How to write job screening criteria an AI tool can help review

Before asking an AI tool to screen resumes, agree what a useful review should establish. "Find the best candidates" gives the system and the hiring team too much room to invent priorities.

Write criteria around the actual work, then describe what evidence would support each one.

Separate required capabilities from preferences

Start with the responsibilities in the job brief. Ask which capabilities are necessary on entry and which can be learned. Keep convenience preferences, such as familiarity with one particular tool, separate from essential qualifications unless the role genuinely requires them.

Avoid using a former employee's background as the template for every future hire. A successful person may have credentials that were not the reason they succeeded.

If a requirement affects eligibility or legal obligations, have the responsible specialist confirm the wording for the locations involved.

Define evidence and uncertainty

For a reporting role, you could adapt a rubric like this:

Criterion	Evidence to examine	If the resume is silent
Builds recurring reports	A described reporting task or relevant project	Mark unconfirmed
Explains findings to colleagues	An example of communicating analysis	Ask a focused question
Uses a required technical method	Work, study, or project evidence of the method	Check through the agreed assessment

These are examples to adapt. The resume may not be the right place to establish every criterion. Some capabilities need a conversation or practical exercise.

Tell reviewers what to exclude

Keep criteria related to the job and avoid inferring personal characteristics from names, photographs, or unrelated biographical details. Decide how career breaks, nontraditional experience, and equivalent qualifications should be considered by the human team.

Ask the tool to distinguish "not found in the supplied material" from "candidate does not have this capability." Those statements are not interchangeable.

Test the rubric before a live campaign

Have reviewers apply the criteria independently to authorized examples. Discuss disagreements and revise ambiguous wording before using the rubric at scale.

Check whether the tool can point to the evidence behind each result. If it invents a qualification or treats an uncertain field as a firm rejection reason, record the error and change how the workflow uses that output.

Keep a version of the agreed criteria with the vacancy. If the requirements change, decide how earlier applications will be reviewed under the new version so candidates are not compared against different standards without anyone noticing.

Further reading

• Why AI in hiring tech fails without structure |UK
• Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile

Related guides

• Create a candidate assessment scorecard
• Interpret matching scores carefully
• AI resume screening: how it works and where it can go wrong