Introduction
One of the core AEO pillars that I've written about relates to building trust. I believe:
- Even if you have properly structured data, that alone is not enough to get cited. Structure is an important multiplier, but it does not replace credibility.
- Analysts, trade media, and influencers have established themselves as trusted sources, so they should be more likely to be cited.
- It's not as easy to trick AI by gaming the system as was done with SEO.
So why do vendor sites and listicle sites dominate the citations in my studies of CAD and Healthcare tech? Why the absence of the 'trusted' sources? It turns out there is a rather nuanced set of mechanisms in place, many of which are yet to be fully uncovered.
The answers that are emerging are both disturbing and encouraging.
Disturbing because it seems that you can game the system (for now). And equally disturbing that consumers of AI are mistaking clever phrasing and crazy good data retrieval for real intelligence.
Encouraging, because the challenge with trust and authority is widely recognized. AI is improving by leaps and bounds. Every week, every month. It is hard to fathom that it will not get better at delivering better results. Encouraging because I personally know that journalists and thought leaders like Roopinder Tara, Michael Finocchiaro, Jim Brown and Monica Schnitger have the goods. They have decades of experience between them in the PLM and Tech space and content like theirs is gold that the AI tools will soon figure out how to mine (provided access stays available).
Before we unpack some of these thoughts, a quick note: This blog and most of the opinions I voice are specific to the B2B market. So don't expect a conversation here about Reddit or B2C citations. Let's dig in.
Rose Colored AEO
I started my AEO journey drinking heavy doses of the Kool-Aid: Credibility and authority are what earn citations. Structure and the technical heavy-lifting make good content more accessible to AI tools. This wasn't just me dreaming this up; this viewpoint was a result of researching the best thinking I could get my hands on. The foundational academic work pointed the same way. The Princeton-led GEO study (KDD 2024) that named the field found that citing authoritative sources and adding expert quotations lifted a page's visibility in AI answers by 30 to 40%. And it makes sense. The same marketing motions around brand, content, and messaging should be driving the trust signals for AI. As a veteran marketer, I view this movement as a renaissance in the value of a fully integrated marketing approach, after years of watching paid and SEO get treated as the whole game. Vindication that you can't simply game the system. (I made that case in my first post, AI rewards the marketing that was always worth doing.)
And AI tools like Claude present themselves as nearly unimpeachable, noble purveyors of trustworthy answers. But in a recent interaction, I asked Claude about launching a comparison site for tech products, a highly referenceable site built to draw citations as a live test of AEO theory.
Me: I'm thinking of launching a comparison site for tech products and I think I can build a highly referenceable site that draws citations, etc., that I can use for a live test of AEO theory.
Claude: AI systems do cite comparison content heavily, but they cite trusted comparison content. G2, Gartner, Capterra get cited because of authority and review volume, not just because they compare things. A new site comparing LIMS vendors starts with zero authority signals. The comparison format helps, but it doesn't substitute for the harder problem: why should an AI trust your categorization over the vendor's own site or an established analyst? Your answer needs to be either genuinely superior structure (plausible given your AEO skills) or proprietary signal (harder, e.g. actual usage data, original testing, expert curation).
Me: But, I ran empirical prompt tests across CAD and HealthTech. Both returned consistently high citations from vendor sites, listicle sites, and resellers who built comparisons and had minimal authority.
Claude: I was reasoning from how human trust signals work (Gartner, G2, review volume) and projecting that onto what AI systems actually do at retrieval time. Retrieval rewards whatever is structured, topically dense, and directly answers the query. A comparison page that cleanly lays out "X vs Y vs Z by feature" is easy to lift a citation from regardless of who published it. The model isn't doing due diligence on the publisher; it's finding the chunk that most usefully answers the question. Amateur sites that happen to be well-structured for the query win. That's exactly the gap your AEO skills exploit, and it means the citability moat is more available to a newcomer than I gave it credit for.
There is recognition and acknowledgement that these trust signals are not yet on solid ground. In How generative engines define and rank trustworthy content the author, Casey Nifong, states that "engines elevate a familiar set of qualities that have long defined reliable content, the same traits marketers and publishers have focused on for years." But it also says that generative engines are increasingly recognizing signals of first-hand expertise, content from subject-matter experts, original research, or lived experience, which I see as an opening for great content from less well-known sources to have a shot at getting cited.
What My Data Actually Showed
Besides bringing myself up to speed with published articles and studies, I wanted to get first-hand experience of how companies get mentioned and how AI tools are using sources for citations. To illustrate the citation behavior, I am sharing two examples of how this played out in the table below:
Two dramatically different segments showed a strong preference for vendor generated content. This is rather good for marketers and vendors but as a consumer, it does not carry the same level of authority that one would expect from researching a Gartner, IDC, etc.
The healthcare revenue prompts were, however, expected to surface vendors in the mentions. What I found surprising is that the source data came from those who had the most to gain, not the neutral, trusted sources.
So Which is It: Trusted Authority Signals or Brute Force?
It seems like a cop-out to say that both are right. Empirical evidence from my examples and what I am seeing from the AEO wizards who can 'get you mentioned in ChatGPT within weeks' are a bit disheartening. It sure seems like you can game the system. For now...
We all like using AI because it is fast (most of the time). But the price of the fast retrieval is a loss of precision. AI pulls candidate content by semantic match and structure. The reliability comes in afterwards when the answers are ranked before the magic appears on your screen. In a Visively article, Pedro Dias writes: "Some systems add a checking step after generation. The response is compared against cited sources, and citations that don't hold up are either replaced with better matches or removed. Claims without adequate support may be rewritten or cut entirely." Notice the hedge: some systems. It is not universal, and it is not reliable yet. If you want to geek out on how the systems rank and cite, this is a great read. Don't summarize it, read the whole article. It's a gem.
If trusted content exists, it wins. But here's the thing: what if trusted content does NOT exist? How many times has ChatGPT or Claude just said, "yeah, I don't have an answer, Google it"? None. That being the case, you end up with vendors, resellers, and listicles dominating the citations. Another great read is this 2026 paper from Heidelberg University called Whose Facts Win, which tells us: "we find that LLMs prefer institutionally corroborated information (e.g., government or newspaper sources) over information from people and social media. However, these source preferences can be reversed by simply repeating information from less credible sources."
And let me correct myself: if trusted content exists, it only wins if it is structured and hits the radar of these A.D.D.-like scanning tools in AI. This is rather an important point. There is good information that we already trust as humans. That's why we look to analysts or tech media for opinions, or academia and thought leaders. And it is not as if AI ignores that authority everywhere. Muck Rack's Generative Pulse study, which analyzed more than a million AI citations and was reported by Nieman Lab in July 2025, found journalistic sources made up more than 27% of all citations, rising to 49% on queries that implied any recency. Authority clearly asserts itself at scale. Which makes its near-absence in my B2B studies all the more telling. In the CAD and healthcare niches I looked at, the trusted names went missing and the vendors filled the vacuum. The Princeton GEO study points to why: a page sitting in fifth position saw its visibility in AI answers jump by 115% once it added proper citations and structure, while the top-ranked page actually lost ground. Structure lets a lesser-known page leapfrog a more authoritative one. And in my investigations of trusted authority sites, I've found a surprisingly high number that are not structured in a way that makes them easy to read for AI.
What are the Opportunities?
The Return of Trade Media
Trade media like EngTechnica and Develop3D already have established authority. But they're losing the retrieval contest on structure alone. If they were to address this leaky bucket, they could displace the amateur sites filling the vacuum. And they hold a second advantage in that they produce original reporting and testing, which is exactly the non-redundant, information gain signal retrieval awards. But it's more than structure.
There is a content gap that trade media has not traditionally filled: comparative data. This is a sticky topic. Trade media has been reluctant to fill this gap because of the dependency on advertising. Comparative articles, when done correctly, position one company as less capable than the other, and the loser is not likely to keep buying ads.
Comparison and Review Sites
These sites are built in such a way as to invite comparisons and answer the question why X is better than Y. But they are structured, maybe intentionally, to favor humans and keep the crawlers at bay. G2 and Capterra are JavaScript-heavy single-page applications. The reviews, the comparison grids, the star ratings, most of it renders client-side after the page loads, gets paginated, sits behind "load more" buttons, or is tucked into tabs and accordions. And the content is aggregate, not written in the Q&A style that makes it easily accessible for AI. In fact, it appears to me that they may not want to be cited.
Big Media
Big Media earns their revenue through subscriptions and advertising. The moment that AI can access their data, the revenues can and are already taking hits. In this article from Mark Stenberg, Stenberg reports that after decades of fighting to rank as high as possible in Google, a handful of influential publishers are now "laying the groundwork for what was once unthinkable," removing themselves from Google Search entirely.
Take this a step further. If the big guys team up and hide their data from Google (and subsequently AI) that could render AI results to be even worse than they are now. Lower tier publishers (like Trade Media) could still find their way into AI citations, but companies like Gartner will keep the magic quadrant and other pertinent research behind their walls. My take is that AI still can provide a valuable service of scraping data together, but it will never have the ultimate authority to be the trusted source in this scenario.
Marketers
As I've said in other articles, I see AI as not just a part of a marketing strategy, but significant enough to be THE marketing strategy. I believe that marketers have the opportunity to build strong authentic brands. They can both influence trusted authorities and exist as trust signals by carefully protecting and projecting their messages. At the same time, technical discipline is important to make sure that great content gets picked up by AI tools. And the investment in PR, analyst relations, all stand the chance of paying off by providing not only meaningful but structured information.
Conclusion
AI tools are evolving faster than I can write this blog. Today the system can be gamed, and it is worth understanding why, and why that will not last.
- The retrieval mechanisms of AI are going for speed, not perfection. They want to produce an answer.
- In the absence of trust signals, structured content that answers a question will be cited.
- As trusted sources get more structured, they will dominate citations.
- AI is not standing still. As I write this, it is getting better at reading trust signals, and it may soon make the system a lot harder to game.
Structure is what gets you cited today. Trust is what will keep you cited tomorrow. The gap between the two is where the opportunity lives right now, and it is closing.
