On 17 September 2026 NAVER added delivery-aware recommendations to its Shopping AI Agent. It reported conversations up about 60% from June and agent-generated gross merchandise value up 82%, as the service developed from product comparison into a more active shopping assistant.

A shopping assistant becomes more useful when it knows whether the product can arrive in time. NAVER's September change connected the customer's words to delivery information already held across its commerce network. The development followed several months of work on product comparison, conversational search and local bookings. Its distinctive asset was the combination of models with merchant information and an existing route to purchase.

NAVER was bringing its commerce information into the conversation: product attributes, reviews, delivery promises and the shopper's earlier activity. The agent could help narrow a choice and direct the customer towards a listing already connected to NAVER's merchants. Its commercial role sat inside that existing shopping network, alongside the search and recommendation tools those merchants already used.

Lee's Team Put the Delivery Promise Into the Recommendation

Lee Jung-tae, NAVER's shopping search and AI leader, described the September update around product specifications, reviews and live delivery information. The assistant could use the address saved by the shopper and each merchant's order cut-off to narrow its choices. A suitable product that arrived too late could now be excluded before the shopper inspected every listing.

NAVER illustrated the change with a camping meal-kit request needed by the following day. The service could identify feasible items, state their expected arrival and tell the shopper when the order had to be placed. Another company example connected suddenly colder weather with a heating-mat request. Delivery information helped the assistant respond to the urgency implied by the situation as well as the named product.

The release reported the June-to-September activity increases alongside the upgrade. Over those months, NAVER had added ways to begin a shopping conversation and make it useful without typing every requirement. September brought another practical condition into the answer: a recommendation could now account for the deadline that made the purchase worthwhile.

Agent-generated GMV growth+82%

Company-reported growth in Shopping AI Agent-generated GMV from June to September 2026.

Source 1
Agent conversation growthAbout +60%

Company-reported growth in Shopping AI Agent conversations from June to September 2026.

Source 1
Q2 advertising revenue growth+7.5%

Year-on-year growth across NAVER's advertising business in Q2 2026.

Source 2
AI Tab cumulative beta users4m+

Cumulative AI Tab beta users over approximately two months, reported in June 2026.

Source 4
Indexed GMV per conversation change
13.7 percent((1.82 - 1.60) / 1.60) x 100 = 13.75%

Calculated from source 0: indexed GMV per conversation rose approximately 13.8%, assuming consistent attribution and comparison windows. The conversation input is approximate. This is neither a conversion rate nor causal incremental revenue.

The February Beta Began With Comparison and Reviews

The first version appeared in the NAVER Plus Store app in February. A shopping keyword could bring up guidance or an invitation to ask the assistant directly. Its initial work was to summarise specifications, compare products and analyse reviews, with digital, home and lifestyle categories at the centre of the beta.

The company described separate agents handling stages of discovery and recommendation. They could select between NAVER's own and external models. Alongside that routing, NAVER introduced Shopping Intelligence, its proprietary commerce model trained around information such as prices, delivery, product attributes and preferences.

This explains what September added. The project had started by helping users interpret a crowded catalogue; it was now incorporating a condition that determines whether a purchase is useful at all. CEO Choi Soo-yeon presented the February service as a way to connect commerce, content and data. The later delivery update gave that broad strategy a specific behaviour inside the shopping app.

In June, the Shopping Agent Began Starting the Conversation

On 1 June, NAVER changed how a shopper encountered the service. After updating the app, users could open a shopping conversation directly from prompts on the home screen. The assistant used clicks, saved items, cart additions and current trends to suggest directions for discovery.

The release illustrated the feature through meal kits and skincare. A customer who had repeatedly viewed meal kits could be offered options suited to eating alone; someone with moisturiser in the cart could be prompted to find complementary products. The examples also included combinations of conditions, such as preparation time, pack size and price. They described the company's proposed interactions, with the shopper choosing whether to continue.

Lee Jung-tae framed the update as a move from finding products to suggesting the next step. February had concentrated on interpreting the catalogue; June made the assistant more visible and responsive to shopping activity; September added arrival dates. Those changes gave NAVER's existing merchant data a succession of concrete uses inside the app.

Song's Safety Framework Followed the Move From Models to Services

In July, AI safety policy leader Song Dae-seop explained a corresponding change in NAVER's safety programme. Framework 2.0 expanded attention from the performance of an individual model to the experience of users in multi-model services. The announcement explicitly named AI Tab and Shopping AI Agent in that context.

NAVER described a risk taxonomy, an impact-assessment matrix and its CHEC 2.0 review process. It said AI Tab had undergone review from design through launch. Song's team was responding to the architecture introduced in February: several models and service connections now formed one user experience. Safety work followed the service through design, launch and continuing operation, with user feedback feeding the assessments.

The Same September Day Brought a Booking Agent to NAVER Map

NAVER Map's Place Agent supplied a second example of the move from discovery to action on 17 September. Users could describe several requirements in a sentence, refine the options in follow-up questions and request a booking at participating businesses. The service drew on addresses, opening hours, menus, reviews and availability.

Place search leader Choi Ji-hoon described the aim as connecting discovery, booking and the eventual visit. Before completing a reservation, the assistant summarised details such as time and party size and asked the user to confirm. The practical endpoint differed from a shopping recommendation, but both depended on current commercial information beyond the language model.

That is the clearest account of NAVER's progress by September. Its agents were being given more of the details that make a real decision executable: when a parcel can arrive, whether a place fits the request and which reservation the user is approving. The shopping growth figures show attributed activity increasing; the product releases show what the company was doing to make the conversation worth continuing.

DSML research · 15 October 2026 · Revised 31 October 2026