To say there’s a lot of content on YouTube is a huge understatement.
If you watched each of the site’s almost 4 billion videos back-to-back, it would take roughly 81,500 years to finish, according to media resource Photutorial
That’s not including all the new videos uploaded after you’d started watching, at a rate of six hours’ worth per second.
Pixability
For marketers, this makes it profoundly difficult to serve ads to the right YouTube audiences at scale, but Pixability believes its new AI-powered tools are up to the challenge.
On Monday, the YouTube-centric video advertising platform announced the launch of a product called GenAI Contextual Segments (CGS), with a goal to reach specific audiences across a wider range of YouTube content.
The product is intended to target niche audiences at scale, which, Chief Product Officer Jackie Swansburg Paulino acknowledged, feels like a contradiction.
“YouTube’s such a vast platform with so much content that there really is enough very specific content to your brand to reach scale,” she said. “But it’s very difficult to find.”
A fistful of data
Pixability is notable for getting in on the ground floor of YouTube back in 2010, when it was still considered an emerging platform.
In 2016, Pixability became one of a few companies granted access to the YouTube measurement platform (YMPT), which reportedly has 30,000 times more data than the public-facing Data API.
Now Pixability uses that data, as well as the historical data it’s collected for the past 14 years, to score and classify every monetized YouTube video based on brand safety and subject relevance.
The generative AI model sifts through 782 different attributes for each video it analyzes, including titles, tags and descriptions.
That data – or, more specifically, being able to act on that data – “is critical for us in terms of ensuring that we’re making our campaigns work, flourish and deliver against whatever the KPIs might be,” said Jeremy Cornfeldt, president of performance marketing agency Tinuiti.
According to Pixability CEO David George, since beta testing of the new YouTube targeting began in July, the team has seen performance improvements of up to 50%, especially with hard-to-find-audiences for brands in the financial services and pharmaceutical categories.
Staying on target
It’s natural that a makeup company, for example, would try to advertise on beauty-related videos. But with so many competitors vying for that same category, it’s hard (and often expensive) to break through.
The idea is similar to the basic promise of programmatic. ESPN is expensive, but you can reach the same visitors more efficiently across lesser-known sites.
“People who buy beauty products are into all sorts of other types of content,” said David George, CEO of Pixability. “We enable them to do that so you can actually get a broader scale, because you’re not just in a very narrow target of content.”
To do that, the Pixability One platform offers curated and custom audience segments, filtered via parameters like GARM risk tier, kid-friendliness, subscriber count and even creator generation (e.g., millennial, Gen Z, etc.).
The platform then uses an open-source language learning model (with the fairly inscrutable name “intfloat/multilingual-e5-large”), as well as dozens of other machine learning tools, to interpret user prompts and recommend relevant YouTube channels on a sliding scale of similarity.
The vehicle brand Jaguar, for instance, could type in “Jaguar car” and get results related to automobile enthusiasts, but not to the Jacksonville football team or rainforest predator of the same name.
It’s not just keyword matching, Paulino said. “It’s understanding the context, and that’s where those 782 pieces of data come in.”
After identifying the proper segments, users can copy the channel IDs all at once and plug them directly into Google’s DV360 or Google Ads for targeting.
Man may not be replaced
As in every community of content creators right now, machine learning and artificial intelligence tech is a controversial subject among YouTubers.
There are the obvious concerns over potential denial of job opportunities and low-quality AI content overwhelming the system. But many popular creators, such as SciShow’s Hank Green, have also expressed concerns about their content being scraped to train generative AI models. If they aren’t going to be compensated for the use of their work, they argue, then they should at least be able to opt out of being included.
Although the Pixability team is using generative AI components, they stress that they have no intention of using their data (which is provided to them directly by YouTube) for further content generation. Rather, the generative AI is meant to help better interpret user requests and create custom taxonomies for categorizing content.
And for marketers who are similarly squeamish about AI, Pixability soon hopes to better identify and sort the “synthetic” from human-made – something YouTube itself now requires accounts to disclose upon uploading, too.
“YouTube and other platforms have realized this is going to be a problem as well,” said Paulino.
