Recruiters now read hundreds of applications that were clearly written by the same invisible hand. The sentences are fluent, the verbs are strong, and nothing in them is memorable, which is a worse outcome than a plainly written CV with real numbers in it. Meanwhile the same tools, used differently, can save you hours and genuinely improve a weak application.

The difference is what you delegate. AI is good at structure, phrasing, and finding what you left out. It is bad at knowing what you did, which is the only part that matters. This guide covers what to hand over, prompts that produce usable material, how generated text gets spotted, what you must always write yourself, and where the real risks sit.

What is AI actually good at here?

Three jobs, all of them editorial rather than creative. The first is reorganising material you already have: turning a messy list of duties into tight bullet points, or cutting a rambling summary to three lines. The second is gap-finding, since a tool asked to compare your CV against a job advert will reliably spot requirements you failed to address. The third is variation, producing five ways to phrase one achievement so you can pick the one that sounds like you.

What it cannot do is supply the content. A large language model predicts plausible text, so when you ask it to describe your accomplishments without giving it facts, it produces confident sentences about work you never did. That is not a small stylistic problem, it is a lie on a document that gets checked in a reference call. Feed it your real material and it becomes an editor; ask it to invent, and it becomes a liability.

Prompts that produce usable material

Generic prompts get generic output. Give the tool the raw facts, the target job, and a constraint, then iterate.

  • Rewrite with facts supplied: "Here are five things I did in this role, with numbers. Turn each into one bullet point under 20 words, starting with a verb, keeping every number exactly as written."
  • Gap check: "Here is the job advert and here is my CV. List the requirements my CV does not address, in order of how prominent they are in the advert."
  • Compression: "Cut this summary from 90 words to 40 without removing any number or job title."
  • Keyword check: "Which terms from this advert are missing from my CV? Do not add any I cannot back up."
  • Interview prep: "Based on this CV and advert, give me the eight questions most likely to be asked, hardest first."

Two rules keep the output honest. Always give it the numbers yourself, and always ban invention explicitly, because a model asked to make a CV "stronger" will happily add a percentage that came from nowhere. Then check every line against reality before it goes anywhere near an employer.

Iterate in small pieces rather than asking for a finished document. One bullet at a time, one section at a time, with you deciding what survives, produces a CV that still reads as yours. Asking for a complete rewrite in a single request produces something fluent that you will then have to unpick line by line, which takes longer than doing it properly and usually leaves at least one invented detail behind.

How do recruiters spot AI writing?

Not by using a detector, which is unreliable, but by pattern recognition after reading the same phrasing forty times in a week. The tells are consistent: uniform sentence length, ornamental adjectives, abstract claims with no numbers, and a handful of words that generated text loves and humans rarely choose. "Spearheaded", "leveraged", "meticulous", "passionate about driving impact" and "proven track record" now function as signals rather than as description.

The bigger giveaway is emptiness. A generated bullet says "streamlined operational processes to enhance efficiency", while a human one says "cut invoice approval from six days to two by removing a duplicate check". The second is shorter, more specific, and impossible to have been written by someone who was not there. Anything a stranger could have written about your job is a line worth rewriting, and that test catches AI text and lazy human text equally well. The same applies to generative artificial intelligence in cover letters, where a fluent paragraph that never names anything specific about the company reads as automated even when a person typed it.

What you should always write yourself

Write your own achievements, in your own plain words, before any tool touches them. It does not matter if the grammar is rough, because that is the raw material and only you have it. Sit down and answer three questions for each role you held: what did you own, what changed because you were there, and what number can you attach to it. That list is the whole value of your CV.

Also write the specifics that prove attention: why this company, what you noticed about their product, the name of the person who referred you, the detail from the job advert you can speak to. Those sentences carry the entire cover letter, and delegating them defeats the purpose, since the reader is checking whether you bothered. Keep your final voice too. Read the CV aloud, and change anything you would never say in a sentence, because that same document becomes the script for your interview answers, including your resume summary and your account of your strengths. If your CV sounds like someone else, the interview will expose the gap in about four minutes.

Where are the real risks?

Three, in order of how often they bite. Invented detail is the first: a percentage, a title, or a date the tool improved for you, which becomes a fireable problem if it survives to a reference check. Verify every figure and every job title against your own records before sending.

Confidentiality is the second. Pasting your current employer's internal documents, client names, unreleased figures, or anything under a non-disclosure agreement into a public tool is a real breach, however convenient. Describe the work in general terms and keep the identifying details out. The third is sameness. If your application is optimised toward the same phrasing everyone else is generating, you become indistinguishable at precisely the moment you need to be memorable, and screening still rewards the specific over the polished, as covered in getting your resume past an ATS.

A workflow that works

Do it in this order. Write your raw achievement list yourself, without help, and include every number you can verify. Run the gap check against the specific job advert and note what is missing. Rewrite the missing parts yourself if you have the evidence, and drop them if you do not. Then use the tool only to tighten phrasing and to cut length, one bullet at a time rather than the whole document at once.

Finish with three human passes: read it aloud for voice, check every number against a payslip, report, or email, and delete any sentence that could describe someone else's job. Then use the same tool for the part where it genuinely shines, which is rehearsal: ask it for the hardest likely questions and practise your answers out loud, alongside the routine in interview preparation. Used that way, AI saves you the two hours you used to spend rewriting bullet points, and none of the credibility you would lose by letting it speak for you. If you are also applying for a career change, keep the same discipline about honest framing described in changing careers without starting over. More guides are in the article library.