Which AI engine you are optimising for in Korea now makes more difference than which keyword. A new report from Seoul consultancy 적정마케팅연구소 (Jeokjeong Marketing Lab), published on 15 October, found that Naver Blog accounted for 46.5% of the sources cited in Naver’s AI Briefing, 7.7% in Google’s AI Overviews, and none at all in more than 42,000 ChatGPT and Gemini citations. The same Korean content ecosystem looks completely different depending on which AI is answering.
The firm runs a GEO tool called BrAInd, which it opened to outside clients this month, so read this partly as a product launch. The method is still more transparent than most GEO marketing, and the main finding is easy to check against what we already know about how Naver runs its platforms.
What the study measured
According to write-ups in StartupN and Digital Times, the firm took 635 non-branded queries connected to 26 brands it monitors and collected AI answers between 26 August and 24 September. That produced 21,225 answers and 85,392 linked citations across four surfaces: Naver AI Briefing, Google AI Overviews, Gemini and ChatGPT. The headline numbers:
- Naver Blog share of citations: AI Briefing 46.5%, AI Overviews 7.7%, Gemini 0 of 9,597, ChatGPT 0 of 32,832.
- Each platform has its own favourite source type. AI Overviews leaned on video and social at 29.4%, mostly YouTube. ChatGPT favoured public bodies and associations at 18.0%. Gemini went to non-Naver blogs such as Tistory and Brunch at 13.1%. News sites were only 1.4–3.5% everywhere.
- Owned channels drive brand mentions, except on Naver. When a brand’s own site or blog was cited, the brand was named in 83.7% (AI Overviews), 90.6% (Gemini) and 92.3% (ChatGPT) of answers, compared with 18.0–29.6% when it wasn’t. On AI Briefing the figure was only 31.9% even when the brand’s own channel was cited. Owned-channel citations appeared in just 5–19% of answers anywhere.
- Ranking overlap varies. Of the cited sources, 50.2% also appeared on the first page for the same query in Naver’s AI Briefing, and 40.5% in Google’s AI Overviews. For Gemini it was 21.3% against Google’s first page, and for ChatGPT just 4.2% against Bing’s, though the ChatGPT sample covered only three brands.
- Headings, tables and FAQs made no clear difference. Comparing cited and non-cited pages from the same 400 Google SERPs, the page with more of each element was the cited one only 46.3%, 46.9% and 46.2% of the time. That is no better than a coin flip.
- Some engines repeat themselves more than others. AI Briefing reused the previous answer’s sources 74.5% of the time, against 51.4% for AI Overviews, 36.2% for Gemini and 16.0% for ChatGPT.
The zero is a robots.txt story
The zero isn’t surprising. As we reported last week, Naver Blog’s robots.txt blocks GPTBot, OAI-SearchBot, Google-Extended, ClaudeBot, PerplexityBot and others from the whole site, while still letting Googlebot crawl posts. That fits the numbers neatly. Google’s AI Overviews, built on the Googlebot index, can still cite Naver Blog (7.7%). ChatGPT and Gemini can’t, because ChatGPT depends on the blocked OpenAI agents and Gemini honours the Google-Extended opt-out. Naver’s own AI Briefing has the whole corpus to itself. This is the clearest evidence yet that Naver’s walled garden reaches into AI answers as well as search results.
It also explains why the study disagrees with another one. The AIBA AI Citation Index from Hamshout and Delight Communication found Naver Blog citations rising 180% between July and August across ChatGPT, Gemini and Perplexity combined. Those figures pool three engines with different access, including Perplexity, use a different question set and cover July and August. We also don’t know exactly when Naver added its blocking language. We first checked it on 6 October. You can’t compare a pooled total with a per-engine count. Anyone quoting “Naver Blog is winning AI citations” needs to say which AI they mean.
What it means for brands in Korea
First, a single content plan for “AI search” won’t work in Korea. Naver Blog is still the main source for AI Briefing, but it does nothing for you in ChatGPT or Gemini. Tistory, Brunch, public-sector pages and YouTube do. A Korean brand needs at least two content footprints: one inside Naver and one on the open web.
Second, the AI Briefing finding is the uncomfortable one. Even when Naver cites your own channel, the answer names your brand less than a third of the time. That matches our earlier recap on AI Briefing citations sliding inside Naver. Naver uses brand content as raw material and then writes a generic answer from it. Being cited and being chosen are different things.
Third, stop selling clients FAQ schema as an AI-citation lever. Here, as in most honest studies, it is no better than chance. And with Naver repeating three-quarters of its sources from one answer to the next, the hard part is getting into the Briefing in the first place. Once you are in, you tend to stay.
Opinion: the vendor caveats are real. It covers 26 brands, one month, only correlation, and it comes from a company selling the measuring tool. The finding that matters doesn’t depend on any of that, though. Naver has decided its content feeds Naver’s AI and nobody else’s, and the citation data now shows that policy working. International teams that treat “Korean GEO” as a translated version of their ChatGPT playbook are optimising for an engine where Korea’s biggest content platform simply doesn’t exist.