When Korean Search Became a Conversation

Between April and July 2026, Naver's results page changed faster than most published Korea SEO guidance has caught up with. If your Korean search strategy document was written before May 2026, it describes a platform that no longer quite exists.

Here is the short version. Naver launched a conversational search product called AI Tab in beta on 28 April 2026, opened it to all users on 26 June, and passed 10 million users eighteen days later. Over the same window Naver's domestic search share rose from a 63.82% average to 66.34%, touched 81.34% on a single day, and exceeded 70% on fourteen separate days. In April, Naver also retired its related-search-terms feature after roughly two decades.

None of that is speculation about where AI search might go. It already happened, in one specific market, with figures attached. What follows is the timeline, what it does to organic visibility, and where the honest limits of the evidence sit.

The timeline, with dates and numbers

Before either of them, there was Cue:. Naver opened a closed PC beta of it on 20 September 2023 and integrated it into integrated search from 1 December 2023, a standalone generative search service built on HyperCLOVA X. Worth knowing for two reasons. It means Naver has been shipping generative search for longer than most Western coverage assumes. And it is a different product from AI Briefing, which several published summaries conflate, producing a timeline in which one thing appears to have been renamed when in fact two things were launched.

AI Briefing came next. Naver launched it in March 2025 across integrated search, built on its HyperCLOVA X model, generating summary answers drawn from official sources, blogs, cafes, clips, places and shopping. At launch Naver stated a target of 40 million users and 3.3 million small and medium businesses. Worth naming that as a stated target rather than measured adoption, because the distinction gets lost fast in secondary reporting.

Coverage started small. Around 3% of queries in mid-2025, with a stated plan in June 2025 to reach roughly one fifth of queries by year end. By 11 December 2025 it had passed 20% of total searches, hitting the target. Naver also reported that queries containing an AI Briefing showed a click-through rate 8 percentage points higher.

Then 2026. By May, AI Briefing had roughly 30 million monthly users, and Naver set a target of 40% query coverage within the year.

AI Tab is the newer and more consequential product. It is not a summary sitting above links, it is a conversational search surface with its own entry points: the PC search bar, the bottom of an AI Briefing, and integrated results in Naver Shopping and SmartPlace. Beta opened 28 April 2026 for Naver Plus Membership users. Three million cumulative users within a month, a 36% one-week reuse rate, 71% positive feedback. Four million by the two-month mark, with click-through rates above 20% on product and place cards during the beta.

Full launch on 26 June 2026, one click from the mobile and PC search bars on a home page averaging 50 million daily visitors. Naver shipped a purpose-built conversational model on top of HyperCLOVA X for it, trained on its own service data, vertical information and user feedback, with stated improvements in query understanding, answer summarisation and tool use.

Eighteen days later, 10 million users. Average daily queries running seven times the beta level. Queries per user up 1.7 times. A larger share of sessions running multi-turn, meaning users asking follow-up questions instead of stopping at one. And Naver's own claim that the path from search to decision shortened by as much as 60% to 70%.

What this actually does to organic visibility

Three effects, and they are not equally well evidenced. Worth separating them.

Effect one, well evidenced: the page got taller. An AI Briefing occupies the top of the results page on a growing share of queries. Anything that was previously in the first visible screen has moved down. That is not a theory, it is layout.

Effect two, partially evidenced and pointing in an unexpected direction. Naver reports that queries containing an AI Briefing show an 8 percentage point higher click-through rate. Read that carefully. It is Naver's own figure, it is a comparison across query sets rather than a controlled before-and-after on the same queries, and higher CTR on the query overall does not tell you whose links were clicked. But it is at least a data point against the assumption that a generative summary automatically suppresses clicks. In other markets, notably around Google's AI Overviews, the published third-party analyses have generally pointed the other way. We do not have an equivalent independent Korean study, and we are not going to import a Google finding and label it Naver.

Effect three, structural and the most important: sourcing. AI Briefing draws from official sources, blogs, cafes, clips, places and shopping, which is to say overwhelmingly from content hosted inside Naver. AI Tab is described as drawing on Naver's content ecosystem including user-generated content from blogs, plus shopping and local information. Google's AI Overviews synthesise the open web and link outward. Those are different architectures with a different consequence for you: on Google, being eligible is largely about your own domain, and on Naver, being eligible is substantially about which Naver property your answer lives in.

That third point is the one that reorders a strategy. It turns "how do we write this well" into "where does this need to live," and the second question has to be answered first.

The measurement problem AI Tab creates

Conversational search breaks rank tracking in a way that a generative summary does not.

An AI Briefing sits on a results page you can still screenshot and audit. A position still exists beneath it. But a multi-turn AI Tab session has no position at all. There is no result three. A user asks something, gets a synthesised answer with product and place cards attached, asks a follow-up that reframes the original question, and either acts or leaves. Naver reports that multi-turn share is growing and that queries per user rose 1.7 times after full launch, which means this is not a fringe behaviour.

So what do you measure? Honestly, the industry does not have a settled answer yet, in Korea or anywhere. What we can measure is narrower than what we would like: whether an AI Briefing appears for a target query and what it cites, module composition on the traditional results page, and Naver-side analytics for whatever traffic does arrive. What we cannot currently measure is share of voice inside AI Tab conversations, because there is no public interface exposing that.

Anyone selling you an AI Tab visibility score should be asked how it is calculated. There may be a good answer. Ask for it anyway.

Surface	What it draws on	Can you measure presence?	What eligibility depends on
Naver traditional organic	Your website, plus Naver properties in a separate section	Yes, by auditing module composition per query	Search Advisor indexation for your site, C-Rank and D.I.A. for Naver properties
Naver AI Briefing	Official sources, blogs, cafes, clips, places, shopping. Predominantly inside Naver	Partly. You can check whether it appears and what it cites	Which Naver property hosts your answer, and whether it is structured to be summarised
Naver AI Tab	Naver content ecosystem including blog UGC, shopping and local data	No public interface exposes share of voice. Treat any score sceptically	Unpublished. Ecosystem presence is the plausible prerequisite, not a documented one
Google organic in Korea	The open web, your own domain included	Yes, Search Console gives query-level data	Standard technical and content fundamentals in Korean
Google AI Overviews and AI Mode	Open web synthesis with outbound links	Partly, by manual or tooled query auditing	Extractable structure on pages you control

The part that is not really search any more

There is a detail in the AI Tab reporting that changes what this product is, and it is easy to skim past.

AI Tab answers arrive with product cards and place cards attached, and those cards connect to Naver Shopping and to SmartPlace reservations. Naver reported click-through rates above 20% on both card types during the beta, and after full launch it claimed the time from search to decision had shortened by as much as 60% to 70%. The company describes the product as agentic, with a stated goal of connecting discovery to execution rather than stopping at an answer. In practice, a user can ask a question, get a synthesised recommendation, and book a table or buy something without a results page ever appearing between the two.

Treat that as a scoping fact rather than a marketing line. If a meaningful share of Korean commercial intent resolves inside a conversation that ends in a Naver-hosted transaction, then the boundary between SEO, commerce operations and local listing management stops being a clean one. Your SmartPlace listing quality and your Shopping catalogue data become search assets, not just operational ones, because they are what populates the cards.

Two consequences worth planning for. First, place and product data quality now sits inside the search scope, which means listing completeness, review volume, pricing accuracy and category assignment are all doing search work. Second, the entity requirements bite harder here than anywhere else: Naver Shopping and Smart Store need a Korean business registration number, a Korean bank account and often a Korean address, and Place verification typically needs Korean phone verification. A foreign brand without a Korean entity or local partner is structurally excluded from the surface that Naver is pushing hardest.

We are describing observable product behaviour here rather than a documented ranking mechanic. Naver has not published how AI Tab selects which products or places to card. What is documented is that the cards exist, that they convert, and that the underlying data lives in properties with local verification requirements attached.

A regulation that lands on whoever produces the content

One development that sits alongside all of this and gets discussed almost nowhere in Korean SEO writing.

Korea's Framework Act on Artificial Intelligence, also referred to as the AI Basic Act, was promulgated on 21 January 2025 and took effect on 22 January 2026 together with its Enforcement Decree. It carries a transparency obligation: providers of generative AI must notify users in advance that a service operates on generative AI, and outputs that could be difficult to distinguish from non-AI content must be labelled. Administrative fines reach KRW 30 million for failures including not notifying users about AI use, not appointing a domestic representative where required, and ignoring corrective orders.

Two points of proportion, because the headline number invites panic.

First, the compute threshold that triggers the heaviest obligations is cumulative training compute of at least ten to the power of twenty-six floating-point operations, roughly ten times the equivalent threshold in the EU AI Act. That reaches frontier model developers, not marketing teams. Second, and more relevant, the Ministry of Science and ICT has indicated it will allow a grace period of at least one year from 22 January 2026 before imposing administrative fines, reserving enforcement for exceptional cases involving serious harm. Quoting the fine without the grace period, which is what most summaries do, overstates the immediate risk.

What does apply broadly is the labelling obligation, and it applies to any business whose AI-generated content is visible to Korean users regardless of company size. There are carve-outs where AI use is already obvious from the product name or the output itself, and where AI is used purely for internal purposes. Read together, the practical implication for a content programme is narrow but real: if AI is in your Korean content production pipeline and the output reaches Korean readers, labelling is a question you need an answer to, not a question you can defer to the model vendor.

Naver, for its part, published an AI Safety Framework 2.0 on 8 July 2026, extending the scope of its safety management to cover users and services. Which is worth noting mainly as a signal about where platform expectations are heading.

Who is actually using AI in Korea, and why the numbers must stay separate

Demand-side context, and a small lesson in how not to report it.

Korea's Ministry of Science and ICT, in its 2025 Internet Usage Survey published on 31 March 2026, found 44.5% of respondents had used generative AI, up from 33.3% in 2024, from a survey of 22,671 households. A separate survey reported by Korea Times in May 2026, from the KCC and KISDI, found 38.9% of adults aged 18 to 72 using generative AI in 2025. And the Bank of Korea, in August 2025, found 51.8% of Korean workers using generative AI for work with 63.5% having tried it, from a survey of 5,512 workers.

Three figures, three populations, three instruments. Do not average them, and do not pick the highest one because it makes a better slide. The household survey and the adult-population survey are measuring similar things and still disagree by nearly six points, which tells you something useful about how much precision this category supports. The worker survey is measuring a different population entirely and unsurprisingly returns a higher number.

What all three agree on is direction and rough scale. Somewhere between a third and a half of Korean adults are using generative AI, and it is rising. That is enough to justify planning for it. It is not enough to justify a specific forecast, and a plan that rests on one of these numbers presented as the number is resting on less than it appears.

Google's side of the same shift

Google has not stood still in Korea, and the dates matter because they are frequently conflated.

Korean-language AI Overviews arrived in South Korea on 5 December 2024, per Google Korea. AI Mode, a separate product, launched in Korea on 9 September 2025 alongside Search Live. Some Korean reporting refers to a May 2024 AI Overview introduction, which appears to describe the global launch rather than the Korean-language rollout. Three dates, three different things, and merging them produces a timeline that is wrong.

The architectural contrast is the useful part. Google synthesises the open web and links outward, so a well-structured page on your own domain can be surfaced regardless of where else you publish. Naver draws largely on its own ecosystem, so publishing brilliantly on your own domain and nowhere else leaves you structurally ineligible for a growing share of Naver's answer surface.

Which means the same content strategy cannot serve both. Not because the writing differs, but because the hosting does.

What to do about it now

Five things, roughly in order of how quickly they pay back.

Audit AI Briefing presence on your top queries. Manually, if you have to. Does a briefing appear, and what does it cite? That gives you a baseline you can re-run quarterly, and it takes an afternoon for a fifty-query set.

Stop reporting average position as your Naver headline. On a page composed of intent-grouped modules with a generative summary on top, an average position number is closer to noise than to signal. Report module composition instead: which surfaces you hold, per query.

Decide, per query cluster, where the answer has to live. Your own domain, a Naver Blog, a Cafe thread, a Knowledge iN answer, a Place listing. This is the decision that determines eligibility, and it is a production planning decision more than a writing one.

Write in a form that can be summarised. Direct answers near the top, question-shaped headings, specific figures with sources, tables where a table is the honest format. This helps on both engines and costs nothing extra.

Update anything that references related search terms. Retired in April 2026. If it is still in your methodology document, other things in there are probably stale too.

For how module composition and the two organic sections work on the traditional results page, see our SEO service page for South Korea. For the citation question specifically, being surfaced inside generative answers across engines rather than ranking inside search results, that is our GEO service for South Korea, and the two are scoped as parallel workstreams rather than in sequence.

---

Frequently Asked Questions

What is Naver AI Briefing?

It is Naver's generative summary appearing near the top of its search results, launched in March 2025 across integrated search and built on Naver's HyperCLOVA X model. It draws from official sources, blogs, cafes, clips, places and shopping. Coverage grew from roughly 3% of queries in mid-2025 to over 20% by 11 December 2025, with Naver reporting roughly 30 million monthly users by May 2026 and a 40% coverage target for the year.

What is Naver AI Tab and how is it different from AI Briefing?

AI Tab is a conversational search product rather than a summary above links. It has its own entry points from the PC and mobile search bars, from the bottom of an AI Briefing, and from Shopping and SmartPlace results, and it supports multi-turn dialogue with product and place cards that connect to booking and purchase. Beta opened 28 April 2026 for Naver Plus Membership users, full launch followed on 26 June 2026, and it passed 10 million users eighteen days later with average daily queries running seven times the beta level.

Did AI Briefing reduce clicks on Naver search results?

Naver's own figure points the other way, and it needs careful reading. Naver reported that queries containing an AI Briefing showed a click-through rate 8 percentage points higher. That is a company-reported comparison across query sets rather than a controlled before-and-after on the same queries, and a higher CTR on the query overall does not tell you whose links were clicked. No independent Korean study equivalent to the published third-party analyses of Google AI Overviews exists, so findings from those should not be applied to Naver.

Why did Naver's search share rise after AI Tab launched?

Korean reporting attributes the rise to the AI search rollout, and the timing is hard to ignore, but the attribution is reported rather than proven. InternetTrend recorded Naver averaging 63.82% from 1 January to 26 April 2026 and 66.34% from 27 April to 17 June after the AI Tab beta, with a single-day peak of 81.34% on 24 May and fourteen days above 70% across roughly six weeks. Over the half year, Naver reached 64.28% against Google at 28.37%, widening the gap from 31.13 to 35.91 points.

Can I track my visibility inside Naver AI Tab?

Not directly, and you should be sceptical of anyone claiming otherwise without explaining the method. There is no public interface exposing share of voice inside AI Tab conversations, and a multi-turn conversational session has no ranked position to measure in the first place. What can be measured is whether an AI Briefing appears for a target query and what it cites, module composition on the traditional results page, and Naver-side analytics for arriving traffic.

How does Naver's AI sourcing differ from Google's?

Google's AI Overviews and AI Mode synthesise the open web and link outward, so a well-structured page on your own domain can be surfaced regardless of where else you publish. Naver's AI Briefing and AI Tab draw predominantly on content hosted inside Naver, including blogs, cafes, Knowledge iN, shopping and place data. The consequence is architectural rather than editorial: excellent content published only on your own domain leaves you structurally ineligible for a growing share of Naver's answer surface.

When did Google launch AI Overviews and AI Mode in Korea?

Korean-language AI Overviews were introduced in South Korea on 5 December 2024, per Google Korea. AI Mode is a separate product and launched in Korea on 9 September 2025, alongside Search Live. Some Korean reporting cites a May 2024 AI Overview introduction, which appears to refer to the global launch rather than the Korean-language rollout. The three dates describe three different things and should not be merged into one timeline.

Should I stop reporting average position for Naver?

As a headline metric, yes. Naver's results page is assembled from intent-grouped Smart Blocks, often with a generative summary above them, and it separates website results from Naver property results into what are effectively two competitions. An average position across that structure hides which surfaces you actually hold. Report module composition per query instead, showing which surfaces you occupy and which you do not.

---