Is it obvious when AI has written a job application?
Usually, and the tell is rarely the writing quality. 23 stock phrases and three formatting habits give it away, and the same list turns up again and again: I hope this email finds you well, passionate, leverage, delve, seamless, synergy, thrilled, perfect fit, proven track record, team player, and so on, plus exclamation marks, em dashes and markdown. The more serious risk is not style at all. Asked to sell you, a model reaches for credentials you never mentioned, so it writes that you hold a clearance or a ticket you do not. That gets found at vetting and it follows you.
The stock phrases
23 of them, plus three formatting habits, and the list is stable enough to check a draft against mechanically:
I hope this email finds you well · passionate · leverage · delve · seamless · synergy · dynamic · thrilled · excited to apply · perfect fit · hit the ground running · fast-paced environment · proven track record · results-driven · detail-oriented · team player · I am writing to · utilize · spearheaded · esteemed · keen to · furthermore · moreover
Also out: exclamation marks, em dashes, and any markdown formatting. All three are tells in an email that is supposed to have been typed by a person.
Asking a model to avoid them does not work
It complies for two paragraphs and drifts back, because the banned phrases are what enthusiasm looks like in its training data. The rule has to be enforced rather than requested: check the finished draft against the list and rewrite anything that hits. Doing that automatically is most of what the software behind this site is for.
The risk that actually matters
Style is the small problem. Asked to sell you, a model reaches for credentials you never mentioned, because that is what selling looks like in the text it learned from. It will write that you hold a security clearance because you said Royal Navy, or a ticket because you named the industry.
That gets found at vetting, and it is the one mistake here that follows you. The fix is to write down what is not true about you before you draft anything — a ticket that lapsed, a degree you paused, a clearance that expired, a licence you are working towards — and check every draft against that list as well.
What to keep
Plain English you would say out loud. Confident, warm, slightly informal, no grovelling. The test is whether a tradesman would read it aloud without wincing.
Where these numbers come from
86 cold emails to real UK employers over four weeks, sent by one person looking for work, logged as they went. 22 replies. The full method and every table is in the playbook, free, no account.
They are engineering and technician roles in Scotland, so a different trade may behave differently. Where a row is thin it says so on the row.
Since then: 171 applications across 137 employers, 32 replies, 19%. Live, and updated after every run.
Type a company, get the real address Other questions
The address finder is free and needs no account.