How Do You Ask Gemini About a Long Video or Recording?
Upload it with Add files, and ask for specifics with timestamps in MM:SS form. Without Google AI Pro or Ultra, you get 5 minutes of video or 10 minutes of audio. Those plans raise it to 1 hour of video and 3 hours of audio. Each video can be up to 2 GB.
This is a [recording of a meeting between X and Y / lecture on Z]. Give me: 1. A summary of up to [5] bullets. 2. Every decision made, with the timestamp in MM:SS. 3. Every action item, with who owns it and the timestamp. 4. Anything shown on screen that wasn't said aloud. If you can't make out a word or a speaker, write [unclear] and give the timestamp. Don't guess names.
Try it now
- 1Click Add files, then Upload Files, and select a recording within your plan's limit. Paste the prompt above.
- 2Jump to two of the timestamps it gives. Check that the moment really is there.
- 3Ask a follow-up about one moment, such as "What did they say at 14:30 about the budget?"
- 4You'll know it worked when the timestamps lead you to the right spot.
Use it when
A recorded meeting, a lecture, a customer call or a how-to video where you want decisions, action items or one moment. Reach for it when listening again would take longer than checking the answer.
Skip it when
Anything where one word matters and you can't check the recording, such as a legal statement or a medical instruction. Also think twice about recordings of other people. Uploads are saved under Keep Activity and can be reviewed. Google says audio and Live video aren't used by default, but check your settings.
Why it works
Gemini takes in both picture and sound, so it can answer questions about what's on screen as well as what's said. On the API, Google's docs say video is sampled at 1 frame per second by default, and costs about 100 tokens per second at default resolution, about 300 at high resolution. A 1 million token window fits about 3 hours at low resolution or 1 hour at high. Google warns that fast motion or quick scene changes may lose detail at the default rate. So a slide talk is a good fit, while a fast-moving sports clip is less reliable.
Example
Say you have a 45-minute recording of a client workshop, which needs AI Pro or Ultra at this length. Attach it, then ask for decisions and owners with timestamps, and for anything shown on the shared screen. When it says that the launch moved to June at 31:12, go there and listen. Anything marked [unclear] is a place to check by hand, and a good reason to ask for it to be marked instead of guessed.
Going further
For YouTube, Google says Gemini can find a video and answer questions about it, for example "Find videos on X", then "Summarize the first video". It requires Keep Activity on, and doesn't work in Live chats. On the API, you can pass public YouTube URLs. For long videos, Google offers "processing": "agentic", which it says can use up to 88% fewer tokens on long-form content. Static mode lets you set start_offset and end_offset, or fps, for example 0.5 for a frame every two seconds.
Common mistakes
- Uploading a 2-hour video on a plan with a 1-hour limit, and assuming the 2 GB size limit is the only one.
- Trusting timestamps without checking two of them. A confident wrong time is easy to miss.
- Expecting Gemini to catch fast action at 1 frame per second. On the API, raise
fps, or send a shorter clip. - Letting it guess speaker names. Tell it who's in the recording, or have it use [unclear].
Sources
- Upload and analyze files, Gemini Apps Help
- Video understanding, Gemini API docs
- Understand and count tokens, Gemini API docs
- Find and ask about YouTube content in Gemini Apps
- Gemini Apps Privacy Hub
Checked 6 November 2026. Products change often, so confirm details with the vendor.