Most AI interview assistants work in the same way. Knowing how they work makes it easier to spot them and to plan questions they have trouble with.
Listening to the question
The assistant listens to the computer's audio or the microphone and turns the interviewer's question into text. By the time the question has been asked, it has already been sent to a language model.
Reading the screen
When the interviewer shares a slide or a coding problem, the assistant takes a screenshot and reads the text in it. This lets it answer questions that are never said out loud, such as a written brief or a coding challenge.
Showing the answer
The answer appears in a small window on the candidate's screen. Some tools draw this window in a way that screen sharing doesn't capture, so the interviewer doesn't see it even when the candidate shares their whole screen. Other tools run on a second device, such as a phone placed under the webcam, or in a browser extension.
Where the tools are weak
Each step depends on something the interviewer controls.
- The assistant can only answer what it can read. If the text on screen is hard for a machine to read, or the screen shows a question the interviewer didn't intend, the assistant answers the wrong question.
- The assistant needs a few seconds to produce an answer. Quick follow-up questions leave the candidate less time to read it.
- The assistant's general answers are good, but it has trouble with specifics. Questions about the candidate's own past work, or about a change to their last answer, are hard to prepare for.
dontapplyai is built around these weaknesses. The question you share is drawn so that people can read it easily and AI tools have a much harder time, and a decoy question is shown beside it. A tool that reads the screen often answers the decoy, and an answer that doesn't fit the real question is easy to notice.
