As of 13 August 2026, AI can only partly compare candidates against your job criteria.

What to actually do

1. Hand it to a person

   The route this page recommends

   A person who owns the outcome does this end to end, worth it when the failure is dear.

2. Use a tool built for this

   Second choice

3. Do it yourself

   The distant third

   A chat interface, chat-fluent skill, and roughly 30 minutes until you can act on the result.

   How to actually do it

   1. Open the current job description and your agreed selection criteria, then separate essential criteria from desirable criteria and remove requirements that are not genuinely needed for the role.
   2. Write down the scoring rules for each criterion, including what counts as evidence and how missing evidence will be treated; if you cannot define a rule, mark that criterion for human review.
   3. Gather the CVs, application answers and other permitted selection materials, remove unnecessary personal details and label each applicant with a neutral identifier such as Candidate A.
   4. Paste the prompt and the criteria, scoring rules and labelled application materials into a chatbot, then ask it to produce the evidence table and flag ambiguous or potentially unfair criteria.
   5. Compare every cited passage in the output with the original application materials, correcting scores where the model has misread, invented or over-interpreted evidence.
   6. Ask an HR-qualified colleague to check the criteria, the treatment of missing information and the proposed shortlist for consistency and possible discrimination before anyone is rejected or invited to interview.
   7. Record the human reasons for each shortlist decision and send only the approved candidate communications after the responsible hiring manager has made the decision.

   Prompt

   Compare the candidates below against the job criteria. Use only evidence explicitly present in the materials and do not infer age, sex, gender, race, ethnicity, nationality, religion, disability, health, pregnancy, sexual orientation, marital status, caring responsibilities or any other protected or private characteristic. Do not use names, photographs, addresses, schools or other proxies unless they are directly relevant, objectively justified job criteria. Do not make a final hiring decision.
   
   First, list the essential and desirable criteria separately. Then create a table with one row per candidate and these columns: criterion, evidence from the application, whether the evidence meets the criterion, confidence, and missing information. Score only where I have supplied a scoring rule. If no scoring rule is supplied, leave the score blank and explain what evidence would be needed. Treat absence of evidence as missing information, not evidence that the candidate lacks the skill. Flag criteria that are vague, overlapping, subjective or likely to disadvantage a group. Flag any candidate comparison that cannot be made fairly from the supplied information.
   
   Finish with a neutral shortlist for human review, grouped as meets the stated evidence, unclear from the evidence, and does not meet the stated evidence. For every recommendation, cite the exact application evidence supporting it. Ask me questions about unclear criteria before applying them.
   
   Job criteria:
   [PASTE THE AGREED JOB CRITERIA]
   
   Scoring rules, if any:
   [PASTE THE AGREED SCORING RULES OR WRITE NONE]
   
   Candidate applications:
   [PASTE REDACTED CVs, APPLICATION ANSWERS OR OTHER MATERIALS, LABELLED CANDIDATE A, CANDIDATE B AND SO ON]
   
   Return a comparison for human review only. Do not recommend rejecting or appointing anyone, and do not claim that the process is legally compliant.

   Open it prefilled in ChatGPT or Claude, or copy it into Gemini, which takes no prefill link.