I've lost count of the job adverts I've rewritten over the years that started life as a copy of an old spec from 2019, got a new salary band stapled on, and went out the door hoping for the best. Most of them attracted plenty of applicants, just not the right ones. That's the problem AI can genuinely help with, and it's also the problem AI can quietly make worse if you use it lazily.

Here's the short version before we go deep. Using AI to write job ads works when you treat the tool as a fast first drafter and a ruthless editor, feed it real information about the role, and keep a human accountable for every word that goes live. It fails when you type "write me a job advert for a marketing manager" and publish whatever comes back. The difference between those two approaches is the difference between a shortlist you're excited about and a pile of one hundred and forty near identical applications you have to wade through.

This guide is written for UK employers, in house recruiters, and agency folk. The law here is different from the American rules most AI tools were trained on, the salary transparency picture is different, and even the vocabulary is different. I'll flag all of it as we go.

Why Your Current Job Adverts Are Underperforming

Before you blame the job boards, look at the advert itself. Candidates make the decision to apply or move on astonishingly quickly. A survey by The Ladders, cited in Datapeople's analysis of long job posts, found jobseekers spend an average of just 49 seconds reading a job ad before deciding whether to apply. That's roughly 100 to 150 words of actual reading. Meanwhile the average advert runs past 600 words, and around 30 percent exceed 1,000.

Length isn't a neutral choice either. Datapeople's own data shows that once a post passes 600 words, the proportion of one click applications rises, and those applicants tend to be less qualified because they've applied without properly reading the role. So a bloated advert doesn't just bore good candidates, it actively fills your pipeline with weaker ones.

LinkedIn's behavioural research points the same way. In LinkedIn's analysis of job post statistics, short posts of 1 to 300 words persuaded candidates to apply 8.4 percent more often than average, while medium length posts of 301 to 600 words performed 3.4 percent below average. More than half of job views on LinkedIn happen on mobile, where a wall of text is even less forgiving. One caveat: LinkedIn's dataset is global and skews American, but I've seen nothing in UK hiring data that contradicts the basic pattern. People everywhere skim.

Then there's the pay question, where the UK evidence is unambiguous and entirely home grown. Research from Reed found that 78 percent of jobseekers are less likely to apply if the advert doesn't show a salary, and 22 percent exclusively apply for roles that do, according to People Management's coverage of the Reed salary transparency research. Reed's own platform data showed adverts displaying salary attracted 27 percent more applications, and their later 2024 analysis put the uplift at around 60 percent. A quarter of candidates say the word "competitive" in the salary field actively puts them off. On this one, the UK is actually ahead of its neighbours: Indeed's data shows 71 percent of UK adverts now list pay, compared with 50 percent in France and just 16 percent in Germany.

So the diagnosis is usually some mix of three faults: too long, too vague, and coy about money. AI can fix the first two in minutes. The third is a business decision no model can make for you.

What AI Actually Does Well Here

Let me be specific, because "AI writes job ads" covers several quite different jobs.

First, drafting. Given a decent brief, a general purpose model like ChatGPT or Claude will produce a structurally sound first draft in under a minute, saving the hour or two a recruiter might otherwise spend. That's real time back for the work that actually fills roles, which is talking to people.

Second, cutting. This is the underrated one. Paste in your existing 900 word advert and ask the model to get it under 350 words while keeping the salary, the four genuinely essential requirements, and the honest description of the day to day. Models are far better editors than most humans, because they have no emotional attachment to the paragraph about your founding story.

Third, language checking. AI tools can scan an advert for gendered or exclusionary wording. The research behind this goes back years. Textio, the best known platform in the space, analysed millions of job posts and found that in jobs where a man was ultimately hired, the original advert contained nearly twice as many masculine toned phrases as feminine ones, with the reverse true where a woman was hired. Words like "fearless", "exhaustive" and "enforcement" skew male in application data; "transparent" and "in touch with" skew female. Textio's data is largely US based, but the underlying psychology travels, and there's a striking UK example: Accenture's UK and Ireland recruitment team credited de biased job descriptions as part of why female applicants rose from 34 percent to 50 percent since 2014, as reported in Stem Women's piece on inclusive job adverts.

Fourth, variant testing. Ask for three versions with different opening lines, run them against each other on your careers page or across two job boards, and keep what wins. Almost nobody does this with job adverts, which is odd given that every marketing team on earth does it with everything else.

What AI Does Badly, and Why That Matters More Than Vendors Admit

Now the honest bit. A raw AI draft, unedited, has three predictable failures.

It invents things. Ask for a job advert with a thin brief and the model will confidently generate responsibilities, benefits and even salary figures that don't exist at your company. I've seen AI drafted adverts promising "flexible hybrid working" for a role that was five days on site. That's not just embarrassing, it's the sort of misleading claim that costs you a new hire in week three when reality bites.

It sounds like everyone else. Models are trained on the internet's existing job adverts, which means they reproduce the internet's existing clichés: fast paced environment, passionate self starter, wearing many hats. Candidates have read a thousand of these. When every advert reads the same, the only differentiators left are salary and brand, and if you're not winning on either, you've just made yourself invisible. This matters more now than it did two years ago, because candidates are drowning in sameness from the other direction too. CV Library's March 2026 research, covering 424 recruiters and over 1,000 UK candidates, found 40 percent of jobseekers abandoning applications because of AI heavy processes, and more than a third of recruiters admitting AI tools are causing them to miss strong candidates. An advert that reads as machine written signals a process that will treat applicants as machine input. Good candidates notice.

It applies American defaults. This is the one UK employers underestimate. Models trained mostly on US content will happily reference at will employment, EEO statements, 401(k) plans, and "compensation" in dollars. None of that belongs in a British advert, and some of it actively confuses UK candidates. Always tell the tool explicitly that this is a UK role governed by UK employment law, and check the output anyway.

If you take one section seriously, make it this one. The rules that govern AI assisted job adverts in Britain are British, and they have teeth.

The Equality Act 2010 makes it unlawful to discriminate in recruitment on the basis of protected characteristics, and that includes indirect discrimination baked into an advert's language. Crucially, you remain liable even when a third party tool produced the problem. As the employment law firm Davidson Morris puts it in their guide to AI recruitment risks, AI use in hiring now sits squarely within UK GDPR, the Data Protection Act 2018 and the Equality Act 2010, and the question for employers is whether their use of it can be justified, documented and defended if challenged.

The Equality and Human Rights Commission, the EHRC, which is the regulator for discrimination law in Britain, has warned specifically that generative AI can produce discriminatory job adverts, for example gendered titles or masculine coded language, because it reproduces bias from its training data. So the same technology that can de bias an advert can also introduce bias, depending entirely on how you use it.

Then there's the government's own guidance. The Department for Science, Innovation and Technology published Responsible AI in Recruitment in March 2024, and it's genuinely useful rather than box ticking. It flags that general text generation tools not specialised for job descriptions can produce outputs that aren't legally compliant, or that use vague or dissuasive language which discourages good applicants. It also recommends risk assessments for AI tools used in targeted job advertising, giving the example of a system that might inadvertently create barriers for applicants over 40 with lower digital engagement.

The Information Commissioner's Office, the ICO, is the data protection angle. If you're only using AI to draft advert copy, your data protection exposure is modest. The moment your tooling starts processing candidate data, targeting adverts at profiled audiences, or screening applications, you're into high risk processing territory and impact assessments stop being optional.

One American import to resist: you'll see US articles quoting fines from the EEOC or referencing New York City's automated hiring law. Interesting, irrelevant. Your regulators are the EHRC and the ICO, your statute is the Equality Act, and your framework is UK GDPR.

The Tools Worth Knowing About

I'm deliberately not going to give you a table of fifteen tools, because for writing job adverts the realistic shortlist is short.

General purpose AI assistants. ChatGPT Plus costs £20 a month in the UK. Claude's paid plan is priced in US dollars at 20 dollars a month, which works out at roughly £16, and since the platform bills in dollars that conversion is approximate. Either will handle drafting, cutting, tone adjustment and language checking if you prompt them properly, and for a company hiring a handful of times a year this is all you need. My honest preference for advert copy is Claude, which tends to produce less florid, more natural prose out of the box, but the gap is small and the prompt matters far more than the model.

Best for: SMEs and anyone hiring occasionally, where £20 a month covers every writing task in the business, not just recruitment.

Specialist job ad platforms. Datapeople, now owned by Payscale, is the serious enterprise option. It sits inside your applicant tracking system, checks adverts against inclusion and compliance rules in real time, and draws on what the company says is over 100 million job data points. Pricing is quote based rather than published, there's no free version, and it's built for organisations running hundreds of live vacancies with consistency problems across teams and regions. Reviewers on G2 and elsewhere praise the real time language feedback but note the scoring can feel arbitrary and the tool is designed for English, so multilingual hiring is awkward. Textio plays in the same space with a heavier emphasis on bias analytics; it's an American platform, its data reflects US hiring outcomes, and again you're into sales call pricing. For most UK employers below enterprise scale, I genuinely don't think these platforms justify their cost over a well prompted general assistant plus a human editor who knows the Equality Act.

Best for: large employers with distributed hiring teams, global brands worried about consistency, and anyone whose legal team wants an audit trail.

Free gender decoders. Before you pay anyone anything, know that free browser based gender decoder tools exist that check adverts against the academic list of masculine and feminine coded words. They're crude compared with Textio's statistical approach, but crude and free beats sophisticated and unbought.

What I'd skip: the AI writing add ons bundled into cheap job posting platforms. In my experience they're thin wrappers around the same underlying models you can access directly for less, with worse prompts than you can write yourself in ten minutes.

A Workflow That Actually Produces Better Applicants

Here's the process I use and recommend, start to finish. It takes about 45 minutes for a role the first time and 20 minutes once you've got templates.

Step one, interview the hiring manager before you touch any AI. Ten minutes, five questions. What will this person actually do in a normal week? What does success look like at six months? Which three requirements are genuinely essential, as opposed to nice to have? What's the salary band? What's the honest downside of this job? That last question is the secret weapon. An advert that admits "the first three months involve untangling a messy legacy system" filters out people who'd hate that and attracts the strange, wonderful people who enjoy it.

Step two, write a proper prompt. Not "write a job advert for an operations manager". Something like: "Write a UK job advert for an Operations Manager at a 40 person logistics firm in Leeds. Salary £42,000 to £48,000, hybrid with three days on site. Use British English and UK employment norms. Keep it under 350 words. Plain, warm, direct tone, no clichés, no buzzwords, no phrases like fast paced or self starter. Structure: two sentence hook about the actual work, what you'll do, what you'll need (only these four essentials: ...), what we offer, how to apply. Do not invent any benefits or details beyond what I've given you."

Step three, generate three variants and butcher them. Take the best skeleton, then edit like a human. Replace every generic claim with a specific one. "Great culture" becomes "the whole company finishes at 1pm on Fridays". Cut anything you couldn't defend in an interview when a candidate asks about it.

Step four, run the bias and compliance pass. Ask the AI directly: "Check this advert for gendered language, age coded phrases like digital native or recent graduate, unnecessary requirements that could indirectly discriminate under the Equality Act 2010, and anything that assumes the reader has no disability or caring responsibilities. Explain each flag." Requiring a driving licence for a desk job, demanding "excellent spoken English" where written communication is what matters, insisting on a degree for a role that doesn't need one: these are the quiet filters that shrink your applicant pool and create legal exposure at the same time.

Step five, check the mundane stuff the model gets wrong. Salary present and honest. Location and hybrid arrangement stated plainly. Closing date real. Job title searchable, because candidates search for "Financial Controller", not "Numbers Ninja", and a clever title is a discoverability tax you pay for a joke nobody asked for.

Step six, human sign off with a name attached. Someone accountable reads the final advert as a sceptical candidate would. The DSIT guidance and every serious UK legal commentator lands on the same principle: AI assists, a human answers for the output.

Step seven, measure and iterate. Applications per view, quality of shortlist, time to fill, and where your eventual hire actually came from. Feed what you learn into the next prompt. This is the compounding advantage: your prompts get better every quarter, in a way a copied old spec never does.

Titles, Search and the Boring Craft AI Can't Do For You

A quick word on the parts of the job that sit around the writing, because a beautifully worded advert nobody finds is a beautifully worded waste of an afternoon.

Job titles are search terms first and branding second. Candidates on Indeed, Reed, Totaljobs and LinkedIn type the ordinary name of the role they want. If your advert is titled "Growth Wizard" or "Customer Happiness Hero", it simply won't surface for the people searching "Marketing Executive" or "Customer Service Advisor". AI tools will occasionally suggest these cute titles because they've absorbed a decade of startup job boards; overrule them every time. There's an inclusion angle too. Stem Women's research on inclusive adverts notes that combative or laddish titles like ninja and guru can signal a hostile environment to candidates outside the majority demographic, which means your clever title is quietly filtering the wrong way.

The first two lines of the advert do disproportionate work, especially on mobile, where most candidates will see only your opening before deciding whether to tap. Lead with the actual work and the actual money, not "About Us". Nobody ever tapped "see more" because paragraph one described a company founded in 2011 with a passion for excellence. Your founding story can live on the careers page.

And keep a house style prompt. This is the single habit that separates teams getting compounding value from AI and teams getting a slightly faster version of their old mess. Write one master prompt containing your tone rules, your banned phrases, your standard benefits list, your salary transparency policy and a reminder that every advert is for the UK market under UK law. Save it somewhere shared. Every new vacancy starts from that prompt plus the hiring manager's answers, which means your fiftieth AI assisted advert is meaningfully better and more consistent than your first, and a new team member can produce on brand adverts on day one. The DSIT guidance calls this kind of thing governance; I call it not solving the same problem eleven times.

One last craft point: read the finished advert aloud. It sounds daft, and it's the fastest AI detector there is. Machine drafted sentences that survive silent reading fall apart when spoken, all those triplets of adjectives and sentences that begin identically. If you stumble reading it, a candidate stumbled too, about 40 seconds into their 49.

Should You Tell Candidates You Used AI

For advert copy alone, there's no UK rule requiring disclosure, and I don't think a line saying "this advert was drafted with AI assistance" adds anything for anyone. The disclosure obligations that do exist, under the ICO's guidance and the DSIT framework, bite when AI processes or assesses candidates: screening, ranking, chatbot interviews. There the direction of travel is very clear, and candidates are voting on it already. Greenhouse's 2026 research found 47 percent of UK jobseekers had been interviewed by AI, 82 percent weren't told beforehand, and 30 percent dropped out of a hiring process after discovering an AI led interview was involved. As one recruitment chief executive put it in that coverage, candidates aren't objecting to AI in principle, they're reacting to its invisibility. If your use of AI ever goes beyond drafting words, signpost it plainly.

The Reality Check

A few honest limits, because this article would be dishonest without them.

AI won't fix a bad offer. If the salary is under market, the commute is grim and the progression story is vague, the world's best written advert just delivers those disappointments more elegantly. Reed's research is blunt on this: candidates rank salary as the number one reason to apply, and no phrasing rescues a number you're embarrassed to print.

Adoption is ahead of skill. The CIPD's Resourcing and Talent Planning research found 31 percent of UK organisations using AI in recruitment, up from 16 percent in 2022, as reported by Onrec's coverage of the CIPD survey. Most of those organisations, in my experience, are using it the lazy way: generate, skim, publish. Which means the bar for standing out by using it well is currently, wonderfully low.

Nor will AI substitute for knowing your market. A model can't tell you that experienced warehouse supervisors in the North West won't move for less than £34,000 this year, or that your competitor down the road just added private healthcare. That intelligence comes from talking to candidates, reading rejection feedback and benchmarking honestly, and it has to flow into the brief before the AI ever sees it. Garbage in remains garbage out, however fluent the garbage now sounds.

And "more applicants" is the wrong goal anyway. UK graduate roles now routinely attract three figure application counts, much of it AI assisted volume from the candidate side. You don't want a bigger pile. You want a smaller, sharper pile, which is exactly what specificity, honest downsides and a visible salary produce. Some of my best performing adverts got fewer applications than the versions they replaced, and filled faster, because the people who applied were right.

What to Do This Week

Pick your single hardest to fill live vacancy. Take the current advert and run the workflow above on it: interview the hiring manager for ten minutes, rebuild the advert with a proper prompt, cut it under 350 words, put the real salary band on it, run the bias pass, and have a named human sign it off. Post the new version alongside or instead of the old one and compare a fortnight of results. That single experiment will teach you more about using AI to write job ads for your specific roles and market than any article, including this one. The tools are cheap, the evidence is clear, and the recruiters winning with this right now aren't the ones with the biggest budgets. They're the ones who kept thinking after the model stopped typing.