When should recruiters check for AI in applications?
Recruiters need to detect AI in job applications when the application itself is the evidence of skill, or when the candidate’s identity is in doubt. A polished cover letter written with help from a chatbot is rarely the real problem.
- Writing roles, where the cover letter or writing sample is the test of the job itself.
- Take-home tests and coding exercises that are meant to show how a candidate thinks.
- Resumes that list experience in suspiciously generic terms, with no checkable detail.
- Remote video interviews where something feels off: lips out of sync, frozen expressions, a face that blurs when the person moves.
- Phone screens or recorded voice answers in high-volume recruiting.
What are the risks of AI-generated cover letters and deepfake interviews?
The real risks are hiring someone who cannot do the job, and hiring someone using a false identity. The FBI’s Internet Crime Complaint Center warned in a June 2022 public service announcement that applicants were using deepfakes and stolen personal information to apply for remote IT and programming roles, some with access to customer and financial data. Reported signs included lip movements that did not match the audio.
The opposite risk is just as real. Rejecting a strong candidate because a tool flagged their cover letter costs you the hire and can be unfair, especially to non-native speakers whose writing detectors misread more often. In some jurisdictions, decisions based solely on automated processing are regulated: Article 22 of the EU’s GDPR is one example. Check with your legal team before any tool feeds into a decision.
A short AI policy for applicants
A few lines in the job posting prevent most disputes. Adapt these to your roles:
- AI tools may be used to improve the wording of your resume and cover letter. Everything you state must be true and you should be ready to discuss it.
- The take-home exercise must be your own work. If you used an AI assistant for any part, say which part and how.
- Interviews are held with camera on. We may ask simple questions to confirm we are speaking with you.
- We review all applications with people. Tools may help us decide what to ask; they do not decide who is hired.
A practical workflow to detect AI in job applications
The workflow below checks skill and identity at the stages where they matter, and keeps a human in charge of every decision.
- 01Publish your AI policy in the job posting and the take-home instructions.
- 02Screen resumes on verifiable facts: employers, dates, projects, links to public work. An AI written resume checker adds little if the facts check out.
- 03For writing roles, run the writing sample through the AI text detector. Use a “probably AI” or “mixed” result to prepare questions, not to reject.
- 04For coding roles, check the take-home with the AI code detector, then hold a live session where the candidate explains a function, changes a requirement and debugs one case.
- 05In video interviews, ask for camera on and make simple live requests: turn your head, hold up the item on your desk, answer a follow-up that depends on the previous answer.
- 06If something seems off and the candidate consented to recording, analyze a clip with the deepfake video detection tool or the voice detector.
- 07Confirm identity through your normal background check before an offer, and document the reasons for every decision.
Recording laws differ by country and state. Always tell candidates when an interview is recorded and why.
Design a take-home test that AI cannot do for the candidate
The best defense against a take-home test written by AI is a test where AI help does not replace understanding. Detection then becomes a minor step.
- Base the task on a realistic, slightly messy scenario from your own work, with constraints that are not in any tutorial.
- Ask for a short written note on the trade-offs the candidate made and what they would do with more time.
- Plan a live extension: in the interview, change one requirement and ask them to adapt their own solution.
- Keep it short. Long unpaid exercises push even honest candidates toward shortcuts.
The guide on how to detect AI-generated code explains why code detection is harder than text detection, and why a walkthrough beats any score.
Check a take-home test before the follow-up interview
Running the take-home through a code detector before the live session helps you choose which functions to ask about. Focus your questions on the parts flagged as probably AI and see whether the candidate can explain and modify them.
AI Code Detector · Free plan
Check a take-home coding test function by function, then use the result to plan your live walkthrough.
Which AnyFact detectors fit recruiting, and what do they return?
Five detectors map onto the hiring funnel. Each gives a verdict of probably human, uncertain or probably AI, with the signals behind it, a plain-language explanation, its limits and a PDF report you can file.
| Stage | Detector | What it returns |
|---|---|---|
| Cover letter, writing sample | Text (300 characters or more) | A verdict per passage; “mixed” when only part looks generated |
| Resume or portfolio file | Document (PDF, DOCX, PPTX) | A page-by-page result for text and embedded images, plus file metadata |
| Take-home coding test | Code (30 to 5,000 lines) | A score per function; beta, the least reliable detector |
| Recorded video interview | Video (up to 3 minutes) | Suspicious moments on a timeline, a face-swap check when a frontal face is clear, and the soundtrack checked as voice |
| Recorded phone screen | Voice (up to 5 minutes) | A score per segment and a transcript |
Resumes are formulaic by nature, so a document verdict on a resume is weak evidence. It is most useful for writing samples and portfolios that are supposed to be original.
AI Video & Deepfake Detector · Free plan
Analyze a recorded interview clip and see suspicious moments marked on a timeline.
What detectors cannot tell a hiring team
A detector cannot tell you whether a candidate is good at the job, whether their use of AI broke your rules, or who they really are. It reads statistical and visual signals. Short texts, compressed video calls and low-quality audio make those signals weaker.
Treat every result as a prompt for a better question. The guide on how to spot a deepfake video lists what to watch for live, which is often more useful than any tool during an interview. For the bigger picture on error rates, read are AI detectors accurate.
Frequently asked questions
Is it wrong for a candidate to use AI on a cover letter?
Not necessarily. Many employers accept AI help with wording, as long as the facts are true and the candidate can back them up. Decide your position, write it in the job posting, and judge candidates on skills you can verify.
Can you tell if a resume was written by AI?
Sometimes, but it rarely matters. Resumes are short and formulaic, which makes detection unreliable. Checking the facts (employers, dates, public work) is a better use of your time than running an AI written resume checker.
How do you spot a deepfake job interview?
Watch for lips out of sync with speech, blurring around the face when the person moves, and reactions that lag. Ask for simple live actions and unscripted follow-ups. If you recorded the interview with consent, a video detector can mark suspicious moments for a second look.
What should we do if a take-home test is flagged as AI?
Hold a live session. Ask the candidate to explain the code, change a requirement and fix a bug. If they understand the work, the flag does not matter much. Never reject on the detector result alone.
Does AnyFact offer an API for applicant tracking systems?
Not yet. AnyFact works through its web app, which suits case-by-case checks on shortlisted candidates rather than screening every application automatically.