How an AI interview assistant works during a call

There is little variety in how AI interview assistants work. Knowing the sequence helps you recognize one in use and choose questions it copes with poorly.

Step one: hearing the question

Audio from the computer or the microphone is fed to the assistant, which transcribes what the interviewer says. The transcript is with a language model before the interviewer has finished speaking.

Step two: reading what is shared

A shared slide or coding problem is captured as a screenshot, and the assistant extracts the text. Questions that are never spoken, a written brief or a coding challenge for example, become answerable this way.

Step three: displaying the answer

A small window on the candidate's screen holds the answer. In some tools that window is drawn in a manner screen sharing fails to capture, which keeps it from the interviewer even if the entire screen is shared. Cluely is an example. Alternatives include a second device, for instance a phone propped beneath the webcam, and a browser extension.

The weak points

Every step relies on something in the interviewer's hands.

• An assistant can respond only to what it has read or heard. If it reads the screen wrongly, the answer will not match the question.
• Generating an answer costs it a few seconds. Follow up fast and the candidate has less time for reading.
• General answers are its strength and specifics its weakness. Questions on the candidate's past work, or on a change to the answer just given, are difficult to prepare for.

Questions that exploit the second and third weak points are collected in how to stop AI during an interview with the questions you ask. For the look and sound of a candidate reading from one of these tools, go to signs a candidate is using AI.