Your Applicants Are Using AI to Write Their CVs. Here's How to Screen Anyway
A CV is a claim, and until recently the quality of the claim carried information. A well-structured document with clean formatting, tight bullet points, and quantified achievements suggested someone who was organised, literate, and had taken the application seriously.
That inference no longer holds. Any candidate with a phone can produce a well-structured, tightly written, quantified CV in about four minutes. Many do. The signal has not weakened; it has been removed.
This is not a candidate integrity problem. Nobody is cheating by using a writing tool to write. It is a screening design problem, and it belongs to the employer.
What has actually broken
Traditional screening signals lose value when applicants can mass-produce polished documents or use real-time assistance during interviews. Relying entirely on filters that reward polish now systematically selects for candidates who are good at producing applications rather than candidates who can do the job.
Three specific things stopped working:
Cover letter quality. Previously a decent proxy for communication skill and genuine interest. Now a proxy for nothing.
Keyword match against the job description. Candidates paste your posting into a chatbot and get a CV mirrored back at you. Keyword density is now a measure of effort, not fit.
Achievement bullets with numbers. "Increased conversion by 34%" reads as evidence. It is evidence only if it happened, and the number is generated as easily as the sentence.
Six things that still work
1. Ask about a specific decision, not a general responsibility
Bad: "Tell me about your experience managing a sales team."
Better: "Describe a time you decided not to pursue a deal you could have won. What was the reasoning, and what did your manager say?"
The second question requires a real memory. Generated answers to it are generic in a way that is easy to hear, because the specifics that make a real answer convincing are exactly what a model cannot invent about your candidate's life.
2. Introduce material the candidate cannot have prepared for
Give them something flawed and ask them to evaluate it. A budget with an error in it. A short piece of code that runs but does the wrong thing. A campaign brief with a contradiction between the objective and the metric. A candidate account statement that does not reconcile.
You are testing judgment against unfamiliar material rather than recall of a prepared script. This is the single highest-yield change most employers can make.
3. Follow up twice on any claim that matters
The first answer to "tell me about the migration you led" will be smooth. The second question, "who pushed back on it and what was their objection," and the third, "what would you do differently," are where the difference between lived experience and constructed narrative shows.
Two follow-ups. Every time. Make it standard for the whole panel.
4. Use a short, real work sample
Not a four-hour take-home. Nigerian candidates are frequently juggling a current job and a commute, and long unpaid exercises select for people with free time rather than people with skill.
Thirty to sixty minutes, paid if it is longer, and as close to the actual work as you can get. A one-page account of how they would approach a problem you actually have. A single spreadsheet task. One recruitment brief written to a real vacancy.
The point is not the artefact. It is that you can then discuss it, and the discussion is where the assessment happens.
5. Weight the reference check properly
Reference checks in Nigeria have a reputation for being pro forma. They do not have to be. One specific question does most of the work:
"If you were building a team again, and this role came up, would you hire them again? Why or why not?"
The pause before the answer is data.
6. Structure the interview and score during it, not after
Same questions, same order, scored on a written scale by each interviewer before the panel talks to each other. Memory reconstructs itself around the impression you formed in the first two minutes. Scoring in the moment limits that.
What to stop doing
Stop treating typos as disqualifying. A CV with a typo in 2026 is more likely to have been written by the candidate than one without.
Stop asking for cover letters unless you are going to ask a specific question in them, and even then expect the answer to be assisted.
Stop trying to detect AI writing. Detection tools have high false-positive rates and they flag non-native English speakers disproportionately. In a Nigerian applicant pool this is an active fairness risk and it will cost you good candidates. Change what you measure instead of trying to police how the document was produced.
A note on assisted interviews
Some candidates now use real-time assistance during remote interviews. If this concerns you, the fix is not surveillance software. It is question design. Questions that require reference to the candidate's own history, questions that require reacting to something you introduce mid-conversation, and questions that require them to think aloud about an ambiguous situation are all resistant to real-time assistance, because the assistance has no access to the candidate's memory and lags behind live conversation.
An in-person final round for senior roles is also reasonable, and it costs you less than a bad hire.
The underlying shift
Screening used to be a filtering exercise: reduce a hundred documents to five. It is becoming an evidence-gathering exercise: get a small amount of real signal from each of a larger number of people.
That is more work per candidate, which is why the mechanical parts of the process are worth automating. Structured intake, knockout questions on genuine must-haves, and self-scheduling free up the hours you now need to spend on judgment.
The employers who will hire well over the next few years are the ones who moved their effort from reading applications to designing better questions.
Related reading
- 88% of companies now screen with AI. What it means for Nigerian employers
- A practical bias audit you can run on your own shortlist
- Time-to-hire benchmarks for Nigerian roles
CTA: BetternshipHR handles structured intake and interview scorecards so your team can spend its time on the questions that matter. Start free.
FAQ
Can I tell if a CV was written with AI? Not reliably, and detection tools produce false positives at rates high enough to be unfair, particularly for candidates writing in a second language. Design your screening so it does not depend on the answer.
Should I ban AI use in applications? A ban is unenforceable and signals that you have not updated your process. It is more useful to state plainly that you will assess through structured interviews and work samples.
What is a fair length for a work sample in Nigeria? Thirty to sixty minutes unpaid is generally accepted. Beyond that, pay for it. Long unpaid exercises filter for free time rather than ability.