LLMs Source Industry Sites 86% of the Time, Generic Sites like Reddit Just 16% [Data Study]

If you’ve followed the conversation around GEO lately, you’ve probably heard the same advice on repeat: ChatGPT cites domains like Reddit, Wikipedia, and Forbes the most, so get your company mentioned on those. 

Infographics or graphs like these often accompany that advice, showing the top cited domains by “AI.”

The logic is that if those sites appear as top sources, then showing up there should help your company appear in LLM responses, right?

Wrong. This thinking is built on a flawed assumption: that there’s a universal list of most-cited sites that applies to all businesses.

There isn’t.

When someone asks ChatGPT for something like, “What are the best dispatch software options for trucking companies?” or “What are some good alternatives to Asana for project management?”—our data below shows that ChatGPT doesn’t pull from a static list of popular domains like Reddit, Wikipedia, or Forbes. It heavily favors citing industry-specific domains that make sense for that exact question.

That’s a key distinction most of those top-cited domain charts miss. They’re adding up citations across thousands of random keywords or prompts. For example, the fine print on the right-hand chart above says, “Based on 150 thousand citations from 5,000 randomly selected keywords from Semrush database.” 

Who knows what those keywords are, but it stands to reason that a large proportion are general-purpose consumer queries (because statistically, those are the most common searches). So for general-purpose consumer queries like “best washing machine” or “restaurants in Chicago,” it makes sense that Reddit, YouTube, or Yelp would show up often.

But if you’re a B2B company, or as we’ll show below, even a niche B2C business, that data doesn’t apply to your brand. For the product or industry-specific prompts that actually matter for your brand, the sources cited by LLMs are almost always industry-specific sites.

To understand this quantitatively, we analyzed over a hundred prompts across five industries our clients operate in—from trucking software to project management—and logged every citation the model used. We did this using our AI visibility tool Traqer.ai. 

Here’s what we found:

• 86% of all references came from industry-specific domains—often from the vendors themselves on their own blogs.
• Only 14% came from general sites like Reddit, Wikipedia, or Forbes.
• LLMs cited our clients’ content in 88% of the topics we analyzed, often referencing blog posts we created for them.

That last stat is worth paying close attention to. As per our Prioritized GEO strategy, when your own site ranks on Google for product-intent keywords, you’re positioning your company for LLM exposure because LLMs search the web to supplement their answers, especially for product-related topics. 

Our analysis includes results from ChatGPT, Perplexity, and Google AIO.

Below, we break down the data to show what LLMs are actually citing, and why that makes owned, SEO-driven content such a reliable foundation for GEO strategy. We’re sharing it because we think this approach can help any company, even those that don’t work with us.

If you’re interested in working with us to build a content strategy that gets your company cited by LLMs, learn more about our services here. Or if you want help tracking the topics and prompts that matter for your or your clients’ brands—and see the sources most often cited for them—you can check out our tool Traqer. 

LLMs favor industry content for product-specific prompts—and here’s the data to prove it

We analyzed the sources in GEO prompts for these five Grow & Convert clients:

• Toro TMS: Transport management software for bulk hauling fleets
• Wrike: Project and workflow management platform
• Axial: Online marketplace for small business M&A and investment banking
• Climb Hire: Nonprofit career training program for adults entering tech
• ServiceTitan: Field service management software for trades and home services

The group includes four B2B companies and one niche B2C, Climb Hire. 

Each client is in a very different industry, which allowed us to see how LLMs cite sources across different verticals. The analysis was based on real product-intent queries like “bulk hauling software,” “alternatives to Asana,” and “HVAC business management software.”

Traqer.ai is topic-based, meaning it tracks a brand’s AI visibility across a topic area, not a single prompt. For each topic, it looks at visibility on various prompts that approach that topic from different angles. 

For example, here’s the topic “California job training programs” for our client Climb Hire:

This is the data from the topic “California job training programs”:

To understand how LLMs choose their sources, we also looked at the specific domains linked by each prompt. Each prompt was categorized by funnel stage:

• Bottom-of-the-Funnel (BOTF): Prompts explicitly asking for recommendations on products or a list of options.
  • Example (Wrike): “What are the top project management tools for marketers?”
    
• Mid-Funnel: Prompts about products but phrased in a way that invites evaluation or education rather than direct product recommendations.
  • Example (Toro TMS): “How can I improve efficiency in my bulk hauling operations using software?”
    
• Top-of-Funnel (TOF): Prompts that don’t reference products at all, focusing on general information or advice.
  • Example (Climb Hire): “Are there any online jobs for moms who want to work from home part-time?”

Across all five clients, we analyzed 120 total prompts:

• 72 Bottom-of-Funnel (BOTF) (~60%)
• 28 Mid-Funnel (~23%)
• 20 Top-of-Funnel (TOF) (~17%)

Once the prompts were categorized, we analyzed a total of 2,440 domains cited by the LLMs and classified each one into one of three source types:

• Industry: Direct vendors or companies that sell the product or service being discussed.
  • Example: Wrike.com in a project management software query.
    
• Industry-Adjacent: Publications, blogs, or platforms that operate within the same sector but don’t sell the product directly.
  • Example: Transport Topics (a news publication in the trucking and freight transportation industry) in a trucking software query.
    
• General: Domains with little or no direct connection to the industry.
  • Examples: Mainstream outlets (Forbes, Wikipedia), community platforms (Reddit, Quora, YouTube), and occasionally completely off-topic sources that LLMs surface in their responses.

This structure allowed us to compare when and how often LLMs cited general sites versus industry-specific ones. That is, do LLMs cite general domains more often for top-of-funnel queries that don’t ask for product recommendations? 

The Results: 86% of citations come from industry sources

The pattern was consistent across every client and industry.

Out of all the LLM citations:

• ~59% came from direct industry sources (1,444).
• ~26% came from industry-adjacent domains (649).
• ~14% came from general domains (347).

Put differently, about 86% of all citations came from within the industry ecosystem, and only 14% came from general or unrelated sources.

On top of that, LLMs cited our clients’ content in 88% of the topics we analyzed, showing that for nearly every product-related query, their pages were among the sources referenced.

In other words, for product-specific queries, ChatGPT and other LLMs tend to favor companies and vendors in the space, especially those already ranking for product-intent keywords on Google.

Here are three examples directly from our data:

1. Toro TMS

Across all the Toro TMS topics we analyzed—from “trucking management software” to “chemical transport software”—the top-cited domains were all from within the trucking and logistics space. 

About 95% were direct vendors such as AMCS Group, Transvirtual, and Toro TMS itself, and the remaining 5% came from an industry-adjacent publication (Transport Topics).

2. Wrike

For Wrike’s topics, 97% of the most-cited domains came from within the project management industry. About a quarter of those mentions referenced Wrike directly, alongside other tools like Asana, Monday, and Atlassian. 

The only general site appearing in the top results was Reddit, making up roughly 3% of total citations.

3. ServiceTitan

In ServiceTitan’s topics, the top-cited domains were again concentrated in the field service and trades ecosystem. 

Roughly 90% were direct vendors like ServiceTitan, Jobber, and BuildOps, while the rest came from industry-adjacent platforms like Zendesk, Kinettix, and FreightWaves.

These examples aren’t outliers. Across all five companies we analyzed, the pattern was the same: LLMs overwhelmingly drew from vendor content and industry-specific sources rather than more general domains.

The 14%: When LLMs cite general sites—and why it matters

Out of 2,440 total citations in our dataset, only 14% came from general sites. General sources show up slightly more often in TOF prompts (~18%) than in mid-funnel (~11%) or BOTF (~14%), but the differences are minor.

What the data shows, consistently, is that even at the top of the funnel, LLMs still lean heavily on industry sources.

To understand why that might be happening, we expanded our Traqer analysis to include a very different type of brand: Peloton. 

Unlike our clients—four B2B brands and one B2C brand operating in niche categories—Peloton sits in a broad, mass-market B2C category with a much wider audience.

We think it’s because of this that the Peloton data shows a very different pattern. 

To compare it cleanly with our clients, we analyzed four keywords across 20 prompts, with LLMs citing a total of 542 domains in those responses.

Across the Peloton prompts:

• General sites (Reddit, PCMag, CNET, etc.) made up just under 60% of citations—324 out of 544.
• Industry-adjacent sites (cycling, equipment, and workout-gear review sites) accounted for ~34%.
• Industry-specific sources made up only ~6%.

Here’s an example of some of the general sites being cited for the topic “best exercise bike for home”:

Notably, almost all Peloton prompts were product-focused: 15 BOTF, 5 mid-funnel, and no TOF queries at all. The lack of TOF prompts makes the contrast even sharper. Even without those TOF queries—where general sites usually show up the most—LLMs still cited general domains far more often than industry or industry-adjacent ones, and far more than anything we saw in the niche categories of our client dataset.

At first glance, you might assume that B2C or consumer-facing prompts lead LLMs to cite more general sites like Reddit, Wikipedia, or YouTube. But that explanation doesn’t hold up with our client, Climb Hire. Yes, it’s B2C, but it sits in a narrow, expertise-driven niche focused on career training, certifications, and early tech careers. Across all its prompts, general sites made up only about 15% of citations, which is very close to what we see in our B2B clients.

This suggests the biggest driver of general-site LLM citations isn’t whether a company is B2B or B2C, or whether the prompts are bottom or top-of-funnel—it’s how niche the vertical is.

Why? Because we think niche companies operate in niche content ecosystems, while mass-market companies operate in mass-market ones.

Meaning, in niche verticals like trucking software or small-business M&A, most of the useful information online comes from industry vendors and vertical-specific publications. General sites rarely cover these topics because the audiences are small and highly specialized. The New York Times or Forbes aren’t exactly publishing in-depth pieces on trucking management software—or, in Climb Hire’s case, IT training programs in California.

And LLMs seem to recognize that. More accurately, when they search the web, Google surfaces the niche sites that do cover trucking software or IT training programs in California. 

Peloton lives in the opposite environment. Queries like “best exercise bike for home” or “virtual spinning classes online” sit inside a massive consumer market dominated by large generalist publishers and platforms—sites like Wirecutter, PCMag, CNET, YouTube, and Reddit. And since LLMs search the web to ground their responses (especially for product-centric prompts), these general sites show up in Google and are cited by LLMs: 

Said concisely, it’s the niche-ness of the category that determines which domains LLMs cite—not funnel stage or B2B/B2C labels.

Off-site brand mentions in GEO strategy: Start with the right topics, then focus on the right sites

What this data shows is that if you’re doing off-site marketing as part of your GEO strategy—what we call Tier 2 of our Prioritized GEO Pyramid below—you should focus on getting brand mentions on the domains that are actually cited for the prompts and topics you care about. 

In other words, don’t waste time trying to label yourself as “niche” or “general,” or B2B vs. B2C, to guess where you should earn mentions. Start with the prompts and topics that matter, then look at which sites LLMs list as sources for those prompts. It’s simple and evidence-based. You’re not following a vague rule about what LLMs supposedly prefer—you’re looking directly at the domains showing up for the prompts that matter to you and aiming to get mentioned there. 

And to be clear, any mention in a reputable publication helps your overall brand marketing. But most marketing teams are working with limited time and budget. So if your goal is to specifically grow AI search visibility for the prompts that matter, it’s best to start with the prompts that are important to your brand and work backwards.

First, our AI visibility tool, Traqer, makes this easy by showing you which domains are most frequently cited for a given topic: 

That way, you know exactly where to build mentions that actually influence LLM responses in your industry.

Second, we offer a “citation outreach” service to our agency clients. We reach out to the sites most frequently cited for the prompts that matter and work to get our clients mentioned in their articles. You can learn more about working with us here.

The power of owned content for AI search visibility

Finally, one more key observation from this study: LLMs cited our clients’ own content in 88% of the product-specific topics we analyzed across the five brands.

Here are a few examples (among many more):

This is why the foundation of our Prioritized GEO pyramid is owned content. 

As we explain in our Prioritized GEO article, owned content boosts LLM visibility through the exact same mechanism as Off-site Mentions (Tier 2): by first giving you brand visibility in traditional Google Search. 

LLMs “ground” their responses in regular web results because they know their training data alone isn’t enough to guarantee accurate or up-to-date answers, especially for questions about the best products in a particular category or how to solve a specific problem.

A very effective alternative to chasing mentions on other sites is simply ranking your own content on Google for the keywords where you want AI visibility. As the screenshots above show, we’re ranking our clients’ domains for traditional SEO queries that LLMs then discover when they search the web. 

As we mention in our Prioritized GEO article, we put this as Tier 1—above off-site mentions—for several reasons: 

• Content on your site gives you the room and flexibility to explain the details and nuances of who your products help, in what scenarios, and how. ChatGPT in particular knows a lot of details about the user and personalizes its recommendations. So the more specific and complete your on-site content is, the more likely it is to be cited and recommended in the right moments.
  
• Content on your site kills two birds with one stone. It drives traditional SEO results while simultaneously increasing your visibility in LLM responses. The same content that ranks on Google is what LLMs are pulling from when they generate product recommendations.
  
• Content on your site comes with less algorithm-update risk. One algorithm update can nix an entire third-party site (like what happened to Reddit on ChatGPT), but LLMs will always continue to pull from the broader web, including individual brand sites like yours.

Want to explore how we can make this strategy work for your brand? Reach out to us. 

Want more insights like this as soon as we publish them? Subscribe to our newsletter.