The same AI wave that floods employers with generated applications, per AI-written applications, flows the other way too: a growing share of job ads are machine-drafted, and most are indistinguishable from each other - fluent, structured and empty. The tool is not the problem; the division of labour is. AI should draft sentences from your facts, not invent facts inside its sentences.
- AI is excellent at job-ad structure and grammar and terrible at the only things that convert: your pay, your shifts, your floor's reality.
- The generic-slop problem is real - candidates now recognise and skip interchangeable AI ads, and platforms rank duplicated copy down.
- The working method: you supply the seven facts, AI supplies the drafting - never the reverse.
- Every AI draft gets edited against the seven-block standard and the banned-word list before it ships.
What AI actually does well here
Four things, genuinely: structure (given the blocks, it never forgets one), register (plain, warm, jargon-free on request - and it will strip corporate padding if told), speed at variants (feed one fact-sheet, get the board version, the 30-second video script per video job ads, and the two-line feed teaser in one pass), and language (competent native-style drafts across markets - reviewed by a speaker before shipping, per multilingual recruitment). All of it is drafting. None of it is knowing.
The prompt that works
Give the model the seven blocks from the job ad template as raw facts, then constrain it: "Write a 150-word job ad from these facts only. Add no requirements, benefits or claims not listed. No corporate phrases ('dynamic', 'fast-paced', 'team player', 'passionate'). Plain warm English, short sentences, the pay and schedule in the first three lines." The facts-only clause and the banned-word list do the heavy lifting - without them, every model pads and inflates by default. Then iterate on the draft with edits, not regenerations: 'shorter', 'move the schedule up', 'make week one concrete' - you are the editor; it is the typist.
The editing pass no draft skips
- Fact audit: every number, requirement and promise traced to your fact-sheet - anything the model added gets cut or verified.
- Slop scan: banned words, mission-statement prose and adjective padding out; the concrete in - the test is whether a competitor could run the same ad unchanged. If yes, it says nothing.
- Inclusion check: coded language screened out, per inclusive job ads - models reproduce the biases of the ads they trained on.
- The seven-block test: hook, reality, money, schedule, requirements, week one, apply step - present, ordered, honest.
The deeper point: the ad is downstream of the role
AI cannot fix an undefined role - it can only upholster it. If the pay is undecided, the shifts unclear or the must-haves a wish list, the model will generate fluent camouflage for the confusion, and the ad will fail exactly as an artisanal vague ad fails, per why nobody applies. The hour that matters still happens before any drafting: the one-page role definition from define the role. With it, AI drafting is a genuine accelerant; without it, faster slop.
The takeaway
Use AI as the typist it is: your facts in, constrained drafting out, human editing against the template before anything ships. The ads that win the feed in 2026 are not human-written or machine-written - they are fact-written, and only you have the facts.
Ads written from your facts.
Qwiza drafts the ad and the screening quiz from your actual role, pay and shifts - natively in your market's language, structure guaranteed. 48-hour pilot target.
See how Qwiza worksFrequently asked questions
Should I use AI to write my job ads at all?
Yes - for the part it is good at. AI drafts structure, tone and grammar in seconds, translates capably, and never forgets a section. What it cannot know is everything that makes an ad convert: your pay, your actual shift pattern, what week one looks like, why your last person stayed six years. Supply those facts and AI is a fine drafting assistant; ask it to invent them and you get the interchangeable slop candidates have learned to scroll past.
Why do AI-written job ads all sound the same?
Because they are sampled from the same distribution of existing job ads - which were already clichéd before the models trained on them. Left unprompted, every model produces 'dynamic team player' prose with suspicious enthusiasm. The differentiation was never going to come from the generator; it comes from the specifics only you hold. An ad that says '06:00 start, rota out 14 days ahead, 1,450 EUR net, Marko trains you' cannot sound like anyone else's - whoever drafted the sentences.
What should I never let AI do in a job ad?
Invent facts: pay ranges it guessed, benefits you do not offer, culture claims it cannot know, requirements it padded in. Model-invented requirements are especially toxic - they quietly add degree and experience filters that delete good applicants, per the requirement-inflation trap. The rule: AI touches wording, never facts; every number, requirement and promise in the final ad traces to something you typed or approved.
