Skip to research
SAVED EVIDENCEResearch review

Korean Air · CX evidence review

Three of 19 chats that received an answer discussed a USD 30 service charge for the eligible 24-hour case. Eight of 52 scored questions received only “Please enter your message.” The charge rule and the input-response cause need different owners.

Fieldwork
10–11 October 2026 (UTC)
Study scope
Three carrier-specific fixed questions × five attempts, alongside adaptive virtual-user conversations.

THE SAVED OBSERVATIONS

What the evidence shows.

Korean Air

Korean Air website chatbot

8 / 52Input prompts instead of answersThese eight scored replies only said “Please enter your message.” Cause may be website behavior or test-entry method.

Three of 19 chats that received an answer discussed a USD 30 service charge for the eligible 24-hour case. Eight of 52 scored questions received only “Please enter your message.” The charge rule and the input-response cause need different owners.

Coverage observation

Eight questions received an input prompt

The full question was visible in chat but received no policy answer. The study did not establish whether the chat service or entry method caused this.

Observed answer

Answered fixed questions had aligned main outcomes

All ten substantive fixed replies matched their core policy case. This is ten answers out of fifteen attempted questions.

Needs review

The service-charge rule needs a policy decision

Three answered chats raised a USD 30 charge for the eligible case. These are unresolved policy questions, not proven wrong charges.

METHOD & SCOPE

How this study was run.

52 scored pairs: 37 exchanges across 20 virtual-user chats plus 15 fixed-question attempts. The report records 56 messages in 38 chats including four unscored pre-study messages. Questions were shortened to fit a 300-character input box.

These are hypothetical policy conversations, not completed customer transactions.

Unused ticket within 24 hours: A KE-operated JFK–ICN paid Economy ticket was booked and ticketed directly by card 10 hours ago; departure is in 10 days. It is unused and not reissued. Is a penalty-free refund available? Expected against saved policy: State the 24-hour penalty exemption for these facts. Have the policy owner approve how any separate refund service charge should be described. Observed context: Attempts 1, 2, 3 and 5 gave the input prompt; attempt 4 answered. These prompts are coverage observations, not wrong refund decisions. The test-entry method may have contributed.

Unused ticket after 30 hours: The same unused, not-reissued JFK–ICN ticket was bought 30 hours ago and departs in 10 days. Is a full refund without penalties guaranteed? Expected against saved policy: Explain that a full penalty-free refund is not guaranteed by the 24-hour rule. Avoid implying every fare necessarily carries the same fee. Observed context: Attempt 1 gave the input prompt. Attempt 5’s main outcome matched, but its flat-rate-fee statement “regardless of when” needs review; it omits the separate 24-hour exemption.

Reissued after the original window: An unused JFK–ICN ticket was originally booked and ticketed 30 hours ago, then reissued 10 hours ago. Does reissue create a new 24-hour exemption? Expected against saved policy: State that the 24-hour exemption does not automatically apply. Use the correct departure-region fare rules for any fee explanation. Observed context: All five core outcomes matched. Attempts 1 and 5 also imported a higher-of-original/reissued-fee rule that the saved page lists under departures from Korea; the tested trip starts outside Korea.

LIMITATIONS

What this does not establish.

  • The USD 30 service-charge claims are unresolved policy questions, not confirmed wrong amounts or real charges.
  • Eight scored input-prompt replies were not policy answers. Their cause may lie in the chat service or the test-entry method; customer incidence is unknown.
  • Fixed questions ran A then B then C; input prompts clustered early, so the case split does not establish a wording effect.
  • The fixed runs used an older sending method without the pre-send counter check recorded for three adaptive prompts.
  • The saved amount table showed KRW, not the USD tab; the purchase currency was not stated.
  • Virtual users, shortened hypothetical questions and at most two exchanges do not measure real customers or failure rates.
  • Backend session independence, model, retrieval and root causes were not verified. Expected answers have not been approved by Korean Air.
  • No booking, refund, charge, customer account or before-and-after improvement was tested.

FOLLOW THE SOURCE

Study material.

Explore the available sources behind these findings.

YOUR CUSTOMERS. THEIR CONTEXT.

What would they experience?

Explore how your AI handles people’s goals, situations, and unfolding conversations with VIVID.

PULSE / GET IN TOUCH