If you have a website, there’s a pretty decent chance that your traffic has been decreasing over the last few years. And because nobody likes graphs that are a downward ski slope to the right, there's also a chance that someone in your organization is feeling itchy about it. You may have even turned to the web, or your favorite LLM, to seek out best practices, advice, and hopefully find answers. At least that’s how my story started.
The next chapter ended up something like this: AI is the evil villain stealing your traffic because AI overviews harvest your content and serve it up to internet searchers as a new recipe that steals your ingredients and seldom mention your brand. Then LLMs take your content, mash it up with everyone else's, and pass it off as their own, almost like your neighbor who puts grocery store cookies on a fancy plate and lights a sugar cookie candle when company comes over. And there you are, the former blue ribbon baker not even invited to the neighborhood picnic this year.
So your ears start to perk up when people start talking about answer engine optimization (AEO), and you think maybe this is how you’ll get a seat at the table again. The advice, over and over again, is that to get “found” in this new era, the rules have changed. There’s a lot of talk about E-E-A-T, or the idea that experience, expertise, authoritativeness, and trustworthiness are the critical drivers of AEO. Suddenly you’re supposed to focus on Reddit and start doing tons of press releases. You need to do original research (clearly I believed that one, you’re reading it), and you absolutely need to have authoritative backlinks.
But here’s the thing, at least for me, the more of this advice I read, the more it didn’t exactly add up. So I dug a little deeper, and what I found was mostly circular. There were a lot of blog posts, videos, and “thought leader posts” that were all just referencing one another. There wasn’t a lot of data to back up most of the claims. So I went after it. I decided to see what ACTUALLY gets cited by AI.
TL;DR Six Key Findings
|
We started with the 485 top queries about B2B marketing that people are prompting in ChatGPT, Claude, Gemini, and Perplexity. That led to 35,755 AI citations across 5,091 domains. We then ran a whole slew of follow-on tests to look at that data and try to understand what separates pages and brands that get cited from those that don’t.
Frankly, the results surprised me. ~78% of citations are companies’ own content; the “social proof” everyone optimizes for is ~9%.
To put a finer point on it: what’s getting cited on the whole is self-promotional content. It’s companies talking about themselves, or at a minimum creating content on their owned properties.
No — and not by a little. Each engine pulls from a strikingly different mix of sources. ChatGPT draws about 84% of its citations from vendors' own content and barely touches anything else. Perplexity leans roughly four times as hard on third-party platforms like Reddit and YouTube. Claude and Gemini land somewhere in between. Same question, four different answers — sweet summer child, did you think it'd be that simple? Here's the mix, engine by engine:
By most measures, ChatGPT is the most-used AI assistant by a wide margin — roughly half of consumer usage (about 54% of worldwide web visits in May 2026, or 46% of unique app users) — with Google Gemini a clear second at roughly a quarter, and Claude and Perplexity in the low single digits to low teens. One useful way to read this: optimizing only for ChatGPT reaches under half the assistant audience; adding Gemini lifts coverage to about three-quarters, and adding Claude pushes it into the mid-80s. (As with the search figures earlier, these scatter by source and by whether they count web visits or app users, so treat them as directional — but the rank order is consistent.)
That’s only the beginning of a more interesting story though. Of the 35,755 citations in this study, Gemini accounted for more than half, while ChatGPT accounted for less than 4%. So the most popular engine cites its sources the least. While the purpose of this study was to understand citations, I have to admit that at least for me, I’m rethinking how I’ll use each of these tools moving forward. Knowing how much they cite and where those citations come from is revelatory.
Surprisingly, no.
5,091 domains were cited across our 35,755 citations. The number 1 domain on this list? HubSpot. While many in marketing wouldn't be surprised by this at all, this next number is pretty shocking: it earned just 2% of those citations. So what does that mean? Well, from where I sit, AEO is anybody’s game. There is a very long tail right now, and small agencies with fewer employees than I have fingers showed up as the top citations for key prompts.
Ultimately, AI cites widely and shallowly across a large field of self-published content. You don’t need to beat HubSpot, you need to show up with the helpful, current, relevant answers when your ideal prospect asks a given question. If you really understand your audience, that’s a very winnable game.
Not in the way that the E-E-A-T advice seems to promise, and this was the hardest finding for me. I really believed that authority mattered because I think humans and human authorship matter…and I still believe that. I just think we have to remember that humans matter more to humans than we do to LLMs.
We looked at the data several different ways, and we simply could not draw any correlation between authority signals and AI citations. We tested whether cited pages showed more of what "demonstrate expertise" advice tells you to add, versus equally-ranked pages that were not cited: named authors, links to primary sources, named-person quotes, first-hand experience, original data. Across three question types, cited pages were no more likely to show any of these.
Even original research (something near and dear to this researcher's heart) accounted for only 3.5% of all cited pages. It concentrates in independent publications (15%) more than vendors (6%), but as a share of what gets cited, it is small — and cited pages were no more likely to be research than passed-over ones.
Because it reads ubiquity, not authority. An LLM isn't a well-read expert that recognizes a respected brand and rewards it. It surfaces the name that appears most often near your buyer's words across everything it ingested — statistical co-occurrence, not judgment. That rewards publishing volume and web presence, not the trust and esteem that build a brand in a human's mind. A marketer reaches for the tool they respect; the machine reaches for the name that saturates its training data.
Now the honest caveat: we can't yet cleanly prove causation, and I'd genuinely like to — if you've got a rigorous design for that study, let's talk. But we know what it cites today, and for pages competing in search, AI citation is close to a restatement of organic ranking. This isn't an exotic new discipline; the work that earns organic visibility earns the citation.
Just like traditional SEO, on the other side of an AI search is a human asking a question. The difference, of course, is that with answer engines, people are asking a series of related questions in the context of a conversation. So it’s not about keywords and phrases in isolation; instead, it’s about the problems they’d go to an answer engine to help them solve. While you could argue that traditional search was ultimately about the same thing, the tool itself is designed to handle one query at a time, without context, and to return a list of answers (and now an AI overview of that list). So while traditional search requires focusing on a single keyword, phrase, or set of keywords and phrases, AEO requires understanding the conversation that person is likely to have with their chatbot and focusing on the string of questions that person is likely to ask in that context.
That changes the work. When you’re not chasing a keyword, but rather covering an entire conversation, it’s as simple and as complex as this: be the clearest, most useful answer to those questions. Here's the part that should feel like relief. The way you get there isn't complicated, and it isn't new. Two things tracked with getting cited in our data, and both are within reach of a small team.
First, rank in traditional SEO for the question. AI hasn’t replaced traditional search; in fact, traditional search has kept climbing since the onset of AI answer engines. Google now processes over 5 trillion searches a year, up from 2 trillion in 2016. According to our research, AI effectively sits atop traditional search, or at a minimum runs in lockstep alongside it. This means the SEO fundamentals you already know are the foundation of AEO as well.
Second, keep it current. Freshness stood out as a distinct signal, independent of traditional search rank. A page that has been updated and maintained will beat one left to age. And be the source, not a mention of the source. Publish the answer yourself, clearly, on a page you own — because most of what AI cites is companies' own content, not the third-party "proof" everyone chases.
None of this amounts to something that can be “hacked” or any kind of authority score. At least today, there is no equivalent to a domain authority score when it comes to AEO.
Today, owned content is what is being cited by AI, and that’s a reason to breathe a bit easier rather than panic. Traditional search still makes up the bulk of all searches, and AI engines are just another way people get answers. Any strategy you create for AEO should be built on an SEO foundation because you’re actually solving for the human on the other side of the search, and what wins organically also wins in AEO. If you focus on the people you serve, answer the questions they’re actually asking, and provide answers that are truly helpful and well-structured, you’ll be found. If you update that content regularly, you’ll continue to be found. It’s exactly that simple and complex all at the same time.
A caveat, because we're allergic to hype: this study measured what gets cited. Whether that citation becomes a click or an actual recommendation is the next question — and the honest edge of what we can prove today. The full study is where we show our work.