As of 13 August 2026, AI can only partly rank candidates for a job.

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, power-user skill, and roughly 1 hour until you can act on the result.

   How to actually do it

   1. Open the job description and write a short list of essential and desirable criteria that relate directly to the work.
   2. Create a consistent scoring scale for each criterion, including what counts as no evidence, partial evidence and strong evidence.
   3. Gather the same types of application material for every candidate and remove unnecessary personal data before using an AI tool.
   4. Paste the prompt into the chatbot, replace the bracketed sections with the job criteria and anonymised candidate evidence, and ask it to produce the scoring table and provisional order.
   5. Copy the table into a recruitment record and compare every score with the cited CV or application evidence, correcting unsupported scores and recording missing evidence.
   6. Ask an HR colleague to check the criteria, weighting, anonymisation and treatment of employment gaps or reasonable adjustments for consistency and potential discrimination.
   7. Use the checked ranking only as one input to a human-led shortlist and record the reasons for the final decision separately from the AI output.

   Prompt

   Act as a recruitment analyst, not the hiring decision-maker. Rank the candidates below against the job criteria using only evidence in the supplied material. Do not infer or use age, sex, gender identity, race, ethnicity, nationality, religion, disability, health, pregnancy, caring responsibilities, sexual orientation, marital status, or any other protected or irrelevant personal characteristic. Do not treat name, address, school, accent, employment gaps or writing style as evidence of ability unless the job criteria explicitly and lawfully require it. First create a scoring table with one row per candidate and one column per criterion. For each score, quote or paraphrase the supporting evidence and mark missing evidence as missing rather than guessing. Then give a provisional order, explain any ties or uncertainty, list candidates who need the same follow-up question, and identify criteria that may be vague, inconsistent or difficult to justify. Do not recommend rejection, make a final hiring decision, or claim that the ranking is legally fair. State that a human must review the criteria, evidence, adjustments and final decision.
   
   Job title: [JOB TITLE]
   Essential criteria: [ESSENTIAL CRITERIA]
   Desirable criteria: [DESIRABLE CRITERIA]
   Scoring scale: [SCORING SCALE]
   Candidates and evidence:
   [CANDIDATE MATERIAL]
   
   Use the same criteria and scoring scale for every candidate. Keep the output suitable for an auditable recruitment record and flag any information that should not have been included.

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