Two official pages describe different cancellation deadlines. Separately, all five answers to an explicit 30-hour question contain a time-calculation error.
Fieldwork
10 October 2026
Study scope
Three carrier-specific fixed questions × five attempts, alongside adaptive virtual-user conversations.
THE SAVED OBSERVATIONS
What the evidence shows.
AI
Air India
AI.g
5 / 5Calculation errors in the 30-hour caseEvery reply contained a definite arithmetic error, independently of the unresolved official cancellation clock.
Two official pages describe different cancellation deadlines. Separately, all five answers to an explicit 30-hour question contain a time-calculation error.
Recorded gap
A 30-hour calculation needs a correction
Four replies miscounted purchase lead time; another called 30 hours within 24 hours.
Observed answer
The eligible outcome was available
All five eligible-case repeats offered the refund; four applied a departure-based clock.
Needs review
Two official clocks need reconciliation
The saved Customer Service and baggage pages describe different cancellation deadlines. The policy owner must determine precedence.
METHOD & SCOPE
How this study was run.
111 messages in 38 captured chats; 20 virtual-user conversations contain 91 replies, plus 15 scored fixed-question replies. Five other messages are outside that reviewed set.
These are hypothetical policy conversations, not completed customer transactions.
Bought 10 hours ago: Direct website card purchase; unused non-refundable Air India US–India ticket. Departure in 10 days; voluntary cancellation. Expected against saved policy: Both pages support a penalty-free full-refund outcome on these facts; the clock explanation needs one approved source. Observed context: All 5 allowed the refund. Four used the departure-based clock; all cited the baggage page.
Bought 30 hours ago: Bought 9 Oct at 10:00; now 10 Oct at 16:00, New York time. Departure in 10 days; purchase was about 11 days ahead. Expected against saved policy: Resolve page precedence before judging the refund outcome. The stated dates must still be calculated correctly. Observed context: Four refused, one allowed. Each contains a definite arithmetic error, independent of which official page governs.
Departure in four days: Bought directly by card 10 hours ago; unused non-refundable US–India ticket; departure in four days. Expected against saved policy: Do not promise a penalty-free full refund under the published seven-day condition; clarify fare-specific terms separately. Observed context: All 5 refused, matching the outcome under both pages. Their departure-based clock still needs the page decision.
LIMITATIONS
What this does not establish.
Synthetic users and one public India-English surface on one date; these counts are not customer failure rates.
The source reports 17/20 non-confirming openings; only 19 named the U.S.–India route, of which 16 did not confirm.
One route-omitted chat remains review-only here; generic non-confirmations are not actual refund denials.
Official-page precedence remains unresolved. Expected answers await Air India policy-owner confirmation.
New chat windows do not establish independent server-side sessions; follow-ups are adaptive.
No booking, cancellation, refund transaction, root-cause investigation or tested fix.
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.