How to Make a Resume With AI That Actually Gets Noticed

My cousin Bilal applied for forty jobs last year. Forty. Same resume, every time, sent out like leaflets through a letterbox.

Replies: none. Not even a rejection. Just silence, which is worse, because silence gives you nothing to fix.

So one evening he turned up at my kitchen table with his laptop and the embarrassment of a man who suspects the problem might be him. We made chai. It went cold — nobody drank it until the resume was unrecognisable. That started with me reading his opening line out loud, in my most serious interview voice:

“Hardworking team player with excellent communication skills.”

He winced. Good. That’s the correct reaction. So is everyone, Bilal. So is the chair you’re sitting on, if you ask it nicely.

A resume being reviewed and improved on a laptop

Six seconds, and a robot goes first

Here’s the thing most people never get told: your resume is read twice, and the first reader isn’t a person.

Bigger employers run applications through an applicant tracking system — ATS — which files your resume and lets recruiters search it by skills and job titles. If the software can’t read your file, or your words don’t match their search, you’re finished before a human ever gets curious about you.

Survive that, and you get the human. The tired human. Forty tabs open, coffee going cold, giving you six seconds and scanning for three things: what you did recently, whether it matches this job, and whether there’s any proof you were good at it.

Raw material first, then — because the worst possible start is the one he’d been planning: opening ChatGPT and typing “write me a resume.” Instead I made him dig, on the back of an envelope, actual pen:

  • Job titles, employers, dates — checked twice, because his dates were wrong on one job by four months
  • Three to five real achievements per role. Achievements, not duties
  • Numbers. Any numbers. Customers a day, records cleaned, money, hours
  • One actual job advert he wanted, the whole thing, skills list included
  • Certificates he had genuinely finished. Genuinely comes back later.

The master copy, and the leash

We pasted his scruffy resume in and asked for structure only. Exactly this:

“Reorganise this resume into clear sections with strong action verbs. Do not add skills, employers, dates or achievements that are not already in my text. Flag anything that is vague.”

That instruction matters more than the rest combined — left alone, these tools embellish: smooth, impressive, faintly untrue. What came back was a master version: tidy, complete, still recognisably him, and every application after that would be a copy of it, adjusted. “That’s it?” he said. That’s the skeleton. Now the part everyone skips.

One job. This job. Only this job.

We opened the advert — a customer support role — pasted it in next to the master, and asked:

“Compare my resume with this job description. List the required skills I already show, the keywords I use in different words, and the requirements I do not meet. Do not invent experience to fill gaps.”

And there — there — was half his forty silences, in plain sight.

“The advert says client onboarding,” I said. “What did you write?”

He scrolled. “Setting up new customers.”

“Well. There’s your problem.”

Same job, different words — and software is literal in a way no human would ever be. Swapping in the advert’s wording, where it’s honestly true of you, isn’t cheating. It’s translation.

The requirements he didn’t meet stayed off — that line in the prompt, do not invent experience to fill gaps, was law all evening. Mostly. Hold that thought.

“Responsible for helping customers.” How many customers, Bilal?

Then his bullet points — where resumes go to be wallpaper. His actual line, word for word: “Responsible for helping customers and answering questions.”

“How many?” I asked. “A day. Roughly.”

Thirty, forty, it turned out — phone and email — and the billing ones, the angry ones, he passed to finance the same day because they went nuclear if left overnight. All of that was in his head. None of it was on the page. After a small argument the line read: “Resolved 30–40 customer enquiries a day by phone and email, escalating billing issues to the finance team the same working day.”

The rest went the same way — “how many, how often, what came of it?” on repeat. “Worked on social media posts for the company” became “Wrote and scheduled 12 social media posts a month, following the manager’s content plan and checking every post before it went out.” And “Did data entry and helped the team” — helped the team do what, Bilal — became “Entered and checked 500+ customer records in the CRM, clearing out the duplicates found in a two-week data clean-up.”

Nothing invented, nobody promoted to genius — but every number got interrogated, because AI rounds your life upwards. Could you prove it today, if asked? Then it stays. If not: a range you can defend, or out it comes. A recruiter will ask.

The summary was a horoscope — “hardworking and motivated professional.” We replaced it with three lines of actual fact: “Customer service assistant with three years in retail and phone support. Handled high enquiry volumes in a CRM, trained two new starters, and now looking for a full-time support role in a larger team.” The AI drafted three versions from his real experience; the final one he assembled himself, because the choosing is what makes it sound like a person.

The Excel incident

It nearly went wrong about an hour in: midway through the polishing, the tool took his honest line “helped with Excel reports” and upgraded it — quietly, helpfully — to “advanced Excel and financial modelling.”

Bilal wanted to keep it. “I do use Excel—”

“Can you do financial modelling?”

“…Not as such.”

Then it comes out. And it did, under protest. And be clear about why, because the tools will never warn you: it doesn’t feel like lying when a machine drafts the lie and you just fail to delete it, but it is lying — and it works for about eleven minutes, until somebody says “tell me about the financial modelling.” Employers check references; the gap between the paper and the person is the first thing a good recruiter smells. Describe real work better all you like. Invent work, and it collapses the first time anyone leans on it.

Boring formatting things we fixed while arguing about fonts

The rest was dull and decisive. His name and phone number lived in the header, the layout was a two-column thing copied off the internet, and the file itself — I wish I was inventing this — was called final_v3_REAL.pdf. So: standard headings — Experience, Skills, Education — because both readers look for things where they expect them. Details out of the header and fancy columns, where systems read them late, jumbled, or never. Abbreviations spelled out once — “Customer relationship management (CRM)” catches both searches. File renamed Firstname-Lastname-Resume.pdf, and his plan to stuff “sales, sales, sales” into a skills line talked out of him: keyword-stuffing fools nobody. PDF suited this advert; the next asked for Word, so Word it got. Obeying the advert is itself a tiny test.

A fortnight later

Three interviews. Same man, same work history. Three.

He didn’t get the first. The second reached a final two. The third — he interviewed well, and I refuse to jinx it by saying more.

What does he still do wrong? Tired, he still fires the master copy at a job unchanged — the leaflet habit, creeping back at 1 a.m. — and his file names have relapsed: I have seen resume_final_FINAL2.pdf, and I have not forgiven it. But he tailors now, mostly. He checks the numbers. And he deletes the financial modelling, every time.

Questions people actually ask me

Can AI just write the whole thing from scratch?

It can produce a document — smooth, generic, about a person who doesn’t exist. Bilal had been sending that document for a year. Start from your own history, let the tool polish, and check every fact like it owes you money.

Which tool should I use?

ChatGPT and Claude both handled everything here fine; the prompts above work in either.

How do I know my resume is ATS-friendly?

Cheap test: copy the text out of your finished file and paste it into a plain document. Comes out in the right order? Software can probably read it too. Jumbled paste, jumbled chances.

Should I cram in every keyword from the advert?

No. The true ones, in the advert’s wording, where they fit. A gap you can discuss honestly beats a keyword you’d have to defend at gunpoint — see the Excel incident, exhibit A.

Will employers know I used AI?

They’ll suspect it if the language is inflated or it doesn’t sound like you in the room. The real risk is claiming more than you can explain — interviewers are very good at finding exactly that line.

Last thing

AI is an editor, not an author of your life — the one sentence Bilal wrote down all evening. Feed it real achievements, aim it at one job at a time, turn duties into results, keep the formatting dull and the facts bulletproof. And if a cousin offers to help with yours, let him. Just make your own chai. His will be cold.

Leave a Comment