEO/GEO right now is that third-party content does most of the work. Get on the listicles, get cited on Reddit, earn coverage on the pages AI engines already trust, and the mentions will duly follow. The figure often attached to this view is that around 85% of what AI engines cite is third-party content, so third-party content must be what gets a brand recommended.
We wanted to test that theory, so we ran a study across 15,000 AI answers for a B2B software client. The full write-up is published on the Grow and Convert blog, and this post is a condensed version of what we found and what it means for how you prioritize AI visibility work.
What we measured
The study we did covered 112 prompts across 24 topics, with 52 weekly runs for each prompt, spanning ChatGPT, Gemini, Google AI Overviews, AI Mode, and Perplexity. For every answer, we logged which articles the engine cited and whether the client was mentioned by name. (Note: Brand mentions and citations are separate signals, and keeping them apart was important for this exercise).
Third-party pages account for most citations
The 85% figure did show up in our data. Across the answers we looked at, the majority of the citations were third-party pages. So the engines are indeed reading mostly third-party content. That part of the common view is accurate. The open question, however, is whether reading third-party content is the same as being mentioned because of it. Our data suggests these might be two different things.
Owned pages are what get the client mentioned
Of the 7,779 answers that mentioned the client, 87% had the client’s own site among the sources. When the client’s own site was cited, mention rates were high:
90% on Google AI Overviews
89% on Gemini
89% on AI Mode
73% on ChatGPT
51% on Perplexity
When a third-party page that includes the client was cited instead, mention rates were much lower:
25% on Google AI Overviews
25% on Gemini
5% on AI Mode
6% on ChatGPT
28% on Perplexity
Importantly, the client is absent from 66 of the 70 pages the engines cite most in its category, yet it still gets mentioned in 51% of all answers, because its own pages get pulled in. Put simply: third-party presence in AI citations turned out not to be a prerequisite for getting mentioned in this case.
The pattern still appears when you look topic by topic
Individual topics show the same relationship. On one topic where the client’s own site was cited in 88% of runs, the client was mentioned in 79% of runs. On another, where its site was cited in only 44% of runs, it was mentioned in 25%. Mention rates moved with how often the client’s own pages were cited, not with third-party coverage. The brands that get mentioned are the ones with a page that matches the prompt closely enough for the engine to pull it in as a source.
What this means for prioritizing AI visibility strategy
We still think third-party mentions matter. The point here is the order of prioritization.
Produce content on your own site first that closely matches the topic areas you are trying to own, confirm it is being retrieved, then pursue third-party mentions on top of that. This follows the Grow and Convert GEO priorities pyramid, which puts owned content first, then off-site mentions, with on-site technical tactics last. A good example is the company, Redline Capital, which reached high AI visibility on owned content alone, which is how we think about building a brand’s visibility from scratch.
For agencies and consultants managing several client brands, the practical takeaway is to check, per client and per topic, whether the client’s own pages are the ones getting cited. If those pages aren’t being retrieved, more outreach is unlikely to fix a mention problem on its own. Traqer breaks visibility out per LLM rather than as a single blended score and separates brand mentions from citations, so you can see which of the two a client has on each topic and where owned content is doing the work.
A caveat on the data
Results like these can move with the company, the industry, the type of business, and how competitive the category is, so we will keep publishing studies rather than treating one dataset as settled. Each engine also behaves a little differently, which is why we look at them separately rather than averaging them together.
You can read the full study, including the complete methodology, on the Grow and Convert blog.
Track your clients’ brand mentions and citations with Traqer