In an investigation, Lead Stories found that Google Gemini's SynthID AI watermark detector appears to confuse results within the same chat. Over several chats, the chatbot initially claimed a video file had a SynthID watermark, indicating AI generation or editing, but later claimed there was no such watermark for the same file. Testing revealed a pattern where Gemini always returns results for the first image or video uploaded in a chat session when asked about a SynthID watermark, even when a new file is uploaded and the question is repeated. Lead Stories reported the issue via Gemini's built-in feedback tool and is trying to reach Google through a mutual contact. SynthID technology embeds imperceptible watermarks into AI-generated content, detectable by SynthID's technology. Google's help page notes that Gemini can currently only recognize content created by Google AI tools. The testing used an image created with Google Gemini and a photo taken by a human.