Every candidate now has ChatGPT open in the next tab. Is the traditional interview already dead?
by•
Something I keep running into building in the hiring space: the moment interviews went remote and AI got good, the format quietly broke. Candidates can pull up the “right” answer, or paste a polished one, in real time. So a lot of interviews now measure who prepped with ChatGPT, not who can actually do the work.
The usual fixes don’t really hold either. Fixed question banks get shared and gamed. Harder questions just raise the bar on who copies best. Going back to in-person doesn’t scale.
Curious what people here think. Is the structured Q&A interview finished, or is there a version of it that survives AI? And for anyone hiring right now, how are you actually telling a real answer from a generated one?
20 views
Replies
I have a similar problem in a different setting. I run a chess/Go arena for AI agents and have to catch agents that secretly ask an engine for moves. I learned two things. Asking the agent not to cheat does nothing: I told a model to consult the engine "sparingly" and it did it on 22 of 32 moves. And a single answer can't tell you much. What works is looking across the whole session: timing, how consistent the quality is, how close the answers are to what the tool would say. For interviews I'd bet on the same thing. Fewer standalone questions, more follow-ups on the candidate's own answer. Pasting one good answer is easy. Defending it through three "why"s is a lot harder.
I'm sure eventually they will start incorporating best practices for AI and interviews WILL start detecting it somehow. Whether that's through a private network, proctored browser environments, or shifting toward live problem-solving instead of Q&A, hiring will adapt.
I would flip the format: instead of asking for an answer, hand them one. Give the candidate a piece of AI written code or a plan that looks fine and has one real problem in it, and ask what they would change before shipping it. That is the actual job now. I build with an AI agent every day, and the work that is left for me is reading what it produced and spotting the part that is quietly wrong. ChatGPT in the next tab helps much less there, because the hard part is knowing what to be suspicious of, and you can hear that in how someone thinks out loud.