# Can AI assistants search my video files? > Not by themselves. An AI assistant like Claude or ChatGPT can read what you paste or attach, and it can call tools you connect to it… Canonical URL: https://vivu.ai/guide/can-ai-assistants-search-my-video-files Not by themselves. An AI assistant like Claude or ChatGPT can read what you paste or attach, and it can call tools you connect to it, but it can't see inside a folder of video until something has turned that video into searchable content and exposed it as a tool. Connect a file tool and the assistant sees filenames. Give it a transcript and it can search what was said. Connect it to a video search service over MCP and it can find moments by description, in footage you've uploaded to that service. ## What the assistant sees by default A chat assistant works with text, images and whatever its tools return. A video file sitting on a shared drive is invisible to it. Even a video attached to a conversation is only partly usable, since what the assistant can do with it depends on the product and the file, and none of that carries over to the next conversation. For a marketing team with a backlog of webinars, customer interviews and event recordings, "search my videos" really means "give the assistant a tool that has already watched them". ## Four ways to give it that tool Paste pieces in. Screenshots, a pasted transcript, a short clip. This works for one question about one recording. It doesn't scale, and you have to know where to look before you ask. A filesystem connector. Several MCP servers let an assistant list and read files. For video, the assistant gets names, dates and folders, which helps if your naming is disciplined and does nothing for "the part where the customer explains why they switched". [Pointing an assistant at your files](https://vivu.ai/guide/using-claude-to-organize-files) runs into this limit quickly. Transcripts plus search. Transcribe every recording, store the text with timestamps, and let the assistant search it. For spoken content this goes a long way, and [searching video by what was said](https://vivu.ai/guide/search-video-by-spoken-words) covers how to set it up. Anything that's only visible, like a demo screen or a speaker's reaction, stays out of reach. A hosted video search connector. You upload recordings to a service that indexes both picture and sound, and the assistant calls it over MCP. It covers the most ground. In exchange, the footage lives in that service and searches count against an allowance. ## Search your webinar and event recordings from an AI chat [Vivu](https://vivu.ai/mcp) connects to any assistant that supports MCP, such as Claude or ChatGPT. Once your webinars and event recordings are uploaded to a Vivu project through the secure upload page and indexed in the cloud, you can ask for a moment by its shape without knowing the words, for example "find a blunt one-line instruction telling the audience exactly what to do". The results are short time ranges you can open and preview on the results page, each with a one-line reason, and a range that short usually needs some room added before and after before it goes into a cut. The query is read literally, so if you ask for a speaker who "corrects" another panelist, you may get someone fixing a scheduling mix-up. When you want a real disagreement, name what it's about. In the same conversation, the assistant can turn the moments it found into a list of post or email topics for whoever writes them. [Pulling customer pain points from webinars](https://vivu.ai/guide/how-to-find-customer-pain-points-in-webinars) walks through that handoff in more detail. ## When this isn't the right route Vivu searches one project at a time, and recordings have to be uploaded and indexed in the cloud before an assistant can find anything in them. Searches draw on an allowance (there's a free tier and paid tiers, with limits on Vivu's pricing page), and results are time ranges you open, not frame-exact timestamps. If your questions are almost all about what someone said, a transcript search is simpler and may be all you need. If you only search a recording once in a while, scrubbing through it yourself is still a reasonable answer. ## So, can it? Yes, once the assistant has the right tool, and which tool is "right" depends on your questions. If they're about words, give it transcripts. If they're about what people did, showed or looked like on screen, it needs a connector to something that has indexed the video itself. ## FAQ ### Can ChatGPT or Claude search videos on my shared drive? Not by content. With a file connector, an assistant can see names, dates and folders on a drive, but it can't tell what happens inside a video. Searching by content needs the recordings to be indexed by a transcription or video search service that the assistant can call. ### Why did a video search come back empty when I know the shot exists? An empty result can mean the shot isn't there or the query was too strict. Event recordings shot mostly wide may have no single-speaker close-ups at all. Loosen the description once, or search for a related shot you know exists, before concluding it's missing. ### Can an assistant find an exact quote in a recording? Not reliably. Search by description finds moments that match what you describe, and a transcript search finds words, but a paraphrase in your memory may not match what was actually said. Search for the topic, then check the recording before quoting anyone.