What happens when every cover letter sounds the same? Hiring managers start adding a poem about a frog.
A growing number of recruiters have started inserting creative non-sequiturs into job applications - brief prompts with no obvious right answer, like writing a short verse or describing something in an unexpected way. The explicit goal is to catch candidates submitting raw AI output without reviewing it.
The "frog poem" has become shorthand for the whole category. A grammatically perfect, well-structured verse about a frog submitted alongside a resume for an accounting role stands out immediately - not because it's impressive, but because it's recognizably what ChatGPT produces by default. The model defaults to AABB rhyme schemes, safe nature metaphors, and a certain relentless optimism. A human asked to dash off a quick poem is more likely to write something awkward, specific, or strange.
The Screen Works Because Most AI Applicants Are Careless
The candidates getting caught are not the ones who've integrated AI thoughtfully into their workflow. They're the ones treating it as a copy-paste cheat code. Someone who uses Claude or ChatGPT to draft and then edits heavily - adjusting tone, cutting the generic filler - will pass most creative screens. The ones flagged are submitting the literal first response with no edits.
This means the screen is more a filter for carelessness than a filter for AI use. A careful AI user passes. A nervous human who doesn't write well might also pass the poem and fail the cover letter. The signal is inconsistent, and it doesn't obviously predict job performance.
There's also a real problem with deploying creative writing prompts as hiring screens without thinking through who they disadvantage. Non-native speakers, candidates with writing anxiety, and people applying for roles where creative writing has nothing to do with the actual work face friction that doesn't reflect their ability to do the job.
What's Actually Being Filtered Out
What hiring managers are really dealing with is a volume problem. AI-assisted application tools let a single person apply to hundreds of jobs in an afternoon, which means recruiters are seeing far more submissions for the same number of genuine candidates. Creative prompts add friction - they slow down the spray-and-pray applicant. But they don't find better candidates.
The frog poem also has a short shelf life. As the practice becomes widely known, anyone using AI for applications will know to edit the poem before submitting, or to ask their AI tool to write something intentionally imperfect. The underlying dynamic - AI tools collapsing the effort cost of mass applications - doesn't go away just because one screen gets harder to game.
The more durable approach is structured work samples or short async assessments tied directly to the role. Those are harder to fake with AI (for now), actually test relevant skills, and don't disadvantage candidates based on whether they can write a passable haiku under pressure. The frog poem buys time. It isn't a long-term answer to what's happening to hiring.