Klarna's shopping assistant: useful discovery, unproven checkout authority

External-source research synthesis, not an OfferSignal client result. Based on the retained October 2026 research and source-specific dates.

A shopper can find a plausible product and still reach a merchant page where it is unavailable. That gap matters more than a polished chat answer: discovery, offer verification and payment are different decisions. Klarna's shopping assistant is a useful case because it shows conversational product research in a live consumer app without establishing autonomous buying authority or a measured improvement in sales.

What Klarna launched

On 19 September 2024, Klarna announced an OpenAI-powered shopping assistant for product search, preference-based recommendations, brand and category comparison, review and specification questions, price history, and price, stock, delivery and cashback information. The description did not give the model permission to place, approve or pay for an order. 1 2

The conversational layer sits on an older comparison substrate. Klarna's 2022 tool already filtered catalogues, sorted prices, showed price history, competing offers and ratings, and sought coupons at checkout. The releases do not disclose the 2024 assistant's model version, retrieval, product matching, ranking logic, freshness target or evaluation set. 2 4

Klarna described the stack as “LLM intelligence + PriceRunner knowledge + Klarna UX.” PriceRunner's acquisition announcement described 3.7 million products from 22,500 stores in 25 countries, with prices, specifications, tests and reviews among its inputs. That is provenance, not an audit of current coverage or truthfulness. 2 3

Coverage claims are not performance results

Klarna's headline says the assistant had detailed knowledge of 6 million products, while the detailed English release describes a 5.6-million-product price database from thousands of stores. These are company coverage claims, not an accuracy rate or live-purchasability measure. The record does not explain the difference, store denominator, duplicate treatment or refresh status. 1 2

Payment remained user-mediated. Klarna's US app guidance describes browsing offers and choosing how to pay at checkout, and the independent tester used the assistant for research without taking an instalment plan. No reviewed source shows delegated purchase authority, automated payment consent, a spend limit or autonomous checkout completion for this assistant. 2 5 7

What one hands-on test revealed

CNET journalist Amanda Smith tested the app after creating an account, searching for a black laptop-suitable backpack under $300. The assistant produced four appealing alternatives, but the top two were out of stock at the seller and another option was also unavailable. A broader request worked better. One Michael Kors result included a 4.7/5 rating from 379 reviews, which Smith manually verified on the merchant's site. 5

Smith reported a $229 difference in under 30 minutes: a $328 Cole Haan bag she had considered versus a $99 Michael Kors alternative she liked. That is an observed price difference in one research journey, not audited realised savings. No purchase, delivery or tax comparison, cross-merchant baseline or wider sample is documented. The stock failures are the more important finding: product relevance may survive while availability is stale at click-through. Smith's conclusion was to use the assistant as a research partner rather than the sole search tool; she also found very specific prompts less accurate. 5

A separate claim that Klarna's January 2024 support assistant handled work equivalent to 700 full-time agents concerns customer service, not this shopping workflow. 2

A later channel, not a retroactive result

In May 2026, Klarna announced a distinct Shopping Search app inside ChatGPT. Its Product Search MCP server was described as connecting ChatGPT to live data for more than 100 million products and 400 million merchant listings across 13 markets, with purchase redirected to a merchant site. It also described organic relevance and clearly labelled sponsored placements. This shows a later distribution channel and an explicit merchant-site completion boundary; it does not prove that the 2024 assistant had the same scale, freshness, ranking policy or performance. 6

An illustrative business application

For a distributor or manufacturer, the transferable pattern is an evidence-carrying discovery workflow, not an autonomous buyer. This is illustrative, not a reported Klarna deployment. Start with one product family. Give each item a canonical identifier, variant, approved relationships, source, market, timestamp, availability state and price or quote-validity period. Link reviews and specifications to their source, and distinguish a summary from a live offer.

Test product-match quality, stale-offer rate at click-through, total-price accuracy including delivery, specification provenance, clarification rate and recommendation-to-quote progression. Revalidate stock and price before payment approval. The commercial or technical owner remains authoritative for suitability, terms and the transaction. Klarna's unresolved questions—eligibility, availability, accuracy, conversion, model design and refresh—are fields a serious pilot should measure.

Sources and next reading

Explore OfferSignal · Focused Business & AI Workshop · All research

References