Third-Party Sources Are More Than LinkedIn, YouTube and Reddit
In many GEO discussions, the impression arises that AI visibility in B2B runs mainly through LinkedIn, YouTube and Reddit. Our measurement shows something different. On average, social platforms account for only about 16% of the third-party sources that AI systems draw their answers from.1 Of that, around 10% goes to YouTube, just under 4% to LinkedIn and barely 2% to Reddit. The remaining 84% lies elsewhere: on trade and company sites, in directories, in editorial media and in reference works.
On top of that comes the breadth. Across all brands, more than 1,100 different websites supply the content that AI answers are generated from.1 There is no shared, manageable canon behind it.
The distribution shows this most clearly: around 84% of these websites appear with only a single brand each.1 So which third-party sources count for a brand cannot be derived from a general list. It has to be determined for your own brand. This is exactly where GEO work begins: optimizing visibility for generative AI systems.
About the Data
More than 30 B2B brands from the DACH region (industry, IT services, software, FinTech, professional services, logistics, healthcare, public sector), evaluated over three months based on around 160,000 AI answers. Measured across ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Microsoft Copilot and Perplexity, in varying combinations depending on the brand.
There Is a Recognizable Structure Behind Third-Party Sources
Third-party sources are not distributed at random. Sorted by type of source, a pattern emerges that repeats across the brands.
Type of source
|
Anteil |
Company and advice sites (third-party)
|
55.7% |
| User-generated content (social media, forums, communities) |
19.2% |
Institutions and associations
|
8.2% |
Reference works
|
7.9% |
Editorial media
|
5.3% |
| Other |
3.8% |
The largest share goes to company and advice sites, at around 56%. What is meant here are third-party sites, so not your own brand website, but rather blogs, guides, industry portals and provider sites in the wider field. That such sites are drawn on frequently is hardly surprising: they often provide the most fitting answer. More interesting is how broadly the rest is distributed.
Beneath this distribution lies a second observation. Some sources appear across many brands at once, spanning industries. These include large reference works such as Wikipedia, along with individual cross-cutting trade portals and the sites of major providers. They form the base layer.
Alongside this stands the brand-specific part from the previous section: sources that appear with only one brand. Both belong together. The shared base layer is the foundation on which every brand in a category should be represented. The brand-specific part is what sets the visibility of individual brands apart.
Deep Content Pages Count More Than Homepages
Even more revealing than the type of source is the type of page that AI systems draw on.
| Page type |
Anteil |
Editorial content (blog posts, guides)
|
28.0% |
Directories, profiles and category pages
|
23.5% |
Third-party product and service pages
|
19.4% |
Comparisons and lists
|
16.4% |
Homepages (own and third-party)
|
4.8% |
Editorial content, directories and profiles, product and service pages, and comparisons carry the largest part. Homepages, both your own and third-party, together account for just under 5%.3
For your own website, this is a useful clarification. It is not primarily the homepage through which visibility in AI systems arises, but the individual, more in-depth content pages. For five of the brands measured, the homepage does not appear in the AI answers at all, and for none is it the most important page.3 So work on your own website pays off above all when it goes deep into individual topics.
The most frequently drawn-on page type also varies by brand: sometimes third-party editorial content, sometimes their product pages, sometimes profiles or comparisons.3 This again confirms that the picture draws on a broad mix. Anyone who only optimizes their own website covers just a small part of what AI systems build their answers from.
Not Every Mention Counts as a Source
A source can appear frequently and still rarely be drawn on as a citation. Presence and citation weight are two different things.
An example from the measurement: posts from communities such as Reddit are cited as a source, per instance, far more often than posts from business networks such as LinkedIn.4 Reddit thereby reaches a level otherwise seen with established reference works and trade portals. LinkedIn is present, but is used less often as an actual source.
For practice, this means: it is not enough to appear as often as possible. What matters is appearing in sources that AI systems also use as citations. Which ones those are depends, again, on the category and the AI system.
Every Brand Has Its Own Source Profile
How different the sources are shows most clearly in the social shares. Depending on the brand, the share ranges from under 4% to over 30%, averaging around 17%.5 Social is the strongest source for only about half the brands. What carries a large part of the picture for one brand barely matters for the next.
The AI systems differ too: Google AI Overviews rely more heavily on video and social profiles. ChatGPT draws more often on forums and review platforms, and also uses them as citations more often. Microsoft Copilot leans more towards professional networks and review sites.5 So the same brand has a somewhat different source profile depending on the system.
These differences are directions, not exact percentages; our measurement does not allow a reliable share per system. But the conclusion remains the same: anyone who wants to know what AI systems build the picture of their own brand from has to measure their own sources, not adopt those of another brand.
From Measurement to Optimization
If every brand has its own source profile, no universal approach helps. The first step is therefore always an assessment. Our measurement provides a concrete starting point for this: the source gap, that is, third-party sources in which a brand is missing even though it should appear there.
For a good two-thirds of the brands measured, this gap list is well filled. Only a single brand has practically no gap.6 Missing presence in relevant third-party sources is therefore the norm.
Where these gaps lie is similar across the brands. Neutrally reachable, that is, without competitor sites, they fall mainly to third-party company and advice sites, along with institutions, communities, trade media and reference works.6 Recurringly, brands are missing, for example, from large reference works, from relevant communities and on video platforms.
Before this turns into action comes the positioning. Three questions help you read your own source profile.
Three Questions About Your Own Source Profile
- Which third-party sources shape the picture of the category in your own market, and does your brand appear in them?
- In which sources is your brand present but rarely used as a citation?
- Which sources are shared by almost all providers in the category, and which are specific to your own brand?
At andweekly, this analysis is part of Generative Engine Optimization, part of our framework The Signal System™. From it, strategies for visibility in the relevant third-party sources can be derived. How to systematically build presence in independent sources is something we describe in detail in our article on Earned Visibility.
How We Measured
The figures come from our own evaluation and cover a period of around three months in spring and summer 2026. The basis is the most frequently drawn-on domains and URLs per brand, measured across a fixed set of questions typical for the respective category. Which AI systems are covered per brand varies.
Three limitations belong with this. The homepage share is measured across the most frequently drawn-on pages, not across every single URL. The differences between the AI systems are patterns, not exact shares. And the evaluation is not a representative study sample, but a snapshot from practice.
We measured with Peec AI, the tool we use for this purpose for the presence and citation of brands in AI systems. How to capture this visibility in metrics and measure it over time is a separate topic that we address separately: how to measure AI visibility.
Frequently Asked Questions
What do AI systems get their information about a B2B company from?
From many different third-party sources: from trade media, directories, comparison sites, communities and reference works, as well as from your own website. Third-party company and advice sites have the largest share here.
Does every mention count equally for AI systems?
No. A source can appear frequently and still rarely be used as a citation. Presence and citation weight differ, sometimes markedly depending on the type of source.
How does a B2B company's visibility arise in ChatGPT, Google AI Overviews and similar systems?
Through third-party sources mentioning a company and the AI system using them as a citation. Which sources those are depends on the category and the AI system. There is no universally valid list.
Do the AI systems differ in which sources they use?
Yes. Google AI Overviews rely more heavily on video and profiles, ChatGPT more often on forums and review platforms. Exact shares per system, however, cannot be derived from the measurement.
Is it enough to optimize your own website for AI visibility?
Your own website is important, above all in the form of in-depth topic pages. A large part of visibility, however, comes through third-party sources. The two together make up the picture that AI systems paint of a brand.
Conclusion
AI systems paint the picture of a B2B brand from many sources at once. Your own website is part of it, but a large part comes through third-party sources: through trade media, directories, communities and reference works. And which of them count differs from brand to brand.
That is why visibility in AI systems begins with an analysis of your own source profile, not with a single measure. At andweekly, this is part of our work on Generative Engine Optimization and of our framework The Signal System™. If you would like to follow the results of this measurement and further insights on visibility in the digital space, subscribe to the Perspectives newsletter.
Sources
andweekly evaluation (primary data) covering more than 30 B2B brands from the DACH region, period 16 April to 16 July 2026, around 160,000 AI answers, measured across six AI systems (ChatGPT, Google AI Overviews, Google AI Mode, Gemini, Microsoft Copilot, Perplexity). Tool: Peec AI. As of July 2026.
1 andweekly evaluation: share of social platforms, number of third-party websites, distribution across brands
2 andweekly evaluation: distribution by type of source and cross-cutting sources
3 andweekly evaluation: distribution by page type and homepage share
4 andweekly evaluation: frequency of citation as a source, per source
5 andweekly evaluation: social shares per brand and patterns per AI system
6 andweekly evaluation: analysis of source gaps
