The big question: Does video impact GEO?
If you’re working in marketing today, one of the big questions is what role GEO or generative engine optimization should play in your strategy. You can find plenty of opinions on this topic, but evidence is a lot harder to come by.
Dane Frederiksen has been working in video for 30 years and spent the past 15 years helping B2B companies reach their customers through video with his company, Digital Accomplice. Given all the hype around GEO, Dane recently started wondering: Could video be a way for companies to become more visible in this new zero-click environment?
He believed it could, given the fact that platforms like YouTube include transcripts that are searchable for both SEO and GEO purposes. But he didn’t just want to rely on a gut feeling—he wanted to see what evidence he could gather to support his hypothesis.
What you’ll find in this article is the story of how Dane went about attempting to answer this question. First, you’ll meet the panel of experts he spoke with and sought advice from as he designed some experiments. Then we’ll walk through each phase of the experiment, sharing what Dane did and how his conversations with GEO experts led him to revise his approach (Spoiler alert: this happened A LOT!). Finally, we’ll share how these learnings have shaped how Dane is thinking about the role of video in GEO and what that might mean for your own strategy.
Meet the panel of experts
Dane sought out several people who have experience with SEO, GEO, and content marketing strategy to grow his understanding of what a sound video GEO strategy might look like and get their advice on his experiment design. The experts included:
Jessica Hennessey, the Founder of Resonate Online and Better Sites, where she works as both a Web Strategy Consultant and GEO Product Leader.
Adrian Dahlin, the Founder of Search to Sale, a B2B SEO and GEO agency.
Cassie Clark, a Fractional Content Strategist and the host of Found in AI, a podcast about getting your brand found in AI search.
Christopher Penn, the Co-founder of Trust Insights, a data and analytics agency, and a respected authority on AI.
The original hypothesis: Structured video increases a business’s chance of showing up in LLMs
Based on his experience with video and understanding of GEO, Dane felt reasonably confident about his hypothesis: Structured video (which includes clips, transcript-rich landing pages, and schema) increases a business’s chance of showing up in LLMs and should therefore be a core part of GEO strategy for B2B companies.
Here’s how Dane explained his reasoning, “LLMs are looking for answers that have some credibility. And so a claim that is supported by a lot of people agreeing with it—like on a Reddit thread, or like lots of people commenting on a YouTube video—that gives things more weight. And because LLMs look at the transcripts and the other text data on web pages, that’s all information that, if you make it available, will give the LLM something to chew on.”
Cassie Clark has a helpful metaphor for this. She said, “The way these AI systems work, I call them the nosy aunts. Before they recommend something to someone, they’re going to go verify it across different channels.” And when it comes to video specifically, she said, “Something like 93% of Google answers come from the Google ecosystem when Gemini or AI overviews give an answer. And YouTube is part of that. So if you want to show up on Gemini, you really need to be part of a YouTube strategy.”
Christopher Penn also shared a technical perspective on where Google is pulling its insights from: “When you look at inside Gemini in AI Studio, where Google grounding sources come from—YouTube’s in there. You look in the Vertex APIs, which is Google’s enterprise search system—YouTube’s in there. So there’s a lot of different contact points where you can illustrate: we know Google is getting its data from YouTube.”
Video also gives you the opportunity to establish authority by creating more pieces of information that are saying the same thing. Dane said, “If you take a video and embed it into a blog page on your domain, and the blog page has text that mirrors what the video says in the transcript, and the schema is also supporting that, that’s more pieces of information that align, which makes it more trustworthy.”
Dane’s conversation with Jessica Hennessey reinforced this point. She said, “LLMs are lazy. The more clear, pointed, nicely tied in a box kind of content that you give them, the more likely they will be to repeat that fact. They’re not looking at an entire blog post and summarizing it—they’re looking for a quotable fragment they can pull out.”
Finally, Adrian Dahlin made the point that video can help set you up for success in the long-term. He said, “It’s clearly one of the things that influences AI, that appears in AI citations and generated answers, depending on your industry. There’s a reason to do video for its own reasons—plus, if that channel also is one of the things that AI wants, you’re future-proofing.”
A rough draft: Dane’s first experiment design
With his hypothesis and encouragement from his early conversations with the experts, Dane decided it was time to start designing an experiment. He explained some of the constraints he was aware of from the beginning: “If you search in AI for anything, you don’t know what the results are going to be. There’s been a lot of data that it’s deterministic and probabilistic, so the outcome can be different every time.”
So what’s the best way to test this? By doing multiple tries and multiple searches on different AI platforms. Dane built a tool in Claude Code to do this. He fed it 20 prompts and ran them across 5 platforms. He’d track the results manually in a Google sheet over 90 days.
Dane explains his reasoning for this approach: “Let’s say you were trying to get to a definitive answer of what types of things come up when people search for, ‘Where’s the best place to buy bananas in Denver?’ If you search for that on different platforms like Perplexity, Claude, ChatGPT, and the answer keeps coming up pretty much the same, you’re getting closer to some sort of truth.”
Dane’s goal was to produce a snapshot that he could provide to a prospective client and say, “Hey, I did a search on your company, and here’s where you’re showing up and here’s where you’re not.” This visual artifact would also include some of the methodology, so Dane could tell the potential client which searches he’d run on which platforms and which key phrases he’d looked for.
It seemed promising… until he had the chance to talk it over with Adrian.
Adrian’s overhaul: Minor and major changes to the experiment setup
In Dane’s conversation with Adrian, a Seattle-based B2B SEO + GEO agency owner, they discussed what Dane had built so far. And Adrian was quick to point out the flaws in Dane’s initial design. Some of his critiques were relatively straightforward, like ditching the manual sheet Dane was using for tracking and replacing it with a tool like Profound, Peec, or Gumshoe.
And some of his critiques required rethinking the experiment in more substantial ways. These included:
Start with an audit of buyer prompts
Adrian argued that looking at the top-cited sources wasn’t a solid strategy. Instead, he advised: “I want every business to start by doing this audit. Don’t assume that because X website is the number one cited source across all LLMs that that’s where your strategy should be. Figure out prompts worth tracking. Base those on your keyword research. Develop prompts, track those prompts, then look at the domains showing up in the cited sources.”
Recognize that not all types of video perform the same way
Adrian agreed that video does get cited by LLMs, but he was clear that not all video performs the same way. Here’s how Adrian broke it down: “Sometimes you’ll see YouTube high on that list and you’ll look at, okay, what kinds of videos are the ones getting cited? And it’s educational, instructional, tutorial type of content. If you see that in the cited sources, now you have very clear evidence and a rationale to go have a video strategy.”
User-generated content is also powerful, according to Adrian: “AI, the people building these models have chosen to preference user-generated content with some signals of authority.”
Give yourself plenty of time for the experiment
Adrian also noted that it takes time for video to show up in LLM results. His advice? “Construct a good strategy and then try to get buy-in to do it for six months without interruption—because you’re not going to do a one- or two-month experiment with video and change your business.”
Cassie’s fix: Rethinking content quality, production value, and tracking cadence
Next, Dane connected with Cassie Clark, a fractional content strategist who’s been running live experiments on her own brand. Cassie shares Dane’s belief in the power of video. “I think video really is mission critical,” she explains. Dane knew Cassie would be the ideal person to weigh in since she’s been hands on with video and gathering her own evidence. Here were a few of the key takeaways Cassie shared.
Avoid the snake eating its own tail problem
Cassie noted that testing video for GEO using video about GEO content poisons the signal and might be too much like a snake eating its own tail. She said, “If you’re going to test out video in that regard, maybe don’t do it with the GEO series. Maybe do it with something consumer-focused.” Although this was sound advice—to pick a clean sandbox vertical with real buyer demand—Dane opted to continue creating video content about GEO anyway since that’s his area of focus and what he’d like to be known for.
Focus on low production quality, high-value content
Based on her own experimentation, Cassie found the best approach was to consistently deliver content that was low on production quality, but high in value for the viewer. She describes the problem with the old school SEO playbook: “The whole ‘let’s pop out 1,000 blog posts’ is not a good SEO strategy to begin with, especially if you’re doing really thin content. It also doesn’t really influence and generate answers as much either.”
And when it comes to video and GEO, Cassie has the same belief, though adjusted slightly to the medium. Her approach is to choose a specific question that she can answer thoroughly and film it on the fly. Here’s her explanation: “I started making videos—when I say low budget, I mean I grabbed my phone and said, ‘Good enough, let’s post it.’ No editing. Very embarrassing. But I wanted to see: If I took a specific question, like ‘What is AI share of voice?’ and made a video answering it, would that show up in an AI-generated answer? And it did. Bad hair day and all.”
Track results on a monthly cadence
Cassie also had suggestions about tracking results from these experiments. She said, “I would check it once a month. And once it starts showing up in the answer, track how long it stays—because it will be replaced by something fresher.”
The Penn grenade: Rethinking the very idea of AI visibility tracking
Next, Dane sat down with Christopher Penn, co-founder of Trust Insights and a widely recognized AI expert. The conversation with Christopher blew up one of Dane’s initial assumptions: the very idea that it’s possible to reliably track LLM citations.
The big reveal: AI visibility tracking is a lie
Christopher was very clear about the fact that LLM responses are probabilistic and heavily personalized, which means it’s basically impossible for a tool to track AI visibility. He didn’t mince words: “Anyone saying, ‘We know what your AI visibility is’ is just lying. There’s no polite way to say it—they’re lying. And if you’re spending money on enterprise AI visibility software, cancel the subscription, save yourself the five to ten grand—it’s worthless. Because of that level of personalization.”
This is not just Christopher’s opinion. He described one person’s LLM experiment that led to wildly different results: “Garrett Sussman set up synthetic Gmail accounts—five or six fake people—gave them different YouTube subscriptions, different Gmail subscriptions. And the brands that Google recommended varied from account to account based on the personalization. So you cannot predict what brands are going to get recommended. Period. End of story.”
An alternative tracking method to try instead
Luckily, Christopher didn’t just throw this massive wrench into Dane’s experiment and leave him to pick up the pieces on his own. Christopher had an alternative measurement method that he believes is much more reliable: “The gold standard for measuring AI visibility: on your contact form, your call center, your email, there’s a little box that says, ‘How did you hear about us?’ Let it be a free-form text field. Using the AI of your choice, analyze the results every month. If the number of people who say ‘ChatGPT’ or ‘Google’ is zero, your GEO strategy isn’t working.”
And Christopher has real data to back up this assertion. He explained, “11% of Trust Insights’ business comes from AI referrals now. In the last six months. It is astonishingly high.”
This created an awkward problem. Adrian had recommended using tools like Profound, Peec, or Gumshoe to run the experiment—the same category of software Christopher just called worthless. Dane now had two credible experts pointing him in opposite directions.
His working solution: use Christopher’s contact-form method as the primary measure of whether GEO efforts are actually driving business, while using tracking tools more loosely as a directional signal for which content is appearing and where.
What’s next? Watch this space!
Dane has now gone through several iterations of his experiment design, gotten feedback from several GEO experts, and come up with a plan for what he’ll do next. He said that he ran an initial test across several different LLMs for specific search terms and which sources they cited. So far, he was not being cited for any of those terms.
His next step would be to make videos of those questions, publish them on YouTube, and publish the transcripts as individual blog posts. Then he’d run another iteration where he would put embedded YouTube videos on blog posts to see if a video embedded in a blog post performs better in GEO.
Dane has a hunch about what will perform the best. He said, “It’s pretty obvious to me that having an embedded video in a blog post gives you the most coverage of potential ways you could be cited, because it could find the video or the transcript or the description, or it could find the blog post. And having the transcript on the blog page would help. And then just having the additional trust layer of a video embedded into a blog post seems to make that generally a more valuable piece of content, because it’s got a lot more pieces to it than just a video on its own or a blog post on its own.”
To recap, Dane will be testing the pieces of content, the amount of content, the combinations of content, and getting it dialed in and specific based on customer language to see how all those things come together.
The urgency of getting this right extends beyond current AI search behavior. Cassie sees a near-future shift that raises the stakes considerably: “If agentic commerce is eventually going to be making a purchase on behalf of someone, or signing up for a demo on behalf of someone—that’s really going to affect revenue. And there’s not a person in there pushing the button. It’s going to be so much harder to be the answer when that happens if you’ve not already done the work now.”
Jessica, who has been collaborating with Dane on the experiment design, points out that the work never really ends: “LLMs like recency. It isn’t just a one-and-done. You need to keep feeding the engine to stay relevant.”
If this sounds like something you’d like to be involved in—either to share your expertise on setting up the experiment and interpreting the results—or as a business looking to experiment with video’s impact on GEO, get in touch! You can connect with me (the writer, Melissa Suzuno) or Dane (the video guy) on LinkedIn or email Dane at dane@digitalaccomplice.com.
Dane emphasized the fact that even if the impact of video on GEO is hard to prove right now, there are so many benefits to adding video to your marketing strategy: “Video helps in a lot of ways—GEO is just one of them, in addition to things like building trust and visibility. There are a lot of different benefits that all stack up.”
Main image photo by Vanilla Bear Films on Unsplash.