Why do AI-written job descriptions attract the wrong candidates?
AI-written job descriptions attract the wrong candidates when they are generated from a job title, because the tool lists every requirement it has seen for that title, inflates seniority and repeats wording that can discourage some groups from applying.
A fictional group of three physiotherapy clinics, Riverside Physio, with 15 staff, needs a clinic coordinator. The practice manager, Helen, writes: 'need someone for front desk + booking + chasing insurers. Mon-Fri, one Saturday a month. must be good on phones. our system is Cliniko-ish. current person left after 6 months, overwhelmed by insurer paperwork. salary 26-28k. would be nice: medical admin background.' Her first instinct is to ask an AI tool for 'a job description for a clinic coordinator'.
How do you write a job description with AI step by step?
Describe the real job first, separate essential from nice-to-have requirements, let the AI draft in a fixed structure, then run a separate bias and clarity pass and send legal points to HR.
1. Describe the real work
List the main tasks with rough percentages of the week, the tools used, the hours and the hardest part of the job.
2. Split requirements
Essential means the person cannot do the job on day one without it. Everything else is desirable or can be trained.
3. Draft in a fixed structure
About the role, what you will do, what you need, what would help, what we offer, how to apply. Include salary if your organisation publishes it.
4. Run a bias and clarity pass
Ask the AI to flag gendered, age-related, ableist or culturally loaded wording, jargon, and requirements that do not match the tasks.
5. Check legal points with HR
Right to work, working hours, any health or background checks and equal opportunities wording vary by country. HR confirms them; the AI does not.
What does a weak and an improved AI job description look like?
Weak result from 'write a job description for a clinic coordinator': 'We are looking for a dynamic, energetic young professional to join our fast-paced team. You will be a rockstar multitasker with 3 to 5 years of healthcare administration experience and a degree in a related field.' 'Young' and 'energetic' can signal age preference, 'rockstar' is vague, and neither the degree nor five years of experience is needed for this role.
Improved, from Helen's real description and the split requirements: 'You will be the first person patients speak to at our three clinics. About half your week is booking and rescheduling appointments by phone and online, a third is preparing and chasing insurer paperwork, and the rest is front desk work. You will need: confident, clear phone manner; accurate data entry; comfort learning a clinic booking system. It would help if you have worked in medical or insurance administration, but we will train you on insurer processes.' The bias pass flags 'good on phones' as clear enough, and suggests stating the Saturday rota and salary range plainly. Helen sends the right-to-work and equal opportunities wording to the HR adviser before posting.
How do you check an AI job description before posting?
Check that every task matches the real job, that every essential requirement is truly essential, that the wording is inclusive and plain, and that HR has confirmed the legal points. A useful test: would the person who did this job well recognise it from the description?
Accuracy
Ask the current team or the previous postholder to confirm the tasks and the hardest part.
Requirement inflation
Remove degrees, years of experience and skills that the tasks do not need.
Inclusive wording
Remove age-coded, gendered or ableist words and unnecessary physical requirements.
Legal and policy
HR confirms right to work, hours, checks, pay transparency and equal opportunities statements for your country.
What are the common mistakes with AI job descriptions?
The commonest mistake is generating from a job title, which produces inflated requirements and the same text as every other advert. Others: leaving in phrases like 'digital native' or 'young team', listing a long wish list as essential, omitting the real hours, assuming the AI knows employment law in your country, and pasting an existing employee's personal details into the prompt.
Where does this lesson fit in a Tosheo.ai path?
Job descriptions sit in the Tosheo.ai HR and recruiting path, with related lessons on screening, interview questions and fair use of AI in hiring. The hands-on task asks the learner to draft and debias a description from rough notes, and the rubric feedback points out inflated requirements and exclusionary wording.