For Recruiters and Talent Acquisition

Protecting Your Judgement: AI Tools for Recruiters Who Still Want to Hire Well

AI screening tools promise speed but often filter out the candidates who surprise you in the best ways. When LinkedIn Recruiters AI or Eightfold AI handles your first pass, you risk embedding hiring bias at scale before you ever have a conversation. The question is not whether to use these tools, but how to use them in ways that keep your best skill intact: spotting potential that a CV does not show.

These are suggestions. Your situation will differ. Use what is useful.

Download printable PDF

Use AI to Write Job Descriptions, Then Rewrite Them for Humans

ChatGPT will write a job description that optimises for keyword matching. It will list every tool, every framework, every requirement. What it produces attracts people who match the description perfectly but often miss those who could actually do the work. After you generate your first draft, remove the jargon and the long lists. Write in plain language about what the person will actually spend their time on. Test your final version: can someone with adjacent experience understand what they would be doing?

›Generate a job description in ChatGPT, then strip out 40 percent of the requirements and technical terms
›Replace skill lists with three to four sentences about the actual work and the problem the role solves
›Run your rewritten description past two people in similar roles and ask what they would change

Screen for Flags, Not Fit. Let Humans Decide Fit.

HireVue and similar tools claim to assess cultural fit and potential. What they actually do is measure how well a candidate matches patterns from your past hires. This is where systemic bias enters at scale. Instead, use AI screening to filter for red flags only: incomplete applications, missing required credentials, nonsensical responses. Everything else goes to you. Your instinct for spotting potential is not a bias you need to eliminate. It is a skill you need to protect.

›Configure your screening tool to flag incomplete applications and missing certifications, nothing else
›Read the CVs of candidates who fail the AI but have unusual career paths before rejecting them
›When you see a career progression that does not fit the mould but shows progression, that is usually a strong hire

Source Beyond the Algorithm. Hunt for What You Cannot See.

Eightfold AI and LinkedIn Recruiters AI will show you candidates who look like your current team. They will miss the unconventional hire who learned a skill outside the usual path and turned out to be your best performer. Use algorithmic sourcing for volume, but keep a separate sourcing stream for candidates the algorithm would skip. This means university programmes that are not target schools, career changers, people with extended gaps, people whose job titles do not match their actual work. Your best hires often come from places the algorithm never looks.

›Source 30 percent of your pipeline from non-obvious places: bootcamp graduates, career returners, people in adjacent industries
›When you find a strong unconventional candidate, ask your AI tool why it did not rank them. Often the reason reveals a bias you can fix
›Build a separate tracker for candidates the algorithm rejected but you wanted to talk to. Track their performance after hire

Use AI to Prepare for Conversations, Not Replace Them

The early interactions in your hiring process are where candidate experience either builds or breaks your employer brand. If AI handles all initial contact, your organisation becomes faceless. Use AI to prepare your screening call: summarise their CV, note unusual details, flag questions that matter. Then you make the call. Your voice, your questions, your ability to hear what is not written down. This is where you spot the person who fumbles their words in writing but thinks brilliantly in conversation.

›Generate five interview questions from the candidate's CV using ChatGPT, then replace half of them with questions only a human recruiter would ask
›Always make first contact yourself. AI can send the next email or schedule the next step
›When a candidate has an unusual background, spend extra time on the call understanding their choices. That conversation tells you more than any assessment tool

Track What Your AI Gets Wrong. Adjust or Stop Using It.

Most recruiters never measure whether their AI screening tool actually improves hire quality. You need baseline data. For three months, track which candidates the AI rejects that you would have wanted to interview. Track which candidates it advances that fail your screening call. If your tool rejects more than five percent of people you would have spoken to, it is filtering for conformity, not capability. You have the right to turn it off or reconfigure it. Your hiring outcomes matter more than efficiency.

›Every month, pull the list of AI rejections and spend one hour reading the ones you wanted to see anyway
›Calculate your AI tool's false rejection rate: candidates it rejected that you would have interviewed
›If a tool has rejected more than 100 candidates in six months, audit a random sample of 20 to see what it is actually filtering for

Key principles

1.Use AI to eliminate noise, not to choose between candidates.
2.Your instinct for spotting potential is a strength to protect, not a bias to remove.
3.Job descriptions written by AI attract people who match keywords, not people who do the work well.
4.The candidates AI misses are often your best hires.
5.If you cannot explain why AI rejected a candidate, that tool should not be making that decision.

Key reminders

When you hire someone the algorithm ranked in the bottom 20 percent, save their file. You are building evidence of your tool's blind spot.
Ask your AI tool to show you the candidates it ranked first and last for a recent role, then read both CVs side by side. You will see exactly what it optimises for.
Before you buy any new recruiting AI, ask the vendor for false rejection rates from organisations like yours. If they cannot provide this, they do not measure it.
Keep one sourcing channel completely free of algorithmic input. Track those hires separately. Compare their performance to hires from algorithmic sources.
When writing a job description, imagine the person who will actually be best at the role is reading it. Would they apply? If not, rewrite it.