AI tools, like ChatGPT, Copilot, and Gemini are impressive. They’re fast, fluent, and seem to know everything. But when it comes to job ads – they don’t know enough.
They generate polished text – but not precise, purpose-driven content. We’ve tested them across real hiring cases, and the results are often too generic, missing the mark on tone, context, or compliance. In most cases, they even reinforce bias.
Here’s what we’ve learned.
AI is only as good as what it is trained on
Recruitment data on the internet is full of outdated, biased, or badly written job ads – and that’s what large language models train on. So even when they sound confident, they’re often replicating the very problems we’re trying to fix.
They don’t know your company, your teams, or your hiring standards. They don’t know what makes a great candidate for this specific role, right now.
As a result, they often miss the core content a candidate needs to make an informed decision about applying:
– What success looks like in the role
– What skills matter most to the team
– What tone reflects your brand
– What’s legally or inclusively required in your market
A study by our sister company, Develop Diverse found that leading AI tools often generated more biased language than humans. That’s a real concern – especially for companies committed to inclusive hiring or simply looking to attract the most qualified candidates from diverse backgrounds.
Can better prompts solve it?
Prompt engineering is evolving fast. Many teams are getting better at asking the right things – and some are even building their own AI tools. But here’s the catch: most of these tools still run on the same models, with the same blind spots.
Even with sophisticated prompt structures (we use them too), you still need:
– Insights from the hiring manager
– The right voice for your company
– Alignment on expectations
– Human judgment on what truly resonates
As our data lead Linea puts it:
“We almost never meet someone who loves writing job ads. Interviews and matching are just way more fun for most TA’s. And it makes sense; the job ad is not as easy to write as many think. The job ad is a way to express the team’s needs. Digging for real needs – not just what we think we need – is actually a demanding task, that takes detailed context awareness which an LLM of course don’t have for your exact role, team of company. Only TAs and hiring managers can bring that to the job ad.
Candidates will also notice your effort. Every product manager knows the typical responsibilities for that role, but they also know that it can differ quite a lot from company to company. If you fill the job ad with a generic job description, you risk the right candidates not seeing themselves in a poorly described role.”
Job ads aren’t just text
They’re conversations. Between your company and someone considering a big decision: to join you.
That requires more than clean grammar or catchy phrases. It takes real understanding – of the role, the context, and the people.
It starts with structured intake/kick-off: Hiring manager insights, aligned expectations, a clear sense of what success looks like. Then comes tone, compliance, inclusion – and finally, engagement.
Marketing teams have learned this lesson. Many still use AI, but they’re more intentional about when and how. Talent Acquisition teams are moving in the same direction – especially as scrutiny grows around AI-based hiring tools (including recent lawsuits targeting bias in screening algorithms).
AI can support the recruitment process. But it can’t replace it.
Because at the end of the day, hiring is human.
