*This is the eighth instalment of Swipe, a series in which we examine South Korean AI apps and services up close. This week: Zeta, the country's leading AI character-chat platform, made by ScatterLab.*

What it is

Zeta is an AI character-chat platform operated by ScatterLab, the company behind Iruda, an earlier Korean-language conversational AI. After a closed beta in the second half of 2023, it launched in open beta in April 2024. Users design their own characters — setting names, personalities, speech styles and fictional universes — and can also chat with characters created by others, collaboratively building stories as they go. On 28th of this month, ScatterLab launched a "Visual Novel Mode", extending the service from text chat into animated, image-driven storytelling. The company says it operates in South Korea, Japan and the United States, and as of September 2026 has accumulated more than 13 million registered users worldwide.

Core features

Zeta's central conceit is that conversation *is* the story. In practice, users browse hashtag-sorted categories — romance, fantasy, BL (boys' love) and so on — find a character or plot that appeals to them, and begin chatting. The interface includes a status panel displaying narrative details such as date, time, location and emotional state; branching-choice buttons that let users steer the plot; and a "lore book" feature that anchors keywords to a character's fictional world. Beyond a free tier running on a basic model, paid tiers are available: a free option supported by advertising, an ad-free monthly subscription called "Zeta Pass", and an in-app currency system.

The newly launched Visual Novel Mode is the headline addition. Characters' facial expressions, gestures and backgrounds update in real time in response to the conversation; when a new character enters the story, the AI generates their image on the spot. Game-like mechanics — a status panel, branching choices and dice-roll events — have also been layered in. Creators can register costume designs for their characters, which the AI applies according to the story's context, and they can sell those designs to other users, opening a monetisation channel within the platform.

Performance: second in users, first in time spent

Data published by research firm WiseApp·Retail in February 2026 showed that Zeta's total monthly usage time reached 113.41 million hours — 2.2 times ChatGPT's 50.47 million hours. By user numbers, the comparison runs the other way: ChatGPT had 22.93 million monthly users against Zeta's 4.02 million, giving the American chatbot a commanding lead on reach. What the figures reveal, however, is that once users open Zeta, they stay far longer. ScatterLab says the daily average session runs to two hours and 40 minutes. Around 87% of users are reported to be in their teens or twenties.

Compared with Crack — the closest Korean rival in the character-chat category, made by Wrtn Technologies — Zeta's lead is substantial. The same WiseApp survey put Crack's monthly usage time at 9.88 million hours. Business news outlet Dealsite noted that Crack's monthly active users are roughly a quarter of Zeta's, and its total usage time less than one-tenth. These figures should not, however, be read as a comprehensive picture of the overall market.

Strengths and weaknesses

Zeta's key advantage is design for immersion. Because users continuously generate new characters and fictional universes, the content supply is effectively limitless. There are no per-message charges, so conversations can run on without financial friction. The Visual Novel Mode adds cinematic visual production, blurring the lines between chat app, video game and webtoon.

The weakness flows directly from that same strength. According to mobile analytics firm Mobile Index, users aged 19 and under accounted for 58.3% of Zeta's monthly active users last month. That concentration of young users has amplified concerns about excessive use, emotional dependency and exposure to inappropriate content. The National Assembly's Science, ICT, Broadcasting and Communications Committee has summoned ScatterLab's chief executive, Kim Jong-yoon, to testify at parliamentary audit hearings scheduled for the 12th of next month. Representative Noh Jong-myeon of the Democratic Party of Korea reported that, when he accessed the service with his age set to 15, he encountered numerous sexually suggestive and violent expressions. A July report from the National Assembly Research Service concluded that existing content-regulation frameworks — designed around harmful publications and broadcast material — are structurally ill-equipped to address the risks specific to AI chatbots, particularly the kind of emotional dependency that can develop through ongoing relational interaction.

ScatterLab's current response is a two-tier system: a default "Safe Mode" that restricts the generation of content classified as harmful to minors, and an "Unlimited Mode" — permitting more explicit expression — accessible only to adults who have completed identity verification. Its rival Wrtn has announced that it will cap daily use of Crack at three hours and monthly in-app spending at 100,000 won (roughly $74), with parental consent required for any excess. Whether Zeta will introduce comparable usage or spending limits remains an open question. User reviews on app stores also flag intrusive advertising and lapses in characters' conversational memory as persistent irritants.

Verdict

Zeta offers the clearest answer yet to a question this series posed when examining Crack: where is the real battleground in South Korea's AI chatbot market? The answer, it turns out, is not information retrieval but storytelling and emotional connection — and the scale of appetite for both is striking. Yet the same immersiveness that drives Zeta's extraordinary engagement metrics now puts it on a collision course with concerns about child safety, just as parliamentary hearings loom. How the company redesigns its service from the ground up — not just its content filters — will define the next chapter.

★★★☆☆ (3.5/5.0)

In a line "It has proved it can hold users twice as long as ChatGPT. Now it must prove it can handle that power responsibly."