Explainer: How ChatGPT is opening up Korea's 'Galapagos-like' digital enclave

An illustration of app logos of ChatGPT, Naver, Daum, Kakao, Papago, A. and wrtn [KOREA JOONGANG DAILY]

 
Before the arrival of ChatGPT, it was user-driven forums or blogs run by domestic portals Naver or Daum where Koreans looked for information about wide range of issues — from the trivial, like the best restaurant in a neighborhood, to the significant, including facts about inter-Korean relations.
 
Koreans' go-to online translator used to be Papago, a machine translation service developed by Naver, prior to the rise of OpenAI's chatbot.
 
The dominance of homegrown internet services earned the country a nickname, “Galapagos,” compared to other Google-dominated markets, but more users are now turning toward relatively new services like ChatGPT for their queries, due to chatbots' reputation for providing more precise answers than traditional keyword-based searches can often surface.
 
As a result, ChatGPT's monthly active users (MAU) have shot up some 6,180 percent to 3.14 million between May 2023 and January 2025, whereas Naver and KaKao, while still dominating in search and chat, are losing screen time, which is leading them to drive forward with their AI businesses to lock in users.
 
This edition of Explainer explores Korea's shifting online culture and its impact on the tech industry. 
 

Related Article

• OpenAI CEO holds AI summit with chiefs of Samsung, SoftBank in Seoul

• Naver internally releases upgraded AI model HyperClova X

• OpenAI, Kakao announce co-development of localized AI product

Q. Just how dominant are Naver and Kakao in Korea?

 
A. The two Korean platforms stand out in a search market of which Google holds 89 percent, according to data from the web analytics platform StatCounter. Combined, Naver and Kakao made up a comparable share of Korea's market in 2015.
 
That dominance over search has retracted in 2024, but the two companies still see a fair amount of use across their various services. Monthly active users (MAU) of Naver's apps for Android and iOS numbered 43.92 million in January 2025 — meaning a vast majority of Korea's 51.75 million population still uses the app.
 

 
Although Naver and Kakao maintain significant user bases, the monthly average screen time for Naver, has fallen compared to the past, according to market tracker Mobile Index. Naver's averaged 472.62 minutes, a 11.14 percent decrease from November 2022, when ChatGPT took the AI scene by storm.
 
Kakao has attempted to diversify beyond KakaoTalk, making various efforts to turn Korea's leading messaging service into a broader social media platform. Those haven't yet paid off: KakaoTalk had 45.69 million users in January 2025, but its average screen time shrank 9.42 percent to 675.98 minutes per between January 2025 and November 2022. 
 
On the other hand, the MAUs of ChatGPT and wrtn, a Korea-tailored AI platform that incorporates a number of models, including ChatGPT4, rose 6,180 percent and 45,600 percent between 2023 and 2025. Average screen time also surged 703.9 percent and 1,478 percent, respectively, during the same period to reach 63.75 and 297.67 minutes per user. 
 
 
Why aren't Korean companies churning out generative AI? 


Well, they have — but their entry was rather late, and the capabilities of their models are quite not up to the par of the world's most successful competitors. 
 
However, Naver and Kakao are positioning 2025 as the year of their AI transformation, promising to release new AI-driven services to lock in users.
 
What initially set these two companies apart was their ability to provide an online archive tailored specifically to the Korean language — a challenge that initially stalled Google Search, which struggled to process Korean data effectively.
 
With user data they’ve accumulated, they expanded their offerings to include community-building services and diversified into areas such as advertising, e-commerce, entertainment, fintech and mobility.
 
AI chatbots are now excelling at drafting emails, reports and essays, practicing conversations, generating ideas, planning content and even providing coding suggestions.
 
Meanwhile, enterprises are integrating ChatGPT-like models into customer service to automate basic inquiries and streamline data retrieval and analysis.
 
However, Naver and Kakao now face a more complex landscape, owing not only the rise of DeepSeek’s R1, OpenAI’s o3 and DeepResearch, but also as the search engine market rapidly evolves into one driven by AI.
 
The two companies are taking opposite approaches. Naver is expanding outward, developing and marketing proprietary technology, while Kakao is integrating external models into existing services.
 

gg

 
Naver now faces the challenge of expanding its search capabilities beyond its own dataset, requiring high-quality inference models that can reason, summarize and provide context-aware answers rather than simply retrieving indexed content.
 
In contrast, Kakao is folding AI into customer interactions, financial transactions and recommendations across its services. DeepSeek and OpenAI are, therefore, more feasible partners for Kakao. Instead of focusing on research, which is financially demanding due to the high monetary and computational demands of training AI models, it can focus on deployment.
 
Kakao’s partnership with OpenAI formalizes this approach, as the Korean giant integrates the U.S. organization's API into its services.
 

Kakao CEO Chung Shin-a, left, and OpenAI founder Sam Altman pose for a photo at the Kakao's press event to announce the Korean company's partnership with OpenAI held in central Seoul on Feb. 4. [NEWS1]

 
However, even Big Tech firms have yet to establish a clear monetization strategy for AI.
 
Naver and Kakao are pushing to integrate AI-powered search and recommendations into e-commerce as a revenue driver, as it remains the most viable way to profit from AI applications — at least, for now. 
 


What changes are coming?


Naver, Korea’s largest portal site, made the surprise announcement of the return of its founder, Lee Hae-jin, as an inside director via electronic disclosure earlier this month. Lee’s reinstatement, which will be finalized by vote at the shareholder meeting in March, is significant in the sense that the entrepreneur has largely remained absent from domestic operations for seven years since stepping down as board chair in 2017.
 
His comeback signals the reclusive entrepreneur’s determination to drive forward with the portal site’s AI business.
 

Naver founder Lee Hae-jin, far left, and CEO Choi Soo-yeon, far right, pose with Nvidia CEO Jensen Huang at Nvidia’s headquarters in California on June 25, 2024. [NAVER]

 
Naver is expected to release multiple AI-driven services this year. By the first half of 2025, Naver will launch an AI briefing feature, similar to Google’s AI Overviews, that provides AI-generated summaries at the top of search results. Naver’s e-commerce platform, Naver Plus Store, will deploy AI to make tailored product recommendations in a stand-alone app.
 
An AI-driven advertising platform for e-commerce vendors, in which AI calculates a product's advertising costs, target consumer base and key words to offer tailored ads to the users, will also roll out.
 
Kakao, the operator of the nation’s dominant messaging service, highlighted its AI road map when announcing its partnership with OpenAI. 
 
Kakao plans to launch a chatbot, Kanana, leveraging OpenAI's API services, as a closed beta in the first half of this year. Kanana is currently undergoing internal testing.
 
Kakao sees Kanana’s competitive edge in its potential for use in group chats, unlike other similar services that focus on one-on-one conversations, making it particularly valuable for enterprise purposes.
 
Kakao and OpenAI plan to co-develop an agent that will make recommendations and assist users through services such as KakaoTalk, Kakao Map, and Kakao T.
 
Similar to Naver’s move to integrating AI into e-commerce, Kakao is set to release a feature, AI Mate, that will offer context-tailored recommendations for shopping and navigation when selecting products or destinations.
 

A screen capture of Naver's translation service, Papago [SCREEN CAPTURE]

 
What about translation-specific apps, such as Naver’s Papago? Will they still be in demand when generative AI appears to do a better job?


There will likely still be demand for purpose-specific AI in the business-to-business (B2B) space. Take, for example, the success of the German company DeepL.
 
DeepL’s success is unusual because when it came onto the scene, the translation market had already long been dominated by Google Translate and, in Korea, Naver’s Papago. However, the company managed to carve out a significant market share late in the game.
 
One of the most innovative aspects of DeepL’s approach was its ability to adjust tone and formality in translations.
 
More than 82 percent of language service companies utilize DeepL’s services, according to survey the Association of Language Companies released in October 2024.
 
Industries such as law, manufacturing and medicine require highly specialized, precise translations, which means that AI translation engines must be trained with domain-specific datasets. 
 
Many consumers complain about the poor translation quality of business-to-consumer services such as Papago, but such complaints are almost inevitable, as such broad models are not specifically trained for the tasks they've been asked to do. B2B translators, however, are often tailored to industry-specific needs.
 
Automated translation has yet to reach the level needed to fully replace human translators, particularly in fields where precision, tone and contextual nuance are critical. However, as it grows more specialized, enterprise demand for it is likely to grow.
 
 
With keyword-based searches falling out of fashion, how will AI searches evolve?


People who previously made heavy use of Google may come to rely on generative AI in not only in web searches, but also decision-making processes — in a much more personal way.


For instance, if a user asks, “I need to book a flight to Jeju tomorrow, but I have a morning meeting. Can you book a flight that works?” they may expect an LLM not just to Google that phrase, but to cross-check flight timetables with subway and bus schedules to optimize their transportation time to the airport, as well as ultimately reserve the flight and transportation to the airport. A future agent could even sort results based on past preferences, such as the users’ preferred airlines or seat choices.
 
But for this to happen, an LLM would need access to the users’ personal data, similar to the way in which Apple allows different apps, such as Calendar, Maps, Siri, to share user information — evolving, ultimately, from a mere information retriever to a personalized, but highly invasive, AI assistant.
 
Prof. Billy Choi of Korea University’s Human-inspired AI Research Lab and Prof. Choo Jae-gul at the Graduate School of Artificial Intelligence at KAIST contributed to this report.
 

Related Article

• Explainer: Why is Toss suing Kakao? How a text message launched a legal war

• Explainer: Why DeepSeek AI is a breakthrough for some, a threat for others

• Explainer: Why Korea's teens are abandoning KakaoTalk for Instagram

BY LEE JAE-LIM [lee.jaelim@joongang.co.kr]