Naver has made its position on outside AI very plain: you can read Korea’s biggest pile of user-generated content through Naver, and almost nowhere else. The robots.txt files on Naver Blog, Naver Cafe and Knowledge iN now open with the same shouty line: “BOT ACCESS FOR THE PURPOSES OF AI TRAINING AND RETRIEVAL-AUGMENTED GENERATION (RAG) IS STRICTLY PROHIBITED.” And last week Naver told Newsis it will not let third-party AI agents such as Meta’s Muse search or buy on its shopping platform either.
Put those together and you get the clearest statement yet of Naver’s AI strategy. Its own engine gets the data. Everybody else gets a closed door.
What the robots.txt files actually say
We checked the live files on 6 October. Anymorph reported much the same picture after checking them on 4 October.
- Naver Blog (desktop and mobile) disallows GPTBot, OAI-SearchBot, PerplexityBot, Google-Extended, ClaudeBot, Claude-SearchBot, meta-externalagent, Applebot-Extended and CCBot from the whole site. The generic
User-agent: *block only shuts off utility paths, so Googlebot can still crawl posts. - Naver Cafe blocks everything:
User-agent: *is disallowed from/, and Googlebot, Bingbot, Baiduspider, Yandex, GPTBot, OAI-SearchBot, Applebot and Amazonbot are each named and blocked. The only bots allowed in are Facebook’s link-preview crawlers. - Knowledge iN allows Naver’s own Yeti crawler on most paths (though not the
/qna/detailquestion pages) and blocks a long list of AI and data bots. That list includes ChatGPT-User, Bytespider, ChatGLM-Spider and Korea’s own WRTNBot, plus SemrushBot, AhrefsBot, DataForSeoBot and Baiduspider.
One oddity is worth noting. The Blog file also disallows Yeti, Naver’s own web crawler, from the whole of blog.naver.com, and Knowledge iN keeps Yeti off its question-detail pages. The obvious reading is that Naver doesn’t need to crawl its own platforms from the outside to put it in Naver Search. That is our inference, not something Naver has documented. Either way, these files are no guide to how Naver itself treats its own content.
Why this matters more in Korea than anywhere else
These aren’t side properties. According to Anymorph’s summary of a May 2026 Naver briefing, chief data and content officer Kim Kwang-hyun said Naver’s user-generated services made up 70% of the content cited in AI Briefing since January. We have already covered how AI Briefing citations keep sliding further inside Naver. The robots.txt changes are the other half of that story. Naver’s answers draw on Naver’s corpus, and rival answer engines can’t legally draw on the same one.
For a brand, the practical consequence is awkward. A carefully run official Naver Blog can help in Naver AI Briefing and in Google, but it is unlikely to become a source for ChatGPT search, Perplexity, Claude or Gemini app grounding, because those crawlers are told to stay out. Korean users do use those tools (we wrote about ChatGPT narrowing Naver’s grip earlier this year), so the brand’s own website has to do that job.
And now the agents are locked out too
The crawler policy now extends to commerce. On 30 September, Newsis (via Financial News) quoted Naver as saying it “does not allow external AI agents to search for products or access data within Naver” and has “no plans to change this policy at present”. Muse isn’t available in Korea yet. The point is that when it arrives, it is unlikely to be able to search or buy for users on Naver Shopping.
Naver isn’t alone. i-boss’s 6 October news round-up reports that Naver, Kakao and Coupang have all moved to block outside AI from their shopping services. In the US, a16z’s latest consumer AI report notes that Amazon shut Muse out in under two weeks, while Shopify, Instacart, Expedia and others signed official integrations.
Meanwhile Naver is building out its own agent. Newsis says Naver’s shopping AI agent grew monthly users and daily conversations about fivefold between March and August. Daily transaction value through the agent rose about fourfold. In-agent payments will be piloted in some categories before the end of the year.
Our take
This is a walled garden with robots.txt as the wall. It is entirely rational for Naver: its Blog, Cafe and Knowledge iN archive is the moat that keeps HyperCLOVA X answers local, and handing it to OpenAI or Meta for free would be strategic self-harm. It is less comfortable for publishers and brands, who now have to run two strategies:
- Inside Naver: an official Blog, Knowledge iN presence and Place/Shopping data for AI Briefing and the shopping agent.
- Outside Naver: a crawlable website open to OAI-SearchBot, PerplexityBot and friends if you want to be cited by everything else.
The real risk is assuming that content on Naver counts as content on the Korean web. For AI search outside Naver, it no longer does. If your Korea plan lives entirely on blog.naver.com, the rest of the AI internet can’t see it, and on current evidence Naver likes it that way.