BRANDS
AI feature: Auto-tagging on YouTube
Stratechery’s Ben Thompson interviewed Google’s YouTube CEO Neal Mohan at the “Made on YouTube 2025” event in New York City this week.
The interview included discussion of a new AI-enabled feature called “auto-tagging” (read TechCrunch) which speaks to the future of video advertising and virtual placement.
In short, tags on video create ad placements – and sure, you can call them shopping links, too.
From the YouTube blog on Monday explaining the auto-tagging feature:
“We know tagging products can be time-consuming, so to make the experience better for creators, we’re leaning on an AI-powered system to identify the optimal moment a product is mentioned and automatically display the product tag at that time, capturing viewer interest when it’s highest. We’ll also begin testing the ability to automatically identify and tag all eligible products mentioned in your video later this year.”
Recall that Mr. Mohan came to Google with the Doubleclick acquisition in 2007 and eventually moved to YouTube (where content is king) from Google’s ad tech in 2015. Google and Mohan know that an effective monetization strategy (for creators and for Google) is what powers the content.
(From the Stratechery interview this week – lightly edited)
Ben Thompson, Stratechery:
“The one [feature] that jumped out to me though was auto-tagging, where you can tag an item in a video that a creator’s trying to sell, instead of them having to go manually and figure it out and add the appropriate link, they can click and do whatever.
This has long struck me as one of the largest monetization opportunities for both you and Meta. How long until you tag everything? I can imagine a system—like your [music] copyright system— where instead of videos getting taken down because there are copyrighted songs in there, the copyright owner just gets monetized through doing that. You could imagine a world where actually everything in a video is tagged and people can come along and see how many items they have tagged across YouTube and, ‘Do you want to bid on having links put in there?’ Is that on the roadmap?”
Neal Mohan, CEO, YouTube:
“That specific idea, we’re not quite there yet. Right now, as you saw, we’re taking some of the first steps towards this. But, your product vision is definitely on the right track there…
But implicit in your product idea is a question about what actually happens on YouTube, and before I answer the specific question around auto-tagging, I just want to say that the lens through which we look at shopping and commerce on YouTube is not necessarily just about a business opportunity for YouTube, and also therefore for our creators — of course it is that — but actually from the viewer’s perspective.
What I mean by that is just like we would go to the mall with our friends and family on a Saturday afternoon, and you didn’t have any specific shopping list in mind, it was entertainment just as much as it was shopping. That user journey exists on YouTube tens of millions of times a day and so it’s really about serving that particular use case. That’s where auto-tagging comes in: make it so that videos have information about products and one of the biggest challenges — and we’ve been trying this by the way, for three or four years without auto-tagging — one of the biggest challenges we heard from creators is that it’s kind of a pain to actually do that manually in videos.
One way to scale it now that the AI has gotten so good that we can do it at pretty high precision is through auto-tagging, and so that’s why we’re so excited about it. You saw the excitement in the room from the creators around it.
They think it’s a big opportunity, but none of those creators are going to be doing it just for the sake of selling products because they have to be truly authentic to their fans. So I bet you when you talk to creators about shopping, they think about it more as a means to connect with fans first and foremost, and then a commerce opportunity.”
Hear the 56-minute interview on Stratechery. (September 17 – subscription)
From tipsheet: AI-enabled auto-tag ad network, here we come.
This feels very early days at this stage as you can see from Thompson’s thinking versus today’s auto-tag reality at YouTube. Nevertheless, the opportunity for dynamic placements within video generated by the creator economy will only grow.
Auto-tagging is facilitating another type of ad placement, if you will. In a similar vein, see AI ad tech startups like Rembrand.
Mr. Mohan also made note of creators’ brand partnerships today that are happening outside of payments by YouTube to creators. Google wants to help facilitate the brand partnerships (they announced ‘swappable slots’) – or put another way – Google wants a piece of that pie.
BRANDS
Addressing AI Max issues
In a Digiday article titled, “Pitch deck: How Google is responding to advertisers’ concerns about AI Max,” reporter Krystal Scanlon took readers a 26-page update by Google on how it’s addressing marketer gripes about its search ad buying system, AI Max.
Among the updates, Google cautioned against using very small ($50/day or less) budgets. The company also made clear that AI generated text was sourced according to client specs. And Google is planning to put shopping and text ads below the first response in AI Mode soon.
Read more in Digiday. (September 17)
BRANDS
Unlocking news: suitability, violations and blocks
Touting its AI-driven solutions, ad verification firm DoubleVerify released a new case study with its client, Vodaphone, which spoke to “suitability” for the brand on news sites.
DoubleVerify said that “suitability” improved for Vodaphone in comparison to 2024 benchmarks:
“33% lower brand suitability violation rate (with a 46% reduction between Q1 2024 and Q1 2025).
45% reduction in unsuitable category violations (with a 37% reduction between Q1 2024 and Q1 2025).
48% drop in keyword violations year-over-year (with a 48% reduction between Q1 2024 and Q1 2025).
20% lower overall block rate (with a 41% reduction between Q1 2024 and Q1 2025).”
Read the press release. (September 17)
From tipsheet: Marketers want to buy news but don’t want their brand to be associated with the wrong content. Publishers want marketers’ budgets. DoubleVerify stands in between with AI-enabled discovery tools, i.e. verification.
LLMs & CHATBOTS
Developments
- OpenAI outperforms humans and Google at the world’s top collegiate programming contest (September 17) – The Decoder
- xAI’s Colossus 2 – First Gigawatt Datacenter In The World, Unique RL Methodology, Capital Raise (September 16) – SemiAnalysis
- Anthropic irks White House with limits on models’ use (September 17) – Semafor
AGENCIES
Laura Desmond brings together industry
At a Smartly Advance event in NYC, former Starcom MediaVest CEO Laura Desmond, who is now the CEO of AI ad platform Smartly, gathered an impressive list of speakers for her company’s conference yesterday.
In one panel discussion led by Ms. Desmond, executives from TikTok, Google’s YouTube, Meta, Amazon and Pinterest addressed AI in advertising and responsible innovation.
According to ClickZ, which covered the event, Pinterest’s CRO Bill Watkins explained that his company’s AI was powering “hyper-relevant visual search and predictive trends”:
“[Mr. Watkins] cited a moment when Pinterest spotted an engagement trend months before it broke into mainstream headlines, illustrating AI’s role in shaping cultural anticipation. On the commerce side, AI-powered decisioning tools like collages and shopping lists are designed to give consumers confidence in their choices and advertisers confidence that their campaigns align with real intent.”
Read more in ClickZ. (September 17)
From tipsheet: Mr. Watkins AI anecdote sounded like a good pitch to AI model companies looking to crawl large social websites.
SELL-SIDE
New AI training deals for Reddit?
According to anonymous sources, Bloomberg said yesterday that social publisher Reddit wanted more value in future AI deals with companies such as Google and OpenAI.
Reddit is also reportedly proposing that Google somehow refers users back to Reddit in a way which would help create the content Google needs for its AI models to train.
It’s just rumblings at this point as Bloomberg explained that no changes to existing agreements appear imminent:
“Reddit’s first data agreements with Google and OpenAI remain active. In January 2024, Reddit reached licensing agreements, including those two deals, that would generate $203 million in contract value with terms ranging from two to three years, according to a document Reddit filed in advance of its initial public offering last year.”
Read more in Bloomberg. (subscription – September 17)
From tipsheet: If true, Reddit is saying that it no longer wants a straight up licensing deal where the AI crawler comes and goes over a specified time period, etc. Perhaps it will be pay-per-inference, pay per search going forward or a combination? How those types of deals are monitored would be fascinating to see.
The implications could be far reaching for open web publishers as Reddit potentially sets the market. Reddit wants those big AI dollars.
TECH
Agentic AI meets Amazon Ads
Amazon released new AI-enabled ad tools yesterday for ad creative development and media/marketing planning – through its AI-enabled “Seller Assistant” – for Amazon sellers in the U.S. Read the release. (September 17)
The Wall Street Journal noted the trend:
“The move is the latest development in a race among ad sellers to use AI to lure new business, particularly among small and medium-size companies that can’t afford to hire ad agencies or create their own campaigns. NBCUniversal parent Comcast, for example, next week intends to introduce an initial version of an AI tool to generate commercials to run on streaming TV.”
The WSJ also noted that the tools will allow Amazon sellers to create CTV ads and take advantage of Amazon Ads enormous CTV ad footprint. Read more. (subscription)
From the Amazon press release:
On creative: “By analyzing a seller’s products alongside Amazon’s shopping signals, Creative Studio’s AI feature generates tailored ad concepts and thoroughly explains its reasoning, giving sellers complete control while revealing new insights.”
On Seller Assistant and ads: “Seller Assistant will also be able to analyze sales patterns and customer behavior to suggest promising new product categories to explore, recommend optimized advertising and marketing strategies, and provide tailored insights to guide international expansion. In addition to answering questions from sellers, Seller Assistant will proactively develop comprehensive growth plans for sellers to review and take action for sellers when they want it to.”
Amazon called this “The future of selling with Agentic AI” and said:
“This evolution will help sellers spend more time focusing on product innovation and customer relationships while Seller Assistant handles making selling in Amazon’s store more efficient, strategic, and successful.”
From tipsheet: Amazon is offering everything an SMB —and large businesses— needs in one place to reach the consumer.
Also, AI is enabling ad tech table stakes (creative/planning/buying) for all the walled gardens.
TECH
The Trade Desk’s AI ad platform pruned
“Scoop: The Trade Desk is rolling out a redesign of Kokai, its programmatic buying platform, that will partially sunset one of its most recognizable—and debated—features: the periodic table. The visual interface, long polarizing among buyers, weighed on TTD’s revenue too.”
More: The Trade Desk Partially Sunsets Controversial Kokai Feature in Latest Redesign – Adweek (September 17)
TECH
Amazon DSP knows purchase intent
Unrelated to yesterday’s Amazon announcement, Mobile Dev Memo analyst Eric Seufert published a blog post regarding Amazon’s AI-enabled DSP.
Mr. Seufert saw a key point of differentiation (transaction data!) with companies like Meta which shuttered its audience network in 2020:
“The Facebook audience network for web was wholly dependent on whatever fractured picture of identity could be gleaned through cookies to depict a consumer’s purchasing patterns. But Amazon has a complete, unobstructed view of those patterns: the only matching that takes place is whether those behavioral profiles fit into a broader audience segment that can be predicted from the data in a bid request.
This form of targeting is far more durable than what’s possible with cookies. And what’s more, because it generates that data itself, Amazon can also directly facilitate measurement in the Amazon Marketing Cloud. For this reason, Amazon is not just an advertising behemoth as a function of its first-party advertising inventory, but also because of what it can accomplish across third party sites through the Amazon DSP.”
Read more on Mobile Dev Memo. (September 17 – subscription)
Related: Advertisers Can Now Buy SiriusXM Audio Ads Directly Through Amazon DSP (September 16) – Adweek (subscription)
TECH
Opinion: The marketing data engineer
In an op-ed in AdExchanger, Incrmntal CEO Maor Sadra makes the case that as AI and automation take the wheel, marketers need a new direct report to overcome “bad data” being input into their AI-enabled ad systems.
Mr. Sadra wrote:
“This is where a new kind of capability is needed: the marketing data engineer. Not a media buyer, not an analyst buried in spreadsheets, but someone who bridges the technical understanding of data architecture with the strategic awareness of how media is planned and measured. A role tasked with ensuring that campaign inputs are complete, structured and consistent, before optimization tools and machine learning models are allowed anywhere near them.”
Read more in AdExchanger. (September 17)
Mr. Sadra’s company, Incrmntal, provides incrementality analytics which helps marketers determine whether an ad campaign actually caused lift for their campaigns’ chosen outcomes.
PROMPT
What are factors that make a publisher website valuable to AI model companies?
Response from xAI’s grok:
Publisher websites—such as news outlets, academic journals, blogs, and media platforms—provide a rich source of text-based data for training large language models (LLMs) and generative AI systems. AI companies like OpenAI, Google, and Anthropic value this content because it fuels model accuracy, diversity, and real-world applicability…
CORRECTION
-
Roku’s Head of Ad Innovation is Peter Hamilton. Mr. Hamilton clarified his comments from a WSJ article on LinkedIn Tuesday.
MORE
- Zeta Announces Generative Engine Optimization Solution to Help Brands Lead in the Post-Search Era (September 17) – press release
- What do tomorrow’s CMOs think about the future of websites? These 5 insights may surprise you (September 17) – Knotch
- AI advertising reality check: Where marketers are really seeing ROI (September 17) – IAB on Ad Age
- Carvana gives keys to Shaq for AI-powered brand takeover campaign (September 17) – Marketing Dive