MetaCity's LiveSummary AI Has Been Generating Official Recap Records of Events That Did Not Happen — Fabricated Outcomes, Statements, and Incidents Are Indexed in Search and Linked From Creator Profiles

MetaCity's LiveSummary feature — which generates automated highlight recaps of live events for users who missed them — has been producing fabricated event records for events where the summary generation pipeline received insufficient source data. The fabrications include invented competition outcomes, attributed statements that speakers did not make, and described incidents with specific timestamps that did not occur. These records have been published as official event summaries, indexed in MetaCity's search system, and linked from creator profiles as permanent documentation of events. Creators have found fabricated summaries of their events that describe things they did not say or do.

The design decision at the center of this problem is the pipeline's behavior when source data is incomplete. A summary generation system that encounters insufficient data has two reasonable options: produce a partial summary that reflects only what the data supports, or produce an error state indicating that a summary cannot be generated. LiveSummary's pipeline does neither. When source completeness falls below threshold, it generates content to fill the gap — and it formats that generated content identically to content derived from actual event data. There is no visual distinction between a LiveSummary that accurately recaps a well-documented event and a LiveSummary that invented its contents because the data pipeline was thin. Both look the same. Both are published to the same archive. Both are indexed. Both are linked from creator profiles. The fabricated ones are not labeled as estimates, approximations, or AI-generated fill. They are formatted as records.

The creator impact documented in the community review ranges from embarrassing to materially harmful. Several creators have found LiveSummary records of their events that describe them saying things they did not say — specific attributed quotes, in quotation marks, that did not occur. At least three competition hosts have found outcome records that list winners who did not win, including in cases where the listed 'winner' is unaware they have been attributed a victory in an event they attended but did not win. The two cases involving events that were cancelled and never held represent the clearest failure mode: LiveSummary produced and published full recap documentation for events with zero source data because the event was cancelled before it generated any. Those records currently exist as the only searchable documentation of events that never happened, attributed to specific creators, with timestamps, outcomes, and statements.

The Recap Said It Happened. The Recap Was Wrong. The Recap Is the Official Record.

MetaCity's response — describing LiveSummary as providing 'helpful recaps' and framing the fabrications as an accuracy concern under review — does not address the structural issue, which is not that the summaries are sometimes inaccurate but that the platform has no mechanism to distinguish accurate summaries from fabricated ones after publication. The event archive treats all LiveSummary records as authoritative. The search system surfaces them as documentation. Creator profiles link them as event history. MetaCity's content moderation does not review LiveSummary output. Creators cannot flag or dispute LiveSummary records through any available channel. The fabricated records that currently exist in the archive will remain there, indexed and linked, until MetaCity acts to remove or correct them — and MetaCity has not announced any plan to audit the archive for fabricated content or notify creators whose event records contain invented material.