How to Write Must-Have Criteria an AI or a Human Can Screen Against

To write must-have criteria that an AI or a human can screen against, keep to four to six, word each as something visible in a CV such as a duty, tool, qualification or result, avoid vague traits, and separate true needs from nice-to-haves. Good criteria make resume screening software and human reviewers produce the same answer.

Why criteria decide everything

Every screening method, whether a person with a highlighter or software that scores CVs, is only as good as the list it checks against. If the list is vague, people disagree and tools guess. If it is bloated, almost every candidate fails something and good people are lost. If it is wrong, the best-looking shortlist will be a shortlist of the wrong people.

Most teams spend an hour on the job ad and ten minutes on the criteria. That is backwards, because the criteria are what actually filter candidates. This guide shows how to write them so a human and an AI can both apply them the same way, with before and after examples, a template and a checklist. For related reading, see /ai-resume-screening and /job-description-template.

Must-have, nice-to-have and not needed

Sort every requirement into one of three groups.

• Must-have: the person cannot do the core of the job without it, or could not reasonably learn it in the first few weeks. Examples are a legal licence, a core tool the job uses all day, or a type of experience the role is built around.
• Nice-to-have: it helps, and you would like it, but another strength could replace it or it can be learned.
• Not needed: it came from a template, a previous job holder or habit.

A good test for each item: "If a candidate were excellent in every other way but lacked this, would I still interview them?" If yes, it is not a must-have. Another test: "Could they learn this within a month?" If yes, it is a nice-to-have.

Be honest about degrees and years. "Degree required" and "five years" often come from habit. If a person with three years and strong results would be interviewed, five years is not a must-have.

How many is the right number

Four to six must-haves is a sensible range for most roles. With fewer than three, almost everyone passes and the filter does nothing. With more than seven, the combinations get so narrow that you screen out people who would do the job well, and the criteria usually contain things that should be nice-to-haves.

Nice-to-haves can run to three or five. They should break ties and not exclude. In a scoring system like Rezumz, the must-haves carry most of the score and the nice-to-haves a small part, so a missing nice-to-have barely changes the result while a missing must-have changes it a lot. That is another reason to keep must-haves truly essential: each one has a big effect on how a person ranks.

Word each one so it can be found

The test of a good criterion is whether you can point to a line in a CV that proves it. Words like "excellent communicator", "self-starter", "passionate" and "culture fit" cannot be found on a page, so a human will guess and an AI will invent a reason.

Turn traits into observable evidence:

Some traits cannot be seen in a CV at all, such as attitude. Do not put them in the must-have list. Test them later, with structured interview questions and an interview scorecard. See /interview-feedback-form for a way to record them.

Use the words candidates use

A screener, human or software, looks for words. If your criterion says "customer relationship management" and the CV says "CRM", a human will connect them and a simple tool may not. Rezumz allows for slightly different wording, but no tool covers every phrase, and a CV that does not mention a skill scores as lacking it.

So write criteria with the common names for things: the tool name and its category, the qualification and its abbreviation, the duty and the usual job title. For example, write "Excel or other spreadsheets for reporting" rather than "data literacy". If a skill has two common names, include both in the criterion note.

Check the list against five CVs of people who did the job well. If their CVs use words your list does not, add them. This takes ten minutes and is the best single check of whether your criteria will work.

Before and after examples

Here are two job-description lines turned into usable criteria.

Original: "We need a dynamic, results-driven sales professional with excellent communication skills and a proven track record, ideally with a degree and 7+ years of experience."

Rewritten must-haves: 1. Has held a quota-carrying sales role (any industry). 2. Has managed a pipeline in a CRM. 3. Has run prospecting calls or meetings with decision makers. 4. States at least one result in numbers, such as revenue or deals closed. Nice-to-haves: experience in our industry, a second language, a degree.

Original: "Looking for a detail-oriented admin who is a team player and good with computers."

Rewritten must-haves: 1. Has worked in an admin or office support role for at least a year. 2. Has used Word and Excel or equivalents daily. 3. Has handled scheduling or calendars for others. 4. Has handled records, filing or data entry with accuracy checks. Nice-to-haves: experience with a named office system, front-desk experience.

Notice that each rewritten line names something that can be found in a CV, and the unfindable traits move to the interview.

A fill-in template

Copy this word-for-word and fill it in before reading any CVs.

"Role: [title]. Date written: [date]. Written by: [name]. Must-have 1: [duty, tool, qualification or result] (also accept: [other wording]). Must-have 2: ... Must-have 3: ... Must-have 4: ... Gate, if any: [licence or legal requirement] (no gate means leave blank). Nice-to-have 1: ... Nice-to-have 2: ... Ask at interview, not on CV: [traits such as teamwork or attitude]. Review date: after the first 20 CVs, check the list for missed wording."

Keep the filled copy with the job. If someone asks later why a person was not shortlisted, this is the first thing you show.

A scoring checklist for your criteria

Score each criterion 0 for no, 1 for partly and 2 for yes on each question, then add up. Do it for every criterion on your list.

A criterion scoring 10 to 12 is ready. A score of 6 to 9 needs rewording. Below 6, move it to nice-to-have or drop it. Aim for every must-have to reach at least 10 before you open the first CV.

Worked example with example numbers

These numbers are examples only. A company screens 90 CVs for a support role against a first draft of eight must-haves, including a degree, five years of experience and "excellent communication". Only 4 of 90 CVs meet all eight, and two of those are not strong. The recruiter suspects the list is too tight.

She applies the checklist. The degree fails the "would I refuse to interview" question, since two of her best current team members have none. "Excellent communication" is vague. Five years fails the habit test, because three years is enough if the person has handled complaints. She cuts the list to five findable must-haves and moves the degree to a nice-to-have.

Rerun on the same 90 CVs, 31 now meet all five must-haves. She reads the top 20 and a sample of 10 from the rest, shortlists 8 and interviews 6. Because the criteria were dated and saved, she can explain every outcome. In an example like this, tightening the wording rather than tightening the standards is what widens the pool without lowering the bar.

Common mistakes

Mistake one: copying the old job ad. Mistake two: eight or ten must-haves. Mistake three: vague traits in the list. Mistake four: demanding degrees or year counts out of habit, which can screen out capable people with non-traditional paths. Mistake five: writing criteria after reading the first CVs. Mistake six: never checking the wording against how strong people describe their work. Mistake seven: letting the list drift between candidates. Mistake eight: treating the list as fixed after you see it fails, rather than reviewing it after 20 CVs.

Examples for different roles, and keeping criteria fair

The same method works across role types, but the findable evidence differs.

Customer support: handled a support queue, chat or phone line as a main duty; used a ticketing tool; dealt with complaints or refunds; worked to response-time targets. Warehouse or operations: operated named equipment or a stock system; followed safety procedures and held any required certificate; handled deliveries and records. Marketing: ran campaigns in named channels; reported results with numbers; used named tools for email, analytics or design. Software: built and shipped features in a named language or framework; used version control and code review; supported a live system. Finance or admin: prepared reconciliations, invoices or reports; used spreadsheets and an accounting or office system daily.

In every case, ask which line on a CV would convince you. If the answer is "I would just know when I saw it", the criterion is not yet written well enough. Add detail until two colleagues, reading the same CV separately, would give the same answer.

Criteria also need a fairness check, because they decide who gets seen. School names, specific employers and unbroken careers are stand-ins for the work, and they exclude capable people such as a parent returning after a break, a career changer or a self-taught candidate. Replace stand-ins with direct evidence of the skill or result. Where a legal or safety need requires a qualification, keep it and say why. Rules on what you may filter on differ by country, and some places have specific laws for automated tools, so check local law and keep a person making each decision. After the first batch is scored, look at who was filtered out and ask whether the criteria, rather than the candidates, were the problem.

Where Rezumz fits, and who it is not for

Rezumz is an AI hiring workspace for small and mid-size hiring teams. When you add a job, you paste the description and Rezumz lists the must-have and nice-to-have skills it found. The score depends on that list, so you check and edit it, then it is used to score every CV, with 80% of the score from must-haves and 20% from nice-to-haves, and the matched and missing items shown with quoted CV lines. The same criteria become the default interview scorecard, so interviewers rate what the job needs. If scores look wrong, edit the must-haves and add the CVs again. Nothing is rejected for you: people decide.

It does not write your judgement for you, and a CV that does not mention a skill scores as lacking it. It is not meant for large enterprises needing deep HRIS or payroll links or one-click multiposting to job boards. If you hire only occasionally and read each CV yourself, a written list on paper may be enough; see /free-applicant-tracking-system. The Team plan is for four or more recruiters.

Prices: a free plan with 2 free candidate analyses; $29 a month, $199 a year, or $499 lifetime (limited); Team is $79 a month or $699 a year for up to 10 teammates. See /pricing, try /demo, and see /recruitment-software and /compare for the wider picture.