Unfog · Guide
Can employers detect AI interview copilots?
Short answer: often, yes — and increasingly it does not matter, because the interview is no longer remote.
What the data says
An analysis of 19,368 interviews by Fabric HQ in 2026 found that 38.5% of candidates showed signs of AI assistance. Of those detected, 45% were using a dedicated copilot tool. Separately, around 20% of US workers admitted to secretly using AI during an interview in 2025 — a figure that is almost certainly understated, because people under-report cheating.
What recruiters actually look for
| Signal | What gives it away |
|---|---|
| Eye movement | Reading rather than thinking — a consistent off-camera glance before each answer |
| Response timing | An unnatural, uniform pause before every answer, regardless of question difficulty |
| Linguistic markers | Written register in spoken form: fluent, structured, and unlike the candidate's own emails |
| Depth on follow-up | The tell that catches most people — a polished answer the candidate cannot then explain |
Some companies now share candidate flagging data across hiring networks, so being flagged once is not necessarily contained to one application.
The bigger problem: the interview moved
According to Gartner research reported by Computerworld, 72.4% of recruiting leaders are now conducting interviews in person specifically to counter fraud. Google, Cisco and McKinsey have all reinstated in-person rounds.
A copilot cannot help you in a room. Whatever its detection risk, its addressable moment is disappearing.
What actually works instead
In-person interviews are judged on the things that cannot be automated. McKinsey's stated reason for the change is to assess judgment, empathy, creativity and connection — all of which you demonstrate through how you speak, not what you say.
- Speaking pace. 120–150 words per minute reads as considered. Above about 165 reads as rushed, which is usually read as nervous.
- Filler density. Under 3 per 100 words is barely noticed. Above 8 becomes the thing the interviewer remembers.
- Pitch movement. Under about 4 semitones of variation reads as flat and disengaged, however good the content is.
- Recovery. Stalling for four seconds costs you more than a mediocre answer delivered steadily.
These are all measurable, and all trainable in a couple of hours of speaking out loud.
Find out where you actually stand. Speak for 30 seconds and Unfog measures your pace, filler rate and pitch movement from the audio. Free, no account needed, and the recording never leaves your device.
Take the 30-second check