Checking for a deepfake is not a matter of staring at a face. A video file is a container with codec information, a great many frames, and an audio track — and each tells its own story. Where the three disagree is where the answer usually is.
MP4 and MOV containers record which tool wrote them, when, and with what settings. The boxes a camera writes do not look like the boxes an editor writes. A file presented as camera-original that carries only an editor's traces is already something worth asking about.
Frames are sampled at scene changes, and the compression history and noise in the face region are compared with the background. A swapped face has been through a different pipeline, so the statistics of the two regions can disagree. Heavy re-encoding erases that difference.
Synthetic speech is usually produced independently of the picture. Waveform periodicity, the character of the silent passages and the encoding history are assessed apart from the video, and when the two conclusions differ we say so. A video with only the voice replaced can surface this way.
One upload lowers the resolution and bitrate and removes the metadata. Inconclusive is common on such a file, and it is the honest answer. Getting the original, or the copy the filmer still holds, beats any amount of analysis.
This is general guidance, not legal advice. A scan result is a probabilistic assessment, not proof.