How to Write a Job Description for Better AI CV Screening

To get accurate AI CV rankings, write a job description that clearly states the must-have skills, the seniority you need, and any dealbreakers - in specific, concrete terms. AI scores candidates against what you write, so a vague description produces vague rankings and a sharp one produces a sharp shortlist.

Here's how to write a JD that screens well.

1. Lead with the must-haves

List the 5–8 skills or experiences that genuinely matter. Be specific: "3+ years building production React apps" ranks better than "frontend experience". The model can weigh concrete requirements; it can only guess at vague ones.

2. State the seniority plainly

Say whether you want a junior, mid, senior, or lead. This drives the seniority sub-score and stops the model from over-rewarding an over-qualified candidate or under-rating a strong mid-level one.

3. Separate must-haves from nice-to-haves

If everything reads as required, the model can't prioritise. Mark the non-negotiables clearly and list the bonuses separately so they nudge the score without dominating it.

4. Name your dealbreakers

On-site only? Specific certification or right-to-work requirement? Say so. Explicit constraints let the model flag candidates who don't fit, instead of ranking them highly and wasting your review time.

5. Cut the fluff

Mission statements and perks don't help screening - they add noise. Keep the requirements section tight and concrete. You can keep the inspiring copy elsewhere; the screening model wants signal.

A quick before/after

Weak: "We're looking for a talented engineer to join our fast-paced team and help build amazing products."

Strong: "Senior backend engineer. Must have: 5+ years Python, production experience with Postgres and AWS, built and owned APIs at scale. Nice to have: fintech background. On-site in London, 3 days/week."

The second version gives the AI everything it needs to rank accurately.

Let AI help write it

A complete example

Here is a full requirements section that screens well:

Role: Customer Success Manager (mid-level), reporting to the Head of Customer Success.

Must-haves:

• 3+ years managing a portfolio of B2B software customers
• Ran renewal conversations and owned a retention or net revenue target
• Comfortable running onboarding calls and product training remotely

Nice to have:

• Experience with HubSpot or Salesforce
• Worked with customers in professional services

Dealbreakers: Must be able to work UK hours. Hybrid, two days a week in our Birmingham office.

Every line is something a CV either shows evidence of or does not. That is what makes it scoreable.

Words that make ranking worse

• Adjectives instead of evidence: "passionate", "rockstar", "self-starter". Nobody's CV contradicts them, so they cannot separate candidates.
• Years with no context: "5 years' experience" of what, at what level? Tie years to the skill.
• Internal jargon: team names, product code names and acronyms only your company uses.
• Requirements that are really preferences: if you would interview someone without it, it is a nice-to-have.
• Exclusionary phrasing: "native speaker", "recent graduate" or "young team" can breach the Equality Act 2010 and skew screening. Describe the skill instead.

Checklist before you run the ranking

1. Are there between three and eight must-haves?
2. Is each must-have specific enough that a CV can prove it?
3. Is the seniority stated in plain words?
4. Are nice-to-haves clearly separated?
5. Are location, working pattern and right-to-work stated?
6. Have you removed anything that is not a real requirement?

If you want starting points, our job description templates cover six common roles with must-haves already separated.

Frequently asked questions

How long should the job description be? For screening, the requirements section matters more than total length. Three to eight must-haves and a few nice-to-haves is enough. The full advert can be 400 to 600 words.

Can I use the same job description for the advert and for screening? Yes, as long as the requirements section is specific. The selling parts of the advert (culture, benefits) do not hurt, they just are not used for scoring.

What if the ranking still looks wrong? Check the top five and the reasoning. If the wrong things are being rewarded, a requirement is usually missing or too vague. Tighten it and re-run.

If you're stuck, generate a first draft and refine it. SwiftShortlist can draft a structured job description from a job title, which you then tighten with the specifics above. Then upload your CVs and let the ranking do the rest.