Skip to content
KoshKramaniti

ChecklistHuman oversight · Free templateAdded 26 Oct 2026

Shortlist job applicants fairly with AI help

Write down what the job needs and what must never count against anyone before AI sorts applications, and keep every call, reject or hire decision with a named person.

Use it when For small businesses hiring telesales, field sales, support or accounts staff who want AI to help sort a pile of applications without it quietly dropping good people.

Download template

It is a simple text file. Open it in any text editor, or paste it into ChatGPT, Claude or Gemini and ask for help filling it in.

How to use it

  1. Fill in the role and the rules below before you open any application. Rules written after reading the applicants tend to bend towards whoever you liked first.
  2. Give the AI tool only the rules and the application text. Remove photos, and do not ask it to judge things that are not in your must-haves. Ask it to sort each application into "meets the must-haves", "unclear" or "does not meet", and to quote the line from the application that supports each sort.
  3. The decision owner reads every pile, including "does not meet", before anyone is called or rejected. Any reject is the owner's decision, not the tool's. Record each decision and the reason in the log.
What you will need before you start
  • A plain description of the job: what the person will do on a normal day, where, and with whom
  • The things someone truly needs on day one, and the things they can learn on the job
  • The applications you already have, or the place they arrive (a job portal, email, WhatsApp, a referral list)
  • One person who will own the hiring decision and read the AI's sorting before anyone is contacted

The template

Fill in the blanks for your business. This is what you get when you download or copy it.

Role

Job title:

What the person will do on a normal day:

Where they will work (office, field, which areas):

Who they report to:

Decision owner for this hire:

Date the rules were written:

Must-haves

Things the person needs on day one. Keep this list short, and make each line something you can see in an application or check in a call.

Good to have

Things that help but can be learned on the job. These never decide a reject on their own.

Never counts against an applicant

Things the AI tool and the people reviewing must ignore. Check this list with a lawyer or an HR adviser before you use it.

  • Gender, age, religion, caste, marital status, disability, or where someone is from
  • Name, photo or how a name sounds
  • Gaps in work history, unless the role has a stated reason to ask about them
  • English fluency, unless the job itself needs it and it is listed in the must-haves

What the AI tool may do

  • Sort applications against the must-haves into "meets", "unclear" or "does not meet"
  • Quote the line in each application that supports its sort
  • List questions to ask in a first call

What the AI tool must not do

  • Reject, rank by score, or contact any applicant
  • Guess at anything the application does not say
  • Use anything on the "never counts" list, even as a tiebreaker

Decision log

Copy this block for each applicant.

  • Applicant reference (not the full name if the log is shared):
  • AI sort and the line it quoted:
  • Owner's decision (call, hold, reject, hire):
  • Reason in one plain line:
  • Decided by and date:
  • Disagreed with the AI sort (yes or no, and why):

Stop rules

  • Do not use the AI sort until the must-haves and the never-counts list are written down and the decision owner has signed them.
  • Do not send any rejection, call or offer from the tool. A person sends every message.
  • If the AI sort and the owner disagree more than now and then, stop using the tool and rewrite the must-haves.
  • Do not paste ID numbers, salary slips or other private documents into an AI tool.
  • If an applicant asks how they were assessed, the decision owner answers, using the log.

Owner and review

Rules owner:

Review the rules after this hire is filled, or sooner if the role changes.

See an example

A made-up example, so you can see what a filled-in version looks like.

This is an illustrative scenario, not a client case study or a measured result. Do not reuse its details as facts about your work.

A sample rooftop solar installer in Bengaluru is hiring field sales staff to visit homes and apartment associations, explain rooftop systems and book site surveys. Applications come in from a job portal and through WhatsApp referrals. The owner wants AI to help sort them, but worries that the tool will quietly drop good people for the wrong reasons.

Sample inputs

Demonstration inputs only. No applicant names, counts or results are used. Role notes from the owner: visits homes across the city on a two-wheeler, speaks Kannada and either English or Hindi with customers, can explain an electricity bill, logs visits in a phone app. Two sample applications: one from a person with retail sales experience at an electronics store and a driving licence, whose application does not mention languages; one from a recent graduate with no sales experience who has done a summer job doing door-to-door surveys and lists Kannada and Hindi.

Example output

Must-haves

[Fact within this demonstration] Valid two-wheeler licence. Speaks Kannada and either English or Hindi. Willing to travel across the city daily. Any experience of speaking to customers face to face.

Never counts against an applicant

[Fact within this demonstration] Degree or college name, age, gaps in work history, and which part of the state someone is from.

AI sort

[Fact within this demonstration] The retail applicant is sorted "unclear", quoting the retail experience and licence lines and noting that languages are not mentioned. The survey applicant is sorted "meets", quoting the door-to-door survey line and the languages line, with "licence not mentioned" as a question for the first call. [Inference] Neither application gives enough to reject. Both missing details can be checked in a short call, which is cheaper than losing a good applicant.

Owner's decision

[Recommendation] Call both, ask the same questions about languages, licence and travel, and record the reason for each decision in the log. This recommendation is not the business's decision.

How to tell it is done well

The must-haves were written before any application was read; every must-have can be seen in an application or checked in a call; the never-counts list was checked with a lawyer or HR adviser; every AI sort quotes the application; the decision owner read every pile, including "does not meet"; every call, reject and hire has a reason in the log and a named person who made it; no message went to an applicant from the tool.

Where a person should decide

The template helps you set fair rules before AI sorts applications and keeps every hiring decision with a person; it does not give legal or HR advice, prove that a tool is free of bias, or tell you which hiring laws apply to your business. Check your never-counts list and how you store applicant data with a lawyer or HR adviser. A sorted pile is not a decision to call, reject or hire anyone. Keep consequential actions with the named human owner.

The template keeps hiring decisions with a person and sets rules before AI sorts applications; it does not give legal or HR advice, prove a tool is free of bias, or tell you which hiring laws apply to you.

Edit your own copy on this site
About this version

New in Kosh at version 1.1. Provider-neutral Markdown instructions; no installation or platform compatibility is implied. Runtime behaviour has not been verified across AI providers. Reuse terms have not yet been published. Background reading: Make the Decision Route Visible Before You Build and Founder Review Should Be Conditional, Not Constant, which discusses research on people following AI advice in hiring decisions. Working out where AI can help and where a person must decide is part of Kramaniti's Foundation Strategy audit.

How we write Kosh templates