AI-generated product recommendations are drawing on creator videos that brands often have little control over, according to new research from creative intelligence company Sundogs. The findings suggest that the content informing AI-driven product discovery follows a different set of rules from those that typically guide brand video campaigns.
Sundogs analyzed 3,900 videos across the U.S. consumer packaged goods and automotive sectors, comparing 400 AI-cited videos provided by marketing services firm Jellyfish through its GEO tracking platform, with 3,500 non-cited videos from the same creators. Published in July 2026, the study examined which creative characteristics were associated with AI citation, from product demonstrations and comparisons to video length and production style.
Ben Jones, founder and CEO of New York-based Sundogs, says the findings reflect a shift he began tracking over the past year as YouTube and other video sources gained ground in AI-generated recommendations. “I’ve sat in rooms full of CMOs,” Ben says. “They all say, ‘Our search traffic is falling off, and we don’t know what to do about it.’”
For brand marketers, his read on what comes next is clear: the question is no longer how to make videos that attract views, but how to create partnerships and content that help products surface when consumers ask AI what to buy.
How AI Learned to Watch Video
“Last fall, we started to get signals that citation sources were shifting from Reddit and text to video,” Ben says. The shift, he argues, went further than distribution. The models themselves stopped treating video as complex text and began reading it natively, frame by frame, interpreting narrative structure, audio, and visual cues together. Ben calls the older approach “transcoding.”
Video provides between 80,000 and 100,000 times the richness of text as a training source, according to the research, and the volume of new video being uploaded represents, in Ben’s framing, an essentially infinite supply of new material. As models began ingesting that content natively, YouTube emerged as the top citation source across AI engines, first for Google’s Gemini and then across the field.
Ben notes the pattern is clearest in CPG and auto, the verticals Sundogs studied. For food, beverage, and retail categories, the picture appears more mixed, according to the research.
Creators Are Driving Brand Discovery
The first finding that stood out from the Sundogs analysis was not about creative format. It was about who was generating the cited content. In CPG and auto, approximately 60% of AI-cited videos came from creators, around 30% from editorial sources such as publications and aggregators, and just 10% directly from brands, according to the research.
“Here’s this critical business problem for brands,” Ben says. “Product discovery sits with creators and creator video in particular, and this kind of creator video that they don’t make or understand or control.”
For brands that have built their creator strategies around paid campaigns and brand-approved scripts, this means the content AI is drawing on to recommend their products exists largely outside that system.
Image: A representative example from CPG, exhibiting many of the same characteristics: 4 years old (no recency), 17 minutes long, 9k views (substantial but not massive), from a creator. Like the auto video, it focuses on specific product functions and use cases. This is product marketing based on functional differences, rather than brand marketing. This creator, Susan Yara, has a scaled audience (over a million subs), so that might be a contributing criteria, but in our review, the substance of the video mattered more than the audience size of the creator.
The Metrics That Don’t Predict Citation
The Sundogs findings disrupt several assumptions about what makes video valuable. View count is not a reliable predictor. Cited videos can have under 1,000 views while other videos from the same creator, with far higher traffic, go uncited. Recency does not matter either. Many cited videos are one to two years old or older, and publication date carries no significant weight, according to the research.
Production quality is similarly irrelevant. AI voiceovers and stock footage appear frequently in cited content. The research documents one cited video comparing two vehicles that was two years old, ten minutes long, had only 12,000 views, and relied entirely on an AI voiceover with stock footage. Creator credibility does not determine citation either. “A great creator who published a lot of videos on a topic did not get cited more than an AI content farm in Kenya,” Ben says.
What the engines are not measuring is influence. They are measuring deep contextual accuracy.
The Creative Profile of a Cited Video
The videos that do earn citations share a specific profile. Formal product reviews and physical demonstrations appear in 77% of cited videos, a 29-percentage-point lift over non-cited ones, according to the research. Videos that explicitly frame the problem being solved appear in 76% of cited videos, a 35-point lift. Clear product benefits appear in 90%. Mentioning specific ingredients or detailing scientific effects drives citation lifts of 30 to 34 percentage points.
Comparison videos are also common: comparisons between brands appear in 71% of cited content, while comparisons between different products appear in 68%. A practical, self-assured tone is present in 84% of cited videos. Exploratory calls to action, the “check it out” variety rather than a direct push to buy, add a 21-point lift.
Ben describes the format as resembling what YouTubers have long produced independently of brand campaigns: long, specific, comparative reviews that read more like research than advertising. “The engines love that,” he says, “because they’re trying to give the best answer they can to a specific person.”
Format matters, too. Cited videos are commonly eight minutes or longer. YouTube Shorts, Ben says, are cited at a fraction of the rate of long-form video.
Long-Form Creators Are Sitting on Uncaptured Revenue
Ben frames GEO not only as a discovery behavior change but as an uncaptured revenue opportunity for creators. He estimates $11 billion is shifting from traditional search engine optimization into AI engine optimization, and argues that creators, not platforms or brands, should capture most of it. “Creators should be able to monetize it,” he says. “You’re already earning it for them; you should be paid for it.”
He cites Erica Louie (@missejlouie), a creator who participated in Sundogs’ research launch and discovered her videos were being cited to answer product questions for brands she had covered. According to Ben, she recognized the implication immediately. “‘My videos have been cited, and they’re providing product discovery value,’” she said. “‘I need to get my clients to recognize: you’re getting value from this, you should pay me for it.’”
For creators already making long-form, fact-dense content, the research is largely a validation, and in Ben’s view, a new negotiating argument. For those focused on short-form, the adjustment is harder. Ben says he spoke with a company built around YouTube Shorts that believed high view counts would translate into citation volume. High volumes of Shorts do not produce citations at any comparable rate to long-form video, he says, and a shift in production approach is required.
Editorial Freedom Is the Price of Entry
For brands, the research creates two distinct pressures. The first is direct: creator content needs to be part of product discovery strategy. “Anyone who’s not already partnering with creators needs to, 100%,” Ben says.
The second is harder. Citation-eligible content requires creators to talk honestly, specifically, and comparatively about products, including competitors. “They need to be able to say, this shampoo is better for dry scalp than this one, but this one is better for XYZ,” Ben says. “And that makes brands very uncomfortable.”
Brands willing to grant that editorial freedom gain something they cannot produce internally, he argues: verifiable, fact-rich content that answer engines weight as a reliable source. Brands that insist on controlling the message are generating the roughly 10% of cited content attributable to brand-owned video. “Marketers are like, ‘I need to message to you,’” he says. “Don’t. You need to be useful to them with a density of information you’re not used to. Creators are a great way to get there.”
Pilots, Partnerships, and What CMOs Are Asking
Heading into the final months of 2026, GEO is Sundogs’ top priority. Ben says the company is developing pilots with brand clients to identify creators already being cited for their product categories, with the goal of either recruiting those creators or training existing partners to produce content that matches the citation profile. Sundogs is also working closely with Adelaide, which studies attention quality in advertising, and Circana, which holds sales lift data, to connect creative diagnostics to downstream commercial outcomes.
The question Ben hears most from CMOs is not whether AI is reshaping discovery but how quickly brands can adapt. “Every CMO is asking me about it,” he says, “and nobody really knows how to do it.” For creators already making long-form, comparative video, his read is that they are sitting on a revenue stream that brands are benefiting from but have not yet explicitly priced.
“Curiosity is a superpower,” Ben says. “It’s a great time to ask questions. You can get answers faster, and you can make things happen faster.”
Cecilia Carloni, Interview Manager at Influence Weekly and writer for NetInfluencer. Coming from beautiful Argentina, Ceci has spent years chatting with big names in the influencer world, making friends and learning insider info along the way. When she’s not deep in interviews or writing, she's enjoying life with her two daughters. Ceci’s stories give a peek behind the curtain of influencer life, sharing the real and interesting tales from her many conversations with movers and shakers in the space.
AI-generated product recommendations are drawing on creator videos that brands often have little control over, according to new research from creative intelligence company Sundogs. The findings suggest that the content informing AI-driven product discovery follows a different set of rules from those that typically guide brand video campaigns.
Sundogs analyzed 3,900 videos across the U.S. consumer packaged goods and automotive sectors, comparing 400 AI-cited videos provided by marketing services firm Jellyfish through its GEO tracking platform, with 3,500 non-cited videos from the same creators. Published in July 2026, the study examined which creative characteristics were associated with AI citation, from product demonstrations and comparisons to video length and production style.
Ben Jones, founder and CEO of New York-based Sundogs, says the findings reflect a shift he began tracking over the past year as YouTube and other video sources gained ground in AI-generated recommendations. “I’ve sat in rooms full of CMOs,” Ben says. “They all say, ‘Our search traffic is falling off, and we don’t know what to do about it.’”
For brand marketers, his read on what comes next is clear: the question is no longer how to make videos that attract views, but how to create partnerships and content that help products surface when consumers ask AI what to buy.
How AI Learned to Watch Video
“Last fall, we started to get signals that citation sources were shifting from Reddit and text to video,” Ben says. The shift, he argues, went further than distribution. The models themselves stopped treating video as complex text and began reading it natively, frame by frame, interpreting narrative structure, audio, and visual cues together. Ben calls the older approach “transcoding.”
Video provides between 80,000 and 100,000 times the richness of text as a training source, according to the research, and the volume of new video being uploaded represents, in Ben’s framing, an essentially infinite supply of new material. As models began ingesting that content natively, YouTube emerged as the top citation source across AI engines, first for Google’s Gemini and then across the field.
Ben notes the pattern is clearest in CPG and auto, the verticals Sundogs studied. For food, beverage, and retail categories, the picture appears more mixed, according to the research.
Creators Are Driving Brand Discovery
The first finding that stood out from the Sundogs analysis was not about creative format. It was about who was generating the cited content. In CPG and auto, approximately 60% of AI-cited videos came from creators, around 30% from editorial sources such as publications and aggregators, and just 10% directly from brands, according to the research.
“Here’s this critical business problem for brands,” Ben says. “Product discovery sits with creators and creator video in particular, and this kind of creator video that they don’t make or understand or control.”
For brands that have built their creator strategies around paid campaigns and brand-approved scripts, this means the content AI is drawing on to recommend their products exists largely outside that system.
Image: A representative example from CPG, exhibiting many of the same characteristics: 4 years old (no recency), 17 minutes long, 9k views (substantial but not massive), from a creator. Like the auto video, it focuses on specific product functions and use cases. This is product marketing based on functional differences, rather than brand marketing. This creator, Susan Yara, has a scaled audience (over a million subs), so that might be a contributing criteria, but in our review, the substance of the video mattered more than the audience size of the creator.
The Metrics That Don’t Predict Citation
The Sundogs findings disrupt several assumptions about what makes video valuable. View count is not a reliable predictor. Cited videos can have under 1,000 views while other videos from the same creator, with far higher traffic, go uncited. Recency does not matter either. Many cited videos are one to two years old or older, and publication date carries no significant weight, according to the research.
Production quality is similarly irrelevant. AI voiceovers and stock footage appear frequently in cited content. The research documents one cited video comparing two vehicles that was two years old, ten minutes long, had only 12,000 views, and relied entirely on an AI voiceover with stock footage. Creator credibility does not determine citation either. “A great creator who published a lot of videos on a topic did not get cited more than an AI content farm in Kenya,” Ben says.
What the engines are not measuring is influence. They are measuring deep contextual accuracy.
The Creative Profile of a Cited Video
The videos that do earn citations share a specific profile. Formal product reviews and physical demonstrations appear in 77% of cited videos, a 29-percentage-point lift over non-cited ones, according to the research. Videos that explicitly frame the problem being solved appear in 76% of cited videos, a 35-point lift. Clear product benefits appear in 90%. Mentioning specific ingredients or detailing scientific effects drives citation lifts of 30 to 34 percentage points.
Comparison videos are also common: comparisons between brands appear in 71% of cited content, while comparisons between different products appear in 68%. A practical, self-assured tone is present in 84% of cited videos. Exploratory calls to action, the “check it out” variety rather than a direct push to buy, add a 21-point lift.
Ben describes the format as resembling what YouTubers have long produced independently of brand campaigns: long, specific, comparative reviews that read more like research than advertising. “The engines love that,” he says, “because they’re trying to give the best answer they can to a specific person.”
Format matters, too. Cited videos are commonly eight minutes or longer. YouTube Shorts, Ben says, are cited at a fraction of the rate of long-form video.
Long-Form Creators Are Sitting on Uncaptured Revenue
Ben frames GEO not only as a discovery behavior change but as an uncaptured revenue opportunity for creators. He estimates $11 billion is shifting from traditional search engine optimization into AI engine optimization, and argues that creators, not platforms or brands, should capture most of it. “Creators should be able to monetize it,” he says. “You’re already earning it for them; you should be paid for it.”
He cites Erica Louie (@missejlouie), a creator who participated in Sundogs’ research launch and discovered her videos were being cited to answer product questions for brands she had covered. According to Ben, she recognized the implication immediately. “‘My videos have been cited, and they’re providing product discovery value,’” she said. “‘I need to get my clients to recognize: you’re getting value from this, you should pay me for it.’”
For creators already making long-form, fact-dense content, the research is largely a validation, and in Ben’s view, a new negotiating argument. For those focused on short-form, the adjustment is harder. Ben says he spoke with a company built around YouTube Shorts that believed high view counts would translate into citation volume. High volumes of Shorts do not produce citations at any comparable rate to long-form video, he says, and a shift in production approach is required.
Editorial Freedom Is the Price of Entry
For brands, the research creates two distinct pressures. The first is direct: creator content needs to be part of product discovery strategy. “Anyone who’s not already partnering with creators needs to, 100%,” Ben says.
The second is harder. Citation-eligible content requires creators to talk honestly, specifically, and comparatively about products, including competitors. “They need to be able to say, this shampoo is better for dry scalp than this one, but this one is better for XYZ,” Ben says. “And that makes brands very uncomfortable.”
Brands willing to grant that editorial freedom gain something they cannot produce internally, he argues: verifiable, fact-rich content that answer engines weight as a reliable source. Brands that insist on controlling the message are generating the roughly 10% of cited content attributable to brand-owned video. “Marketers are like, ‘I need to message to you,’” he says. “Don’t. You need to be useful to them with a density of information you’re not used to. Creators are a great way to get there.”
Pilots, Partnerships, and What CMOs Are Asking
Heading into the final months of 2026, GEO is Sundogs’ top priority. Ben says the company is developing pilots with brand clients to identify creators already being cited for their product categories, with the goal of either recruiting those creators or training existing partners to produce content that matches the citation profile. Sundogs is also working closely with Adelaide, which studies attention quality in advertising, and Circana, which holds sales lift data, to connect creative diagnostics to downstream commercial outcomes.
The question Ben hears most from CMOs is not whether AI is reshaping discovery but how quickly brands can adapt. “Every CMO is asking me about it,” he says, “and nobody really knows how to do it.” For creators already making long-form, comparative video, his read is that they are sitting on a revenue stream that brands are benefiting from but have not yet explicitly priced.
“Curiosity is a superpower,” Ben says. “It’s a great time to ask questions. You can get answers faster, and you can make things happen faster.”
Subscribe to Our Newsletter
Check Out Our Podcast