Author: admin

Key stat: The US accounts for just 10% of TikTok users but generates 41% of the platform’s ad revenue worldwide, according to a July forecast from EMARKETER.

Beyond the chart:

• TikTok’s US audience is highly responsive to advertising. In fact, 83% of weekly US TikTok users ages 13 and older have taken action after seeing an ad on the platform, including making a purchase (43%), according to a December Edison Research report.
• Meanwhile, the platform’s US user base is expanding beyond its Gen Z core. TikTok adoption among users 45 and older has grown 1,200% between 2019 and 2025, according to an August survey from CivicScience.

Use this chart: Drop this into any deck questioning TikTok’s role in your media mix. Even with the app’s future uncertain, this chart makes the case for why the US market still commands attention and budget. Use it to defend domestic TikTok investment despite regulatory headwinds.

Related EMARKETER reports:

Methodology: Estimates are based on the analysis of various elements related to the ad spending market, including macro-level economic conditions, historical trends of the advertising market, historical trends of each medium in relation to other media, reported revenues of major ad publishers, estimates from other research firms, data from benchmark sources, consumer media consumption trends, consumer device usage trends, and EMARKETER interviews with executives at ad agencies, brands, media publishers, and other industry leaders.

AI companies are prioritizing flash over substance, says Surge AI’s CEO.

“I’m worried that instead of building AI that will actually advance us as a species, curing cancer, solving poverty, understanding universal, all these big grand questions, we are optimizing for AI slop instead,” Edwin Chen said in an episode of “Lenny’s” podcast published on Sunday.

“We’re basically teaching our models to chase dopamine instead of truth,” he added.

Chen founded AI training startup Surge in 2020 after working at Twitter, Google, and Meta. Surge runs the gig platform Data Annotation, which says it pays one million freelancers to train AI models. Surge competes with data labeling startups like Scale AI and Mercor and counts Anthropic as a customer.

On Sunday’s podcast, Chen said that companies are prioritizing AI slop because of industry leaderboards.

“Right now, the industry is played by these terrible leaderboards like LMArena,” he said, referring to a popular online leaderboard where people can vote on which AI response is better.

“They’re not carefully reading or fact-checking,” he said. “They’re skimming these responses for two seconds and picking whatever looks flashiest.”

He added: “It’s literally optimizing your models for the types of people who buy tabloids at the grocery store.”

Still, the Surge CEO said that AI labs have to pay attention to these leaderboards because they can be asked about their rankings during sales meetings.

Like Chen, research scientists have criticized benchmarks for overvaluing superficial traits.

In a March blog post, Dean Valentine, the cofounder and CEO of AI security startup ZeroPath, said that “Recent AI model progress feels mostly like bullshit.”

Valentine said that he and his team had been evaluating the performance of different models claiming to have “some sort of improvement” since the release of Anthropic’s 3.5 Sonnet in June 2024. None of the new models his team tried had made a “significant difference” in his company’s internal benchmarks or in developers’ abilities to find new bugs, he said.

They might have been “more fun to talk to,” but they were “not reflective of economic usefulness or generality.”

In a February paper titled “Can we trust AI Benchmarks?” researchers at the European Commission’s Joint Research Center concluded that major issues exist in today’s evaluation approach.

The researchers said benchmarking is “fundamentally shaped by cultural, commercial and competitive dynamics that often prioritize state-of-the-art performance at the expense of broader societal concerns.”

Companies have also come under fire for “gaming” these benchmarks.

In April, Meta released two new models in its Llama family that it said delivered “better results” than comparably sized models from Google and French AI lab Mistral. It then faced accusations that it had gamed a benchmark.

LMArena said that Meta “should have made it clearer” that it had submitted a version of Llama 4 Maverick that had been “customized” to perform better for its testing format.

“Meta’s interpretation of our policy did not match what we expect from model providers,” LMArena said in an X post.

There were differing opinions over to what extent AI and agentic technologies are already part of programmatic workflows (and there seemed to be no consensus definition for what qualifies as agentic). What the agency executives did seem to agree on, though, was that AI agents are best kept away from the actual point of transaction between media buyers and sellers. 

Instead they should be relegated to pre- and post-transaction tasks, like helping to plan campaign parameters and to organize post-campaign performance data. But even that level of AI involvement will require enhanced training of agency employees not just on AI tools (and their limitations) but also on the fundamentals of programmatic advertising workflows for the human employees to better supervise their AI counterparts.

The following has been edited for space and clarity.

Setting the stage

“I think we all know that machine learning AI … from a buying standpoint is already very integrated. But I think what is new is the gen AI, right? And that’s what’s exciting. The agent idea, and this idea of ideation creating autonomy is where it’s evolving. But I don’t think we all know what we’re talking about, just because it is already integrated within buying across Meta, Google, etc. But from a gen AI aspect, I think that’s what’s really exciting. Where we’ve seen efficiencies in terms of our AI that we’re using is helping with media mix models planning. That really helps with efficiencies, as well as automation in general, in terms of administrative tasks, analytics, and aggregating data.”

AI used as a smoke screen

“Every agency has some level of machine learning being used in their algorithms and optimizations. We’re not afraid of that, but it isn’t very transparent by nature. One of the challenges we face in this industry is now we’re being fed AI as a cover for more opaqueness in pricing, more opaqueness in performance optimizations. Whether it’s [The Trade Desk’s] Kokai’s new algorithms all trying to do stuff for you and saying ‘Just trust us.’ That, I think scares us, because we know it’s not LLMs, but we also know they’re using it as a smoke screen, and we’re not all equipped to poke holes in their theories and push back. I think that might be where a lot of the fear is, because we just see more opaqueness and us having less and less control.”

“We have not gotten there yet, but one of the things that we’re working on on our agency is, how do we use the agentic AI to help us explain what AI is doing in our campaigns? But how do we get it to explain stuff? Because if a person makes an optimization, you can say, Why did you make that optimization? What do you think it was going to do? And what did it do? AI doesn’t do that — it just makes the optimization and has a result or doesn’t have a result.”

Careful with that agent, Eugene

“Agents are built for fetching and gathering. That’s what they do good. They don’t think. You just asked why — they don’t do that. They’re all probabilistic anyway. Technically agents are [Las] Vegas on all the things you’re looking to track, and it tries to get to a better result based on what it’s seen in the past. So if you’re doing something new right now with the agent, don’t do that. They’re not really good at that yet, because they’re working on odds. I feel like we’re probably three or four years away from trusting any of these things.”

“AI can be really good for things with defined parameters. But as an industry, we’re kind of lumping in a lot of AI together. For, say, a programmatic media plan, it’s going to potentially hallucinate because things like outdated information aren’t being taken down from the web, and it’s taking that into consideration. If you just let it to its own devices, you’re not going to get best results. But if you use a very clean data set to find parameters of how things should be evaluated, I think it can be very beneficial.”

What’s to come? 

“From the strategist side, some of the programmatic partners actually have tools that will build your whole media plan, right? But what I haven’t seen is the integration being there. I think their dream of it is, click the one button that builds the plan for you and optimizes the plan. I don’t think that’s upon us yet, but I think that’s what may be coming for us.”

“All of us are being told that’s what people want — let’s do that. But then you have legal compliance, and your finance team. All it takes is for the AI to hallucinate one zero, and you’re in big trouble. And that’s one mistake that’s going to cost you potentially a client, or could cost you your job. Because LLMs will never NOT hallucinate — their underlying architecture design has that flaw. I don’t know how many trillions of dollars have been invested in this, but they haven’t fixed this problem. They’re like a very smart intern — we’re going to give it information. It’s going to be able to do a lot of tasks that are menial and take a lot of time, and do it fast. But before we do action, we’re going to put it through some of the older models, or sometimes trust the older models to do the actual heavy lifting, because they have experience. We know they work, and we know they can work and they can do it better.”

“You were saying smart intern — it’s a six year old. Think of it as your six year old child who will give you an answer whether it’s right or not, because they want to prove to you that they think what you said is important. That’s AI right now. It’ll be different later, but right now, it’s a six year old.”

“We’re using AI in pre- and in post-production. Your whole planning team is pre-production. Your analytics team is post production, giving reports back. That is a lot of the day to day job, so that can save us a ton of time. The actual production part is actually a small percentage of our time. We set up the campaign, we run it, the optimizations come from post analytics, and then we go and make the changes. A lot of the lead up is a lot of time that AI can save us.”

Accountability

“Accountability comes with management not allowing people to use, ‘Well, that’s what the AI gave me’ as an excuse. They have to be accountable to their work no matter how it’s been generated. And I think that’s how we have to move forward. Because if we allow everyone to use that as an excuse, that we’re going tons of errors.”

“My concern is on [the 25-year-old media planner’s] over-reliance on AI for doing a lot of these things. Is that going to impact overall foundational knowledge that allows them to make these strategic decisions creatively and out of the box? Because they’ve been brought up [thinking] ‘Well, I can just do all of these things by throwing it in here.’ They don’t understand the why behind it, and then it snowballs 20 years from now.”

How to train your AI

“How would you train the AI better? The best way I can describe it is, it’s like good barbecue — low and slow. Use older data that’s very specific. It probably comes in slower than what you’re getting from signal-based data, or real-time data and all the rest of that. Not necessarily because it’s better, but because it has less garbage in it.”

Color by numbers

Does the need to market as aggressively as possible during the holidays offer an opportunity for retail media networks and challenger social platforms to capture more holiday and 2026 ad dollars? The answer is yes — if they can close the measurement and trust gap, according to new research from Kantar. Some supporting stats: 

• Retailers see building in-store/omnichannel capabilities (82%) and off-site data monetization (45%) as key ways to compete with Amazon. 
• 87% of brands would be more likely to trust and invest in retail media networks accredited for standardized measurement, but only 24% of retailers are fully aligned with such standards. 
• 67% of brands are ready to invest more if in-store impact can be proved. 
• 80% of brands say unified in-store/online measurement is essential or important.  

Takeoff & landing

• The closure of Omnicom’s acquisition of Interpublic Group resulted into the planned layoffs of about 4,000 staffers, leading to a wave of publicly aired bitter feelings over social media. It also resulted in the shuttering of IPG’s Magna media investment and business intelligence unit, along with the departure of Eileen Kiernan, the highest ranking media executive in IPG’s fold. 
• Havas made two different acquisitions last week: U.K. experiential agency Bearded Kitten, which will be folded into Havas Play, and, earlier in the week Unnest, a French data consulting and engineering firm, to support Havas Media Network’s tech and data abilities. It declined to identify purchase prices for either. 
• Personnel moves: Dentsu named Kara Osborne Gladwell its global product architect officer for Media, and brought back Tia Castagno to become Global Innovation President. Both are newly created global positions and report to Will Swayne, global practice president, Media … Monks hired Thiago Correa to be svp of media for the EMEA region, coming from Publicis where was global client lead for H&M  … Creator marketing agency Influencer hired Ryan Fitzpatrick as CFO, coming over from a similar post at VaynerX.

Direct quote

“This merger is a reminder that scale doesn’t fix fragmentation. AI only works when the underlying customer data is unified, governed, and understood. The organizations that win in the next phase of marketing won’t be the ones with the most tools — they’ll be the ones with the clearest picture of their customers.”

—Tony Owens, CEO of customer data cloud Amperity, on Omnicom’s acquisition of IPG. 

Speed reading