Inside Goodfit's interview proctoring
A look at the proctoring and interview security layers Goodfit runs on every AI video interview, what each one catches, and when it runs.
Inside Goodfit's interview proctoring
An AI video interview on Goodfit is watched by more than one system. Some checks run before the candidate says a word, some run continuously while they speak, and several more run over the finished recording once the call has ended.
- 01Before the interview4 checks
Is this the right person, on a setup that will hold up?
Device and connection check
Camera, microphone and network are measured before anything starts, and the microphone the candidate proved working is the one the interview then uses. A connection too weak to carry a recorded interview cannot start one.
Live person check
On-screen prompts the candidate has to respond to in the moment. A photograph, a looping video or a face on a second screen cannot complete them.
Face reference captured
A still of whoever passed that check, kept as the reference. Every camera check later in the interview is compared against this exact person.
Government ID matchoptional
Photo identification compared against the reference face, and the name printed on the document compared against the name on the application. In India, candidates can verify through DigiLocker instead of photographing a document.
- 02During the interview4 checks
Are they alone, and unaided?
Interview recording
The full call is recorded on our servers.
Who is on camera
Checked every second, for the whole interview: that a face is present, and that there is only one. A second person entering the frame is recorded with the time they appeared.
Where attention goes
Head and eye position measured against how the candidate sat at the start. Attention held away from the screen is recorded with its duration, once it is sustained.
Browser environment lock
Switching tabs, opening developer tools, copy, paste and print attempts, blocked shortcuts and leaving full screen are each recorded with a timestamp. The whole interview screen is captured alongside them.
- 03After the interview5 checks
Did those answers come from them?
Frame-by-frame camera review
A frame every few seconds, across the entire interview, each one described on its own: who is in shot, where the eyes are, hands at the face, a phone, earbuds, notes or paper, a second person, a photograph or screen held up to the camera.
A second look at the whole sequence
Those frames read again in order, as a run rather than as snapshots. This is what separates someone glancing at their desk from someone returning to the same fixed point again and again.
Voice matched to the video
Lip movement checked against the audio, and the audio checked for how many distinct voices are present.
Answer patterns and response timing
The transcript is examined for the marks of reading aloud: repeated phrases, restarts mid-sentence, answers that carry on past what was asked. Response timing across the whole interview is checked separately for a regularity that live speech does not produce.
Hidden browser assistants
The captured interview screen is scanned for AI assistant tools injected into the page, and separates one merely installed from one actually used during the interview.
Before the interview
Nothing starts until we know two things: that this is the person who was invited, and that the setup in front of them can carry a recorded interview to the end.
The candidate is taken through a short check of their camera, microphone and connection. The microphone they prove working is the one the interview then uses, so the call cannot fall back to a muted built-in mic. A connection too weak to sustain a recorded interview is not allowed to start one.
They are then asked to respond to a few prompts on screen in the moment. A photograph, a looping video, or a face on a second monitor cannot complete them. The still we take at that point becomes the reference for the rest of the interview.
Where a role calls for it, we also match a government-issued photo ID against that reference face, and match the name printed on the document against the name on the application. Both have to agree. Candidates in India can verify through DigiLocker instead of photographing a document.
During the interview
The full call is recorded on our servers. While the candidate is speaking, three things are watched continuously.
Who is on camera, checked every second for the length of the interview: that a face is there, and that there is only one. A second person entering the frame is recorded with the moment they appeared.
Where their attention goes, measured against how they sat at the beginning. Attention held away from the screen is recorded along with how long it lasted, once it is sustained.
What the browser is doing. Switching tabs, opening developer tools, copy, paste and print attempts, blocked shortcuts, and leaving full screen are each recorded with a timestamp. The interview screen itself is captured alongside them, so a reviewer can see what the candidate was looking at.
After the interview
When the call ends, the recording goes through several independent passes.
Frame-by-frame camera review. A frame every few seconds, across the entire interview, each described on its own terms: who is in shot, where the eyes are, hands at the face, a phone, earbuds, notes or paper, a second person, a photograph or a screen held up to the camera.
A second look at the whole sequence. Those frames are read again in order, as a run rather than as snapshots. This is what separates a candidate glancing down at their desk from one returning to the same fixed point again and again.
Voice matched to the video. Lip movement is checked against the audio, and the audio is checked for how many distinct voices it contains.
Answer patterns and response timing. The transcript is examined for the marks of reading aloud: repeated phrases, restarts mid-sentence, answers that continue past what was actually asked. Separately, the rhythm of responses across the whole interview is checked for a regularity that live speech does not produce.
Hidden browser assistants. The captured interview screen is scanned for AI assistant tools injected into the page, and distinguishes one that was merely installed from one that was actually used during the interview.
For what the resulting flags mean and how to review one, see How Goodfit measures candidate integrity.