AI for your CV and job search: passing screening without sounding like a robot

In short: AI helps in a job search wherever you need a draft and structure: tailoring a CV to a specific role, drafting a cover letter, rehearsing an interview. But the facts, the numbers and the voice must be yours — AI text left "as is" reads faceless and loses. The working scheme: your facts + a prompt with the job ad + your final edit. Below are four prompts you can run right here, with a button, and see the model's answer without leaving the page.

Where AI genuinely helps and where it hurts

A job hunt is a conveyor of repeating tasks: rewrite the same CV for five different roles, add a cover letter to each, then prepare for the conversation. That's exactly where AI saves hours. But there's a boundary: AI doesn't know your achievements. It doesn't remember that you cut an approval cycle from two weeks to three days. If you don't supply those facts, the model will invent them or replace them with generic filler like "solved tasks effectively".

• Helps: structure, wording, tailoring to a role, translation, tone, question rehearsal, decoding the job ad.
• Hurts: when you ask "write me a CV" with no inputs and paste the result unread. You get text the recruiter has already seen forty times today.

Think before reading on: name three of your results from the last year with a number or a deadline attached. If that was hard — that's the real work, and AI won't do it for you. It only packages it.

We studied the courses that promise to teach this. Here's what's actually inside

Before handing out advice, we went and looked at what the people selling "AI for work" as a course actually teach. On 17.07.2026 we pulled data from Udemy's internal API and from Coursera's review pages. These are our own measurements — the figures below aren't in any public write-up.

The largest course on exactly our topic is ChatGPT: Complete Course For Work by Steve Ballinger: 281,823 students, 130,274 reviews, a 4.46 rating, 16.5 hours of video across 158 lectures, last updated 01.07.2026. Its prerequisites, word for word: "No prerequisites as ChatGPT is a tool anyone can access and use immediately". So: sixteen and a half hours of video for a person who needs to rewrite one CV by Friday.

Here's what the unhappy ones write:

"The majority of the class he just rambles around unimportant subject… There's really no 'meat' in this course. A major waste of time and money!" — Kevin K., 20.02.2026, 1★ (ChatGPT: Complete Course For Work)

That isn't a lone voice. By our count, the share of ratings at or below 3.5★ is 9.61% for Generative AI for Beginners (409,492 students) and 10.36% for The Complete AI Guide (376,845 students, 42 hours, 545 lectures). Roughly one student in ten is unhappy — behind an average rating that looks perfectly lovely.

And here's the interesting part: what people complain about on courses with "prompt engineering" right there in the title.

"There is nothing teached about creating a good prompt. It is just an overview of types of prompts." — Geralt O., 01.07.2026, 2★ (Mike Wheeler, Prompt and Context Engineering 101)

"No specific guidance on prompt engineering… what to avoid while asking, how to organize your thoughts, how to give feedback to AI based on its answers etc." — Bharat Ram A., 05.06.2026, 1.5★ (same course)

"i thought it would go deeper in prompts and have more examples and sessions to master or enhance our current prompts." — Manuel L., 15.06.2026, 2★ (Prompt Engineering for Everyone)

Read what they're asking for: what to avoid asking, how to organise your thoughts, how to give feedback to the model on its own answer. None of that is "the prompt formula". All of it is debugging: the prompt didn't work — what do I change first? That gap is exactly what stands between you and a decent CV, which is why there's a whole section on the debugging ladder below. It matters more than every template on this page combined.

"This is not my voice" — the complaint that lands directly on you

While collecting user pain points, one cluster stood out that not a single course we looked at addresses:

"I've studied best practices to avoid it generating content that sounds like AI, but I'm not having any success. … It is reported as 100% written by AI according to zerogpt." — from the thread "Prompt to insert content without sounding like AI" on the OpenAI forum

"the email writing tools just seems to strip out my personal voice making me sound like I'm writing unsolicited marketing spam." — a comment on Hacker News

A caveat against ourselves: that's three sources, not a sample. People on the OpenAI forum and Hacker News are advanced users, not a typical job seeker. We can't tell you "X% of people hit this" — we don't have that data and we're not going to invent it. But the wording of the second complaint is worth memorising: the tool strips out the personal voice. Not "fails to add" — strips out. The same thing happens to a CV: you hand the model a live fact — "talked three clients out of leaving after a failed release" — and get back "maintained high levels of client retention". Formally identical. In practice you just became one of the forty.

Step 1. Gather the raw facts before opening a chat

Fifteen minutes on paper saves an hour of back-and-forth with the model. For each past role write down: what it was like before you, what you did, what it was like after. Use any verifiable numbers — counts, deadlines, money, volume, headcount. Even "handled 12 clients instead of 7" beats "actively grew the client base". That's the raw material; AI then helps package it.

A working rule: if a fact couldn't survive a phone call to your former manager, it isn't a fact — it's an adjective. Adjectives don't work on a CV because everyone writes them. A checkable claim works precisely because it's frightening to write, and that shows.

Step 2. Tailoring the CV to a specific role

A universal CV almost always loses to a tailored one. Many companies use applicant tracking systems (ATS) that look for overlap between your text and the job description. On top of that, a human recruiter skims on the first pass. The goal: the words that matter for this role should appear in your CV honestly, not by guesswork.

The prompt below is filled in completely — it already contains an invented CV and an invented job ad. Hit "Run" and watch the model take the pair apart. Then swap in your own.

You are an experienced technical recruiter. Compare this candidate's CV against the job ad.

CV:
Maria K., 4 years in customer support for a SaaS product.
- Handled chat and email tickets, around 60 per day.
- Wrote 40 knowledge-base articles; tickets on those topics fell from 22% to 9%.
- Trained 5 new hires; ramp-up time dropped from 6 weeks to 4.
- Replaced a manual repeat-contact report with a spreadsheet, saving 3 hours a week.
- English B2, native Spanish.

JOB AD: Support Team Lead
Requirements: 3+ years in support; experience managing a team of 5+;
working with CSAT and first-response-time metrics; building processes
and documentation; English C1; automation experience a plus.

Give exactly three blocks:
1) What Maria genuinely matches, and how to rephrase each point in the
   language of the ad WITHOUT adding facts that aren't in the CV.
2) Which requirements she does not meet — honestly, no softening.
   For each: is it a hard blocker, or can a cover letter work around it?
3) What to cut as irrelevant for this specific role.

Finish with one line: apply, or don't waste the time. Justify it.

Note the line "WITHOUT adding facts" — without it, the model will happily paint in experience you don't have. That's not paranoia: AI hallucinations in a CV turn into a lie at the interview, and one follow-up question exposes it. And note block 2: a model that hasn't been asked for honesty will default to making you feel good.

Step 3. A cover letter that doesn't get binned

A cover letter works if it answers one question: why you and why here. The model is great at removing the blank page, but you supply the specifics: what caught your eye about the company, which part of your experience solves their pain.

A banned-phrase list is an underrated trick. It knocks the model out of its averaged template — it simply has no option left except finding more precise words. Ask for two variants while you're at it: comparing is easier than judging a lone text.

Write a cover letter on behalf of this candidate.

Candidate: Maria K., 4 years in SaaS support.
Facts you may use (there are no others; inventing is forbidden):
- 40 knowledge-base articles, tickets on those topics fell from 22% to 9%;
- trained 5 new hires, ramp-up dropped from 6 weeks to 4;
- native Spanish, English B2.
Role: Support Team Lead at a company building a doctor-booking service
that is expanding into the Spanish market.
Why she cares: she has spent three years explaining complicated things
in plain words, and here that's needed in a new language and a new market.

Format: no longer than 150 words, four short paragraphs, a living tone.
Banned: "I am a team player", "dynamically growing", "results-oriented",
"I was interested in your vacancy", "extensive experience".
First paragraph — straight to the point, no run-up.
Name the gap outright (English B2 against a C1 requirement) and handle it honestly.

Produce two different versions: one restrained, one bold.
Under each, one line on which kind of hiring manager it lands with.

For how a strong request is built in general, see what a prompt is and how to write one.

Step 4. The emptiness test

This is the fastest way to understand why your text isn't working. Take a finished letter — yours or the machine's — and ask the model to strike out everything that would fit any candidate. The prompt below already contains a specimen of classic AI filler. Run it and see what survives.

Below is a cover letter. Act as a merciless editor.

LETTER:
"Hello! I was interested in your project manager vacancy.
I am a goal-oriented professional with extensive experience in a
dynamically growing industry. I am a team player, I solve tasks
effectively and thrive in a multitasking environment. My strengths
are communication skills, responsibility and a drive for development.
I am confident that my experience and skills will be useful to your
company. I would be glad to discuss possible cooperation with you."

Do three things:
1) Put every sentence in a table, and next to it: would this fit any
   other candidate on earth (yes/no)? Assume "yes" by default.
2) Calculate what percentage of the text survives once all the "yes"
   rows are removed.
3) Ask me 5 questions whose answers would turn this into a letter that
   only one person alive could have written.
Do not rewrite the letter. I want the questions, not your version.

The last line is the important one. Ask a model to "rewrite this nicely" and it returns the same filler in different words. Asking for questions turns it around: now it's extracting facts out of you instead of manufacturing text.

Step 5. Interview rehearsal

This is probably the most underused application. The model can play the interviewer and ask questions — including the uncomfortable ones. The prompt below is self-contained: hit "Run" and answer right there in the chat.

You are the hiring manager for a Support Team Lead role.
Interview me under these rules:
- ask ONE question, then stop and wait for my answer;
- once I answer, briefly (2-3 lines) say what was strong, what was weak,
  and which fact was missing;
- then ask the next question, building on my previous answer.
Eight questions total. Make sure you reach the uncomfortable ones:
a gap in my history, a career switch, why I left, a conflict with a manager.
Don't let a vague answer slide — push back and ask for specifics.
Do not show all the questions as a list. Ask the first one and wait.

Five run-throughs and you won't be hearing any question for the first time in the real interview. Separately, ask it to decode the job ad: "what is actually behind these requirements, what pain do they solve, what will they ask about".

This is where our research is worth revisiting. We looked at how paid courses check that a skill was learned, and found this:

"you send your assignments and immediatly you got your results: 100% correct. I am still speechless. I put all that effort in and have no idea whether I my answer was correct or not." — Shinysheep, 21.09.2023, 1★ (Prompt Engineering for ChatGPT, Vanderbilt, 698,444 enrolled)

That isn't an author cutting corners. It's a structural limit: there is no model inside a Coursera lesson, so there is physically nothing to check your prompt with — which leaves an automatic "100% correct". Hence a simple takeaway worth carrying away: any prompt training that never runs a prompt is reading about swimming. The button under every block on this page exists for exactly that reason.

The prompt didn't work. What to change first

This is the section whose absence those four reviews above are mourning. Walk the ladder top-down and change one thing at a time — otherwise you'll never know what helped.

• 1. You're missing facts, not phrasing. The most common cause of filler. If there isn't a single number in your prompt, the model physically cannot write concretely. Fix: add three checkable facts. Test: did the answer gain nouns and lose adjectives?
• 2. There's no prohibition. The model averages because average is safe. Fix: a list of five banned phrases plus an explicit "don't invent facts". Test: did the boilerplate turns of phrase disappear?
• 3. There's no format constraint. "Write a letter" gives you a page. "150 words, four paragraphs" forces word choice. Fix: a hard limit. Test: length.
• 4. There's no role and no reader. "You are a hiring manager who will read 200 letters tonight" hands the model a rejection criterion. Fix: one line of role. Test: did it start prioritising?
• 5. You're asking for text when you should ask for questions or criticism. The most underrated move on this list. Change the verb: not "write", but "interrogate me", "tear this apart", "find the holes". Test: did you learn something about yourself?
• 6. The task is fused. "Rewrite my CV for this ad and do the cover letter too" is two tasks; the model will do both mediocrely. Fix: split into two requests.
• 7. Only now — the model. Switching tools is the last thing to try, once the first six are handled. Usually by then you no longer need to.

If the answer is still faceless after step 5, the problem almost certainly isn't the prompt — it's that you don't yet have a formulated fact. That's an unpleasant but useful diagnosis: go back to step 1 and go get one.

Eight mistakes that make an AI CV obvious

• One CV for every role. Ten minutes saved at the cost of every application.
• Borrowed experience. The model added "team management", you didn't notice, and it surfaced at the interview. More expensive than any gap.
• Symmetry. Four paragraphs of identical length, three bullets in each. Live humans don't write like that.
• Adjectives instead of numbers. "Significantly improved" carries zero information.
• Words from the ad with nothing behind them. The ATS lets them through; a person asks about them, and you sink.
• A letter that retells the CV. If the cover letter duplicates the CV, it isn't needed.
• Editing with synonyms. Swapping "effectively" for "efficiently" adds not one gram of specificity. Detectors have nothing to do with it — you're just repainting emptiness.
• Zero rough edges. Perfect smoothness is the main tell of a machine.

Will the recruiter know it was AI?

The honest answer: dedicated "AI detectors" are unreliable and regularly flag human writing as machine-made — neither you nor a recruiter should lean on them. But an experienced person often spots the style without any detector: smooth, symmetrical, full of generic claims and free of any detail only you could know. What gets caught isn't "AI", it's emptiness. So the best defence isn't swapping in synonyms — it's specifics: project names, numbers, circumstances, your own phrasing.

• Read the text aloud. If you don't talk like that, rewrite it.
• Delete any sentence that would fit any candidate at all.
• Keep one or two rough edges of your own voice.

The technical side of this — why detectors get it wrong and what they actually measure — is covered separately in our piece on AI text detectors.

What never to hand to AI

A CV is personal data: yours, and sometimes other people's. Don't paste former colleagues' and clients' details, internal company documents, or commercial specifics under NDA. Many services may by default use your input to improve their models — this can usually be switched off in settings, and under GDPR you have rights over how your personal data is processed. A simple habit: replace your name, phone, address and links with "Candidate", "city", "X" before pasting. It changes the quality of the analysis not at all, and drops the volume of leaked data to zero. More in our article on privacy when working with AI.

Is a course worth buying for this?

Since we pulled the prices anyway, here they are as of 17.07.2026. Eleven of the twelve top Udemy courses cost €19.99; one (The Complete AI Guide) is €24.99. And here's the small discovery that made digging into the API worthwhile: there is no discount. The server response returns "saving_price": {"amount": 0.0}, "has_discount_saving": false, "discount_percent": 0. Price and "list price" are the same number: what a storefront dresses up as a vanishing offer is, on the server, simply the price.

Subscriptions: Udemy Personal Plan is €20.00/month (promo €10.00/month). Coursera Plus is €50/month or €343/year with a 14-day refund window. Google's own programmes on Coursera run $49/month after a 7-day trial. One caveat, since we're checking everything: we took the Udemy Personal Plan price off their landing page, not the API — udemy.com returns 403 to a crawler. That's one notch weaker than the discount JSON, and we'd rather say so than pass the number off as a measurement.

And the cherry: Vanderbilt's Prompt Engineering Specialization (138,967 enrolled, 39 hours) requires a paid ChatGPT+ subscription to complete the assignments. So a Coursera student pays €50/month for the platform plus separately for the model — all in order to watch videos and receive an automatic "100% correct".

A caveat against ourselves: a 10% unhappy share doesn't make a course bad — 90% are fine, and Ballinger's 281,823 students found something in it. We didn't measure whether anyone learned anything; we read what they complain about. Those are different things. So the conclusion is a careful one: if you need to rewrite a CV by Friday, 16.5 hours of video is the wrong tool. If you want to understand the field for the months ahead, a course may well fit — just check first whether it contains a single place where you run your own prompt and get feedback from something other than an answer key.

Do it now (20 minutes)

Take one role you genuinely want and run the cycle: facts on paper → the prompt comparing CV to ad → the emptiness test → edits in your own voice → a 150-word cover letter → one interview rehearsal. One pass like that beats ten applications fired off at random. All four prompts above run from a button right here — start with the emptiness test. It takes a minute and it usually sobers people up.

To widen your use of AI beyond the job hunt, start with our article on AI for work. If the durability of your profession is what worries you, see the breakdown of whether AI will replace your job. If the role itself interests you, there's a piece on the prompt engineer profession. And if you want a stock of ready-made starting points, look through our prompt examples.