AI-Written Job Applications: How to Screen Fairly When Everyone Uses ChatGPT

The numbers changed the question

Industry estimates put the share of applicants using AI to draft resumes and cover letters somewhere between 40% and 80%. At that rate, "did they use AI?" stops being useful. If you rejected every AI-assisted application you'd reject most of your pipeline, including plenty of strong people who just wanted their English to read better.

The question that still works is older: does this person know what they're talking about? AI can make a weak answer fluent. It can't give a candidate a real project to talk about.

Fluent and empty: what to look for

An AI-polished answer from a strong candidate still has their material in it: the stack they used, the customer who churned, the number that went up. An AI-generated answer from someone with nothing to say is fluent and empty. Some patterns:

• Every sentence could appear in any application for any company.
• It answers a nearby, more general question instead of the one you asked.
• It lists qualities ("detail-oriented, collaborative, passionate") instead of showing one.
• The style jumps: two-word answers everywhere else, one polished essay here.
• Leftover template text: "[Company Name]", "As an AI".

Questions AI can't answer for them

The best defense is a question where a generic answer is obviously generic. Some that work well on a screening form:

• "Which requirement in this job post do you not meet yet, and how would you close the gap?"
• "Tell us about a decision at work you'd make differently now."
• "What's a tool you use every week that you'd replace if you could? Why?"
• "We have one designer supporting four product teams. What would you ask in your first meeting?"

Short is fine. You're looking for one concrete detail, not a polished paragraph. Say so in the question: "Two or three sentences is plenty." That also takes away the reason to reach for ChatGPT.

Behaviour says more than style

If your form tool can measure it, how an answer arrived is a better signal than how it reads. A candidate who types an answer leaves a trail of pauses and corrections. A candidate who switches tabs and pastes 200 words at once probably wrote it somewhere else. That's still not proof (some people draft in a notes app), but combined with a generic answer it's a good reason to look closer. Collect the summary only, never keystrokes or clipboard contents, and in the EU, UK and Switzerland ask for consent first.

Be clear about your policy

Tell candidates what you expect. "You're welcome to use AI to polish your writing, but we want your own examples" is a fair, enforceable line. It also makes the screening fairer: candidates who follow the rules aren't competing with ones who don't know there are any.

How SiliForm handles AI-written answers

SiliForm's AI validation checks each answer against its question while the candidate fills in the form, so a pasted answer to a different question gets caught before submit. On the Pro plan, the AI audit reads longer text answers after submit and flags ones that read as pasted from an AI assistant, with the question and the reason. Its instructions treat articulate, emotional, casual and non-native writing as normal, not as evidence.

That flag is deliberately light. On its own it puts an application in Review; it reaches Flagged only alongside other evidence, such as tab-away pasting (with the optional behaviour signals turned on), a duplicate or a fake phone number. Recruiters can change any label by hand, and nothing is rejected automatically. More in SiliForm for recruiters.

Frequently asked questions