Does AI write better Job Descriptions than humans?

Human vs. Robot - Who writes the best job description?

Is it possible to use an early generation of Generative AI (GenAI) to write a compliant Job Description (JD) for a senior position at the UN?

That is what we test in this article. I have worked with a senior (anonymized) recruiter at the UN, and I wrote a brief instruction to our generative AI software to draft a job description for a position at the D2 level.

I also choose a JD for a real job that has recently been advertised.

We then sent the two versions of the JD to a UN Job classification expert. We didn't tell them that one JD was drafted by a human and one JD by AI. We asked them to check compliance and let us know which one they preferred.

To draw accurate conclusions, the topic needs more research. Still, with this first test, we have learned that a JD drafted by AI is already at an acceptable standard and possibly even better than a JD prepared by an HR professional.

The robot gets its instruction.

I am a recruiter, not a UN Job Classification expert, but I believe (after about ten years at the UN) that it is fair to say that I have a good understanding of the UN recruitment process and the needs of the UN.

Hence, I assess that the quality of my instructions would be comparable with what other UN recruiters would deliver. To get started, I pasted the below instructions into the message prompt of the selective generative AI tool:


"Write a job description for the position of the ILO Director of the Human Resources Development Department. The job level is D2, the contract type is FTA, and the duty station is Geneva. The tone and voice in the job description must speak to women candidates.

The job description should include the following section headers:

  1. Background about ILO

  2. Introduction

  3. Specific duties

  4. Desirable profile

  5. Competencies

  6. Education

  7. Languages

  8. Experience

  9. Conditions of employment

  10. Recruitment process

  11. Fraud warning

The experience section must include the minimum requirements for a D2 at the UN and a statement that ILO is an employer that encourages applicants with diverse backgrounds.

The text about the conditions of employment must comply with ILO’s conditions of employment.

The section about the recruitment process includes information about the desire of candidates from under-represented or non-represented member states.

The job description should be three pages long and use a language that is easy to understand."


Job description blind test review

With the GenAI-generated JD, I now had two job descriptions: the original ILO, one JD used during a 'real' recruitment process, and one 'fake' drafted by a robot.

I contacted a former UN colleague with profound experience in UN job classification. Hereinafter, I will call my UN colleague, the expert. To ease the project execution, the expert asked to be anonymous as the internal approval process to take part in this type of project would otherwise take too long.

I purposely chose a job description of another UN agency than the one the expert worked for. In my message to the expert, I only mentioned that Impactpool runs a research project where we look at new types of job descriptions.

I named the ILO Job Description JD A (hereinafter referred to as the ILO JD) and the Job Description written by GenAI as JD B (hereinafter referred to as the GenAI JD).

I asked if the expert could review two versions and let me know if the JDs would comply with UN rules.

Judgment Day

The expert sends the response in an e-mail. The expert writes in the reply.

"They both work. JD A (our remark: the ILO JD), I would say is a more traditional approach that some UN organizations would prefer or still prefer to use for their job posting/VN.

I would opt for JD B (our remark: GenAI JD), where the VN is written in a more catchy way, adaptable to the current talent outreach efforts and market."

Although the expert liked the more catchy way of writing, the GenAI JD is a bit too generic and needs some more details. The expert considers the GenAI JD more a vacancy notice (VN) than a "standalone" JD.

"Maybe here I would add a link to the detailed JD for more in depths details to the candidates. The JD B (our remark: GenAI JD) also provides an intro to the organization which I think is important, especially for a candidate that might have not worked in the UN environment, hence before doing an in-depth search about the org at least has a few lines on the mandate and also what the org is interested to attract.

Maybe a combination of the two could work out, thinking out loud to sell it to some UN entities for their job posting or a detailed JD can be attached to the “catch” VN opening."

Quality of the JDs

In the evaluation, the expert was also asked to rate the quality of the two JDs. The expert rated the JD quality on a 1-5 scale:

  • #1, poor, the JD is not possible to use at all

  • #3, ok, the JD is sufficient for UN posting, and 

  • #5, the JD is excellent as is

The expert gave the ILO JD a #3, with the comment:

"The JD A (our remark: the ILO JD) would be sufficient for a UN job posting, but requires additional updates in my opinion."

But interestingly, the expert gave the GenAI JD a #5, with the comment:

Excellent, but it needs an additional link to the detailed JD, but only minor changes are needed."