AI lets anyone produce dozens of pieces of content in a short time. But when every brand scales output at the same pace, feeds get more crowded and each brand becomes harder to see. Jung Bum-jin, CEO of Korean AI marketing company Pion Corporation, who presented a session at MAX SUMMIT 2026 titled “We make more with AI, but we’re less visible,” told Mobiinside the answer lies in treating AI as an operating system.
Jung points to two data points. A Microsoft study found Korea had the fastest AI user growth among surveyed countries in the second half of 2025. Meanwhile, Kapwing’s analysis of popular YouTube channels by country found Korean AI slop channels drew the most views. The country adopting AI fastest is also consuming the most low-quality AI content.
The problem, he says, isn’t using AI. It’s churning out plausible output without deciding who it’s for or why. Making creative used to be the bottleneck; now the hard part is finding a reason for customers to stop and look among all that content.
One good prompt won’t keep producing good ads. If product information is inaccurate, banned brand expressions aren’t documented or nobody owns final review, results wobble. The structure Jung describes includes:
“What matters isn’t what the AI made, but whether our organization can trust and use the result,” he said.
Brands submit a brief or product URL, and the system drafts selling points and a script, then generates voice, video and product cards and edits them in one flow. It produces 10-minute live videos with no filming and can create dozens to hundreds of variations per product. In a same-conditions comparison, AI-hosted lives outsold human hosts, and one broadcast reached No. 1 in the food category’s popular rankings.
A Mom’s Touch campaign in Japan, built from local culture, reached about 500,000 views in a month. Jung says the hard part of global campaigns isn’t one video but continuously adapting language, prices, offers, legal copy and on-screen talent for each market. AI keeps the core scenes and brand identity while swapping only what each market needs, so brands can test messages cheaply and invest in what works.
For Hyundai, Pion built a system that keeps headquarters’ visual and messaging standards while letting regional teams produce their own assets quickly. For a duty-free retailer, it generated fitting backgrounds for about 40,000 product images while preserving product shapes, designing the workflow so generation, review and approval ran inside the client’s existing sign-off process.
Pion’s new AdCatch tool compares competitor ads by appeal, target, call to action and visuals to find leads for the next brief. The goal, Jung says, isn’t copying but finding questions customers care about that nobody has answered yet. Discovery, production, execution and performance learning feeding back into discovery is what Pion calls its marketing agent OS.
Before choosing AI tools, check whether your brand standards, approval steps and performance feedback loop are documented. As Jung puts it, the best marketers won’t be the ones who generate the most output, but those who form the best hypotheses and learn fastest. For how platforms are pushing back on low-quality AI content, see LinkedIn’s AI Slop Detection Hits 94% Precision — How User Reports Train the System, and for embedding AI into agency workflows, read Building an In-House AI Platform at an Ad Agency: How Daehong’s D-AISY Automated TV Schedule Research.
Low-quality content mass-produced with AI without a clear audience or purpose. Kapwing found Korean AI slop channels drew the most views among countries analyzed.
Accurate product and customer data, brand expression standards, production and approval workflows, distribution channels and a loop that feeds performance back into production.
According to Pion Corporation, AI-hosted lives on Naver Shopping outsold human-hosted lives under the same conditions.
To apply what you just read to your own site, start with a free audit of where things are now.
A strategist replies within 24 hours on business days.
Buzzvil's BD strategy team has zero developers but runs 32 AI workflow automations — from 7-minute revenue drop analysis to 90% fewer manual steps. How they scaled from personal hacks to team systems.
Meta’s Muse Wonder and Muse Wonder Full add Active Memory and Active Focus so its AI agent can handle tasks that take days. But a Toronto user’s Marketplace mishap shows the risks of agentic AI.
Pew cloned thousands of panelists as AI digital twins and found a 12.4-point average error. Five ways synthetic respondents fail — and where marketers can still use them in research.