Do AI-Narrated Videos Get Cited by AI Search? 493 Videos, Zero Citations

0:00 AI-Narrated Videos and Citation Reality
Hey, I’m Jeremy with PushLeads, helping small businesses get more leads without struggle and frustration. And today, we’re looking at a really interesting topic that I’ve been personally researching. Do AI-generated videos actually get cited by answer engines, by chatbots, by AIs? And this all goes back to sort of an idea that I came up with. What– And this all goes back to an idea that I came up with, which is basically that, you know, if we could make a PowerPoint slideshow with an actual AI voice, like a real, you know, human voice, but an AI version of it, is that enough to get AI citations? And the short answer is no, it’s not enough. There’s no evidence for that. But let’s dig in. And throughout this video, you’ll see references to PushVid, which is sort of the version one of this product I created, and that, and that product basically is essentially a slideshow video product. So no, there’s– So there’s definitely a good reason to have a slideshow video product. The question is, is putting it on YouTube, will it give you citations? Um, so we started doing this and then later on retracted these because it looks like there could be a possibility that Google could consider it abuse. So I’m not saying that you shouldn’t do slideshow videos, but I was– I’m s– but I am saying you need to be super careful. So let’s look at what the data says about it and jump into it.

1:25 YouTube Dominates AI Citations Not Slideshows
So the verdict in three bullets is that there’s zero documented cases, there’s no first-hand verifiable account of an AI-generated narrated slideshow, a faceless video or an AI avatar video being chi– be-being cited in Google’s AI overview, Perplexity, ChatGPT, Gemini. Um, and so it doesn’t actually exist. Um, YouTube is, of course, cited, um, but in our videos, which we just started doing about a month ago, um, at the beginning of August or so, twenty twenty-six, we’re not being cited, but obviously we just start– There’s a couple of videos we’re cited in, but not very many. Again, sort of my picture is that I’ve been in and out of YouTube for a long time. Really jumped fully into it because, um, you know, essentially there was just so much traffic on the shorts and there’s so many videos that do exactly what I do. Um, there’s so many companies that do exactly what I do, so what’s the point? And then when I looked at my own fleet of forty-five clients that we’re doing, uh, SEO for, mostly service-based businesses, I started realizing that there was one competitor that kept showing up over and over again, um, that’s competing for my clients’ visibility, and that was YouTube. It showed up fifty percent of the time. So YouTube as a domain captures about twenty-one percent of Google’s AI overview citations, but the overwhelming videos are the human presenter, the hands-on demonstration content, or maybe videos like this where it’s an actual video of me talking, walking through a slideshow. This is real me, not AI version of me. This is not HeyGen, right? Um, so anyway, it’s human first, um, and Google, um, and YouTube seems to prefer human-first content. And then our own, um, data has, you know, zero citations as we talked about it. So small channels can be cited according to Utterly AI’s model, where small channels can be cited, so it points to that this will work, doing videos will work, but the reality is, it’s gonna take some time. So let– So let’s get in–

This is part of our AI search visibility guide.

3:31 What Actually Gets Cited Demonstration vs Narration
So chapter one, YouTube dominates AI citations, but You– but not– So ch– So chapter one, YouTube dominates AI citations, but not YouTube slideshow videos. So independent trackers converge on a striking number. YouTube captures about twenty-one point one percent of cite– in Google’s AI overview, and that comes from Ahrefs, uh, as of June of twenty twenty-six, looking at three million US searches. And BrightEdge reports a two hundred ex advantage over the nearest video competitor, which is Vimeo, only having about one percent, point oh one percent. Peak AI’s thirty million source studies ranks YouTube as number two behind Reddit. So this, so this does not mean that a high– So what does this not mean? A high-level domain citation tells you nothing about the types of videos that are being selected. So every study is really looking at the same pattern, the how-to, the tutorial, the visual demo and the product comparison format. So, so let’s keep going. N– Um, so what actually gets cited? It’s demonstration, not narration. So every third party describing a character– So every third-party study looking at cited videos looks at the– came to the same conclusion. So every third party study of character-cited videos came to the same conclusion. MPDigital’s Q one twenty twenty-five analysis of ten thousand SGE overviews reported that the how-to citations were up six hundred fifty-one percent and the visual demonstration videos were up five hundred and ninety-two percent. So ch– my own top fifty cited channels also meets that pattern. So it’s how-to tutorials, which are step-by-step instructions. It’s product demonstrations, which is hands-on comparisons, and then, of course, visual repair and bu– anything with installation, construction and repair. So an AI-narrated slideshow over stock footage cannot replicate this experience where the– the transcript could be readable by an AI, but the– but it just means that the video itself is not gonna get citations. Chapter two, source quality, who founded what?

5:44 Source Quality Who Funded the Research
Chapter two, source quality, who funded what? So we need to look at two studies here, and this actually is a foundation of two tiers. Tier one is the primary research and disclosed methodology, which is Utterly AI looked at a hundred million citation instances over thirty days, six engine– And this was in March of twenty twenty-six. Ahrefs Brand Radar looked at four hundred and sixty-three keywords, which is four million URLs, three million US queries. And SearchE ranking German house study looked at fifty thousand searches of December twenty twenty-five. Peak AI also looked at thirty million sources, five engines, and Digital Applied looked at a thou- A thousand AIOs in April twenty twenty-six. And so this is a ton of data here, and they all have partial to full methodology. And then tier two is vendor content marketing with LR8 AI samples dozens, no raw data, just they sell their own video AI tool. And BrightEdge, which sells AI catalysts, and the– and, and their founder claims Google watches video, which is unconfirmed by Goosh. F-fi– And then EverythingPR/5W, which is a synthesis of other studies, Veed, vidIQ, NP Digital, Ratty Rant, et cetera. So a ton of information that goes into it.

7:06 XLR8 Vendor Claims Zero Independent Replication
So let’s look– So let’s, so let’s look at the XLR8 problem, which is one vendor, and there’s been zero replication of it. So the only source claiming that AI avatar videos get cited at the same rate as the human presenters is XLR8. That’s a mouthful. This is a vendor who sells an AI vi-video generation tool. And in, in their July twenty twenty-six guide, the– it made three specific claims. It say that– It said that every cited video came from a channel with less than a thousand subscribers, and its AI avatar and human presenters work, quote, "both work equally," and it has– and its own near zero subscriber channel outsites established channels. So this got– So this guide’s sample is dozens of videos with no raw data and no disclosed methodology. And of course, they sell this product, right? So if they’re selling an AI video product, of course they’re gonna put out stuff saying that AI video works. It really kind of– This whole thing really seems like a gray area, honestly. And this– and it, and it, and it also contradicts the far larger la– and it also directly contradicts the far larger, uh, Otterly AI study on every subscriber finding. So the verdict here is treat every XLR8 claim as an unverified single-vendor marketing. It’s a sole source of any kind of AI ven– AI avatars get cited, and of course, the thousand-subscriber floor s-floor claims. Neither has been replicated by an, by an independent self, and that should send off alarm bells. And then we’re gonna get into the thousand subscri– And then we’re gonna get into the thousand sub-subscriber floor, which is almost certainly false. And Otterly AI looked at a hundred million citations, which is the largest study on YouTube AI citations, and under thirty-five perce– and thirty-five percent were under ten thousand subscribers. So– So that was the actual number is ten thousand subscribers, and then, and then forty percent were under a thousand views. Of the AI-cited videos, fewer had less than a thousand views at the time of the analysis, and then the median channel had over– had fewer than forty-one videos. So Otterly’s AI– So Otterly’s– So Otterly AI’s expli-thesis is the best answer wins, not the biggest answer. So Otterly’s AI explicit– So Otterly AI’s explicit thesis is that the best answer wins, not the biggest channel, and the thousand-subscriber floor is an artifact of XL, of XLAR’s tiny self-interested sample, not a real threshold. So small channels do get r– s– so small channels do get cited regularly when their content matches the query. So how exactly do AI engines ans- So chapter three, retrieval mechanics,

10:02 Retrieval Mechanics How AI Engines Find Videos
how AI engines actually retrieve videos. So it’s important to understand the technical pipeline because this clarifies exactly what determines the citation eligibility and of course, where slideshows fit in that mix. So the first thing we have is a text crawl where ChatGPT… So the first thing we have is a text crawl where ChatGPT, Perplexity, and Claude are fet-are fetching– where ChatGPT, Perplexity, and Claude are fetching the transcript, description, and chapters. They cannot watch the frames, only a slide, and so therefore a s– and so therefore a slideshow is not structurally penalized. Frame sampling, Google’s Gemini stack can sample visual frames at about one frames per second, but tellingly the G– but tellingly, Gemini cites YouTube the least of any engine, which is point two percent, which doesn’t make any sense to me because it’s Google’s product. And then we third– And then third, we have timestamp parse of it, which is seventy-three percent of all timestamp citations occur in Google’s AIO, twenty-seven percent in Google’s AI mode, and clean chapters, which is, you know, timestamps. And of course, clean chapters with timestamps are structurally imp– We also have the slideshow’s number one genuine advantage, and that’s the ceiling. What transcript-driven retrieval means for slideshow for ChatGPT, Perplexity, and Claude citation runs entirely through the text label. So this is where a well-transcribed, chaptered slideshow is in principle as parsable as a human-presented video. But John Mueller has con– And John Mueller has confirmed that Google uses transcripts and captions to understand video content. So this alone is the strongest genuine point in the favor of a slideshow-only video. And it’s the only one with structural support, which should give you red flags, right? Because Google and AIs prefer things a certain way, and when you try to not– and when you do things not their way, that’s c-considered gray hat or black hat. So again, doing slideshow videos appears to be a gray hat sort of category. So where the argument breaks down is parsability and select… On the how-to and demonstrated queries where YouTube actually earns its citations share, the re-the reason a specific video is actually chosen is b-is because it visibly answers the an– is because it visibly shows the answer. A clean A clean transcript on a B-roll gets you the eligibility to queue, but it does not win the citation selection over a channel that actually demonstrates the task. So structural access doesn’t mean citation success. Super important that you understand that.

12:48 Enforcement Risk Scaled Content and Inauthenticity
Chapter four, the enforcement risk. The real penalty vectors is scaled content and inauthenticity. So Google scaled content abuse policy that came out in March of, in March of twenty– So Google scaled con– So Google scaled content abuse policy, which came out in March of twenty twenty-four and enforced March of twenty twenty-six, and the explicit… the qu– And it’s explicitly method neutral. No matter what content is produced, through automation, human efforts, or some combination, the volume without value is a red line, and Google views that as scaled content abuse. So an automated single voice slideshow deployed at a scale is directly in scope of this, not c- not because it is considered AI-made, but it’s because it’s templated and generic. And funny story, I remember back in twenty eleven, I had someone paying me to do slideshow videos that had keywords, and they were targeting anyone searching for cable installation or internet installation. And so the videos were like cable installation, Charlotte, North Carolina, cable Asheville, North Carolina, internet, internet service provider. And so we were cranking out a ton of these videos and thinking that… I was thinking, "Eventually, Google’s not gonna like this." And of course, I was right. So I don’t– That’s– So that was like fifteen, seventeen years ago, when I was kind of somewhat gray hat. I don’t do that at all now. Obviously, it makes… You obviously you have to play the game. And then, and then YouTube’s inauthentic content policy, which is July twenty twenty-five and January of twenty twenty-six, is the enforcement wave. And so YouTube’s July twenty twenty-five monetization update and reported and updated January twenty twenty-six, th-this enforcement wave targeted mass-produced or repetitive content. So is slideshows with an AI voice sort of repetitive content? Well, it’s a gray area, right? Um, so ratter guidelines have… And then AI disclosure labels are a red herring. So YouTube’s alterner– YouTube’s altered or synthetic label is a transparency signal, not a penalty flag. A single Eleven Labs voiceover on a slideshow generally does not trigger disclosure, disclosure requirements. The policy targets realistic synthetic depictions of real people. So if you’re using HeyGen, then you have to, then you have to explain that it is AI-generated, but if it’s the voice, I– you don’t have to say it’s AI-generated. And of course, chapter five, first-party data.

15:21 PushLeads First-Party Data 493 Videos Zero Citations
So PushLeads, which is our own age, this is the most important data point in this entire analysis. So our current s- so our current channel stats are six hundred and seventy-seven subscribers around a hundred and sixty-one public videos, eighty-two thousand or, or so, eighty-two thousand total views, a hundred and… And then we’re tracking four hundred and ninety-three tracked videos. And of seventeen thousand scanned citation links, zero citations were found. Again, I expect that to start rising in thirty to sixty, ninety days. I am starting to see YouTube search finding me, and I am starting to see some visibility, but it does take time when you start cranking out content. And really, I stopped doing content because I didn’t think that it was even valuable to Google. I had no idea that Google’s AI was actually looking at all of these YouTube videos. So why zero matters so much? Under Otterly’s AI model, channel size is not the barrier. Small channels with ar-around ten K, five K, or even less than a thousand subscribers are cited regularly. But there’s– but we have zero citations across four hundred and ninety-three videos, and that’s because most of these videos have just been generic videos like what is SEO, how does SEO work? They’re not specifically answering specific questions. However, though, some early data which n- is not in this PowerPoint is that when you make videos covering specific points, then you get visibility for that. So it really comes down to the targeting which– versus… And here’s a look, and here’s a look at first-party data versus published benchmarks. So, um, we have twenty percent of my… So my YouTube citations share is around twenty percent versus Ahrefs around twenty-one percent, and then everything PR twenty point nine percent. The YouTube actual citations and answers according to my– across my whole fleet is around fifty percent, which is higher than most domain share figures, but that’s consistent with Bright Edge study of around sixty percent and SEs an- ranking around eighty-two percent. And then how many chat… And, and then how often is ChatGPT citing YouTube? Um, ev- essentially, it’s identical to Otterly’s AI finding that chat just drives around four percent of YouTube citations. So ChatGPT barely uses YouTube. So I always tell my clients, "If you wanna show up for Google’s AI, you need to have ongoing video content. If you wanna just show up for– If you wanna show up for ChatGPT, then you wanna show up on– then you just– then you wanna have Reddit only. Then you wanna answer questions on Reddit." So the real question is, is even making this content even worth it today, right? Because Google Search really accounts for what? Ninety, ninety-five percent of the entire equation, the traditional search. However, though, AI usage has gone up from, I think, five to fifty percent in the last year or so. So that’s really gone up. So chapter six, the traffic reality check.

18:20 Traffic Reality Check Citation Share vs Business Value
Citation… So chapter six, the traffic reality check. Does citation share actually equal traffic equal business value? The answer is today, no, it doesn’t. So even we’re having the citation case instructions descri- The business case for chasing AI citations is actually somewhat deserves somewhat. So even where the citation case is strong, the business case for chasing AI citations deserves some scrutiny here, and the evidence is sobering. We have AI traffic referral being tiny today, right? In twenty-seven, twenty-eight, and thirty, it’s gonna be growing up short exponentially. But A– But Ahrefs study across thirty-five thousand sites found that ChatGPT, Perplexity, and Gemini send roughly the same amount of percentage of traffic as Reddit to websites, which is like point two one two percent of total human traffic. It… And then Reddit’s Steve Huffman says chatbots are not a major traffic driver today. But one of the things we have to pay attention to is that beha– human behavior is changing. More and more people are expecting automatic mode, right? And so this shift in human behavior is causing all of us in marketing to really pay attention to AI citations, even though today it doesn’t have much value. Hard to know. AI, AIO, AIO citations are underperforming, uh, compared to the blue rink. AI ci– AIO citations are underperforming comparing to the blue rink. AIO citations are underperforming comparing to the blue… AI citations are underperforming compar… So AI citations underperform compared to blue… So AIO citations underperform comparing to the blue ranks. the blue links, of course. And s– Lily, and Lily Ray at Search Engine Land said, "In my research from the first half of this year, AIO citations constantly underperformed even compared to traditional blue links near the bottom of the SERP." And then citation volume is not the real constraint. So YouTube’s citation share is real. The business question is whether chasing that sha… Especially through slideshow content that typically earns zero citations, is that the, is that the highest use of the production value for a company? And we can take a quick look at real versus unlisted versus private citation eligibility. Public videos are indexed by YouTube search and AI engines at around two to… And they show up in around two to three weeks, and they’re re– and they’re reliably and fully citable. Unlisted and private videos are citable only if the URL is linked on a public page or in a playlist. And of course, private videos cannot be indexed or cited at all by search engines and having zero vi– and they have zero eligibility. So what should you do from here?

21:17 Recommendations What to Do From Here
So chapter seven, recommendations. What should you do from here? So stage one, stop counting on an AI slideshow video as an AEO asset. The evidence is absent, not merely weak. There is a meaningful difference between weak evidence and no evidence. So an AI slideshow video earning citations. For an AI slideshow earning cite… The evidence. So for an AI slideshow earning its, earning AI citations, the evidence is absent. There’s no known channel, there’s no name channel, no ID, and no screenshots been produced by any source. Our own videos are also showing that it doesn’t work, of course, but our, most of our content until the last month or so has not really been hyper-focused content answering questions. And of course, that’s what we’re doing now. Unless I’m doing this type of video, which is more about looking at the data. So what should you keep on doing? Video earns its keep if it, if it delivers… So what should you keep on doing? Video earns its keep if it delivers on on-page engagement, human audience utility, conventional YouTube and video search, and content repurposing across format. These are not citation benefits, but they are real, but, but they do show that slideshow videos do have real business value, even if it’s not gonna actually drive citations. So after stage one, you’re gonna jump into stage two, which is run one clean falsifiable test over the six– over the next sixty to ninety days. Because retrieval is transcribable, test whether one variable actually works. Produce ten to fifteen videos meeting all criteria, and we can– and track the citation experience. A-a-and attract– and track the citation appearance over six weeks. You’re gonna look at the single que– Make sure that you’re answering one question per video. You’re gonna be doing the verified transcript. Make sure it’s clean, manually checked. Not Google’s, not Goo– not YouTube’s, your own transcript, which is much more accurate. You need to have the proctor– You need to have the proper chapters with the start, the timestamps, and then ten seconds each. And you need to have a structured description, which at least three thousand words with entities and hashtags. When… So on the human presenter question, what the actual data says. The data does not say AI video never works and human video always does. It says that de– It says that videos that demonstrate value and structure are driving sites. And it, when it… A human presenter helps when it enables real physical demonstration and establishes EAT signals. The combination of a real face plus a visual demonstration plus clean transcripts is the configuration that gets mostly aligned with what actually gets cited. A talking head alone is not enough. So a human presenter without a demonstration is n– obviously not, is obviously not better than a well-structured demonstration slideshow. So that’s why we’ve been working on a version two of this that is actually human prese- along with a whole bunch of other visuals that make it interesting because, I mean, this video right now is in the raw recording is up to about twenty-seven minutes. I mean, this final video might be twenty, twenty-two minutes. The question is, is someone gonna watch this big, long video? Some people will, but most people won’t. I’m really making this video for my own documentation or anyone that wants to dig into this. And of course, having this video content on Pushli, um, a-as an enter hidden page with the video schema does add value. And the real differentiator here is demonstrating the value, the transcript structure, and the topical answer fit. These variables matter across all formats. So pivoting to human presenter alone without demonstration content doesn’t actually solve the underlying problem. And the last thing to look at here is key caveats to hold in mind. There’s– An absence of evidence doesn’t equal an evidence of abs– There’s no published study isolating AI-generated versus human-made videos in citation datasets. Otterly AI, BrightEdge, and Ahrefs all lump YouTube together. AI citations could exist and simply just be undocumented. But it’s kind of a risk, right? You wanna deal with a penalty or even having your channel demonetized or potentially even having your channel, you know, removed. I don’t know if it’s worth it personally. Um, I’m working on building my brand, not destroying it, right? Um, so I made this slideshow product, but I’m not, I’m not really actively rolling it out on my main channel. I might do– I might test it on some other channels at one point. Cross-vendor agreement is partially circular. Mostly YouTube domin– releases. And then number two, the cross-vendor agreement is partially circles. Per– is– And then number two, the cross-vendor agreement is partly circular. Most YouTube dominates figures are tracing back to Ahrefs, BrightEdge, and Otterly AI. So your first-party data is much more trustworthy. And then figures are a fast-moving window. So AIO moved from Gemini Three in January twenty twenty-six, and cita– and citation– So AIO moved from Gemini Three in January twenty twenty-six, and citation behavior has been very volatile. The key dates are Otterly AI in March of twenty twenty-six, the Ahrefs domain leaderboard in June, in June of twenty-six, and SE ranking in June– in December of twenty twenty-five. And of course, we have that XLR eight study of July twenty twenty-six. And then site fake doesn’t equal revenue because AI referral traffic is point one two percent of the total at this point. So AIO citations are underperforming, and they’re showing below– So AI citations are underperforming below SERP blue rinks. B-b– So AI citations underperform bottom of SERP blue rinks– blue links for clicks. So I’m Jeremy Ashburn with PushLeads. I hope you found this valuable, and thank you so much