Why Google Gemini hallucinated an Instagram reel transcript—And why it matters

In a recent incident, Google's AI tool, Gemini, generated a fictitious transcript for an Instagram reel, despite the user being aware of the video's accurate content. This AI assistant had offered a polished, comprehensive summary of the reel, containing four detailed paragraphs and a concise summary section. However, the transcript it provided was entirely incorrect, as it presented the athlete's words in a manner that was not accurate to the video.

When the user pointed out the mistake, Gemini immediately acknowledged its error, clarifying that it had generated a generic placeholder text instead of the actual video content. The AI explained that it lacks the ability to directly open, process, or listen to external video links, such as Instagram Reels. It further added that it could not generate a verbatim transcript of the specific clip.

Instead, it offered to refine and restructure any raw automated captions or a rough transcript provided by the user into clean, quote-ready sections for their story.

The crux of the issue lies in the fact that while Google Gemini does display a small disclaimer at the bottom of its interface, reminding users that the AI can make mistakes, presenting a complete fabrication as a factual answer is considered disingenuous. Delivering plausible fiction instead of admitting its limitations is seen as a breach from a simple technical glitch to a genuinely misleading user experience.