Skip to content
SMB Agents AI

Hire & onboard

Screen job applicants against a simple scorecard

Goal: you read a short list of strong candidates, not the whole pile

Illustrative example: a composite of common setups, not a specific business's story.

Difficulty
No-code
Setup time
1 hour
Monthly cost
$20
Impact
Medium

The short version

An agent reads each application against your must-haves and produces a short scorecard; the owner makes every decision.

What they did

An agent reads each application against your must-haves and produces a short scorecard; the owner makes every decision. The owner started with a small pilot: one workflow, a few days of supervised use where every output was reviewed before it went out, then gradually let routine cases run with lighter oversight.

Agent/tool used: ChatGPT / Claude projects

How it went

What to aim for: you read a short list of strong candidates, not the whole pile. These are goals, not guaranteed results; we don't have measured numbers for this setup yet. Most of the value comes from consistency, the task happens every time instead of when someone remembers. Expect a week or two of tuning the prompt and rules.

Worth knowing

AI drafts can be wrong or off-tone. Keep a human approving anything customer-facing at first, don't paste sensitive data (card numbers, health records, SSNs) into general chat tools, and check each tool's data and privacy terms.

Before you switch it on

What can go wrong
It marks down a strong candidate over formatting, or repeats a bias baked into your past hires.
When to stop it
Pause it if its scores keep disagreeing with yours, or it cites things not in the application.
When you step in
You read every low-scored application before rejecting it; every hiring decision is yours.

Reads outside messages. Anything a stranger sends can hide instructions aimed at AI. Never let it pay, share logins or act on an urgent ask without you.

Copy-paste prompt

Role: [job title]. Must-haves: [list]. Nice-to-haves: [list]. Here is an application: [paste]. Give a 1-5 score per must-have with evidence quoted from the application, list open questions to ask in a phone screen, and do NOT consider age, gender, race, or other protected characteristics.

Replace anything in [brackets] with your own details.

Step-by-step setup

  1. 1Write down how you handle this today, in 5-10 bullets.
  2. 2Connect or export data from Gmail, Google Workspace.
  3. 3Set up ChatGPT / Claude projects and paste in the prompt, filling in the [placeholders].
  4. 4Test it on 5-10 real past examples and fix anything that's off.
  5. 5Run it with you approving every output for one to two weeks.
  6. 6Once you trust it, automate the routine cases and spot-check weekly.

Back-of-napkin math

How much time would this give you back?

6.9

hours back per month

$346/mo of your time

Assumes the setup removes about 60% of the manual time. First-pass screening only. You still interview humans. Your numbers, your call. The robot refuses to promise anything.

Part of The Restaurant Pack.

Keep going

Published 2026-10-01.