World Model Readiness
Engraved role-check instrument

For you

Role Checks

Module · Does the filter help you see people

The Recruiter's AI Check

AI can rank a thousand applications before your first coffee, and that is both the promise and the trap. A tool that filters people you will never meet decides careers on your behalf, and in most jurisdictions hiring is a high-risk use of AI for good reason. This module checks the five habits that keep you in the loop: healthy skepticism of the tool, a candidate experience worth having, awareness of the bias the model learned, human contact where it counts, and reasons for rejection you could defend.

Question 1 of 5 · You question the tool

Do you know how your screening tool ranks candidates, or do you trust the score?

A ranking you cannot explain is a decision you cannot defend. Vendors call it objective, but the model learned from someone's past hires, with all their patterns baked in. If you cannot say why it scored someone low, you are outsourcing judgment to a black box.

Question 2 of 5 · Candidates get a real experience

Does AI in your process treat candidates like people, or like inputs to a funnel?

The candidate remembers how it felt to apply. Auto-rejections with no signal, chatbots that loop, interviews with no human until round three: these cost you the people you most wanted, and they talk.

Question 3 of 5 · You watch for learned bias

Are you alert to the bias an AI tool can inherit and amplify at scale?

A model trained on your past hires learns your past preferences, including the ones you would never write down. The difference from a biased human is scale: it applies the same skew to every candidate, silently, all day.

Question 4 of 5 · A human makes the call

At the points that change a person's path, is a human deciding, not the model?

Some decisions are too consequential to leave to a ranking: who gets an interview, who gets rejected, who gets the offer. Automation can inform those, but a person should own them, and be able to say so.

Question 5 of 5 · Rejections have real reasons

For anyone the process rejects, could you give a reason you would stand behind?

The screen dropped them means nothing to the candidate and nothing to a regulator. If the only reason for a rejection is a score you cannot unpack, you have made a decision you cannot explain or defend.

For the statistics · one click each

Three questions for the public picture

These do not affect your score. They feed the anonymised, aggregated statistics; groups under 8 respondents are never shown.

How much of your candidate screening relies on AI tools?

None
Sourcing only
Ranking and filtering
Most of the funnel
End to end

Do you know how your screening tool actually ranks candidates?

No idea
Roughly
Fairly well
In detail
I use no such tool

Has anyone checked your AI hiring tools for biased outcomes?

Never
Informally
Once, at purchase
On an ongoing basis
I do not know

Your context

Used to calibrate the report. Company size and sector remain in the anonymized dataset; your email does not.

What the five levels look like

Every dimension in this assessment is scored 1 to 5. This is what the levels mean, dimension by dimension. The graded report diagnoses where your own answers land and what to do about it.

You question the tool

  1. 1Trust the score blindly
  2. 2Assume it is fair
  3. 3Curious but unsure
  4. 4Understand the basics
  5. 5Probe and challenge it

At the low end: Trusting a score you cannot explain means the tool recruits, not you. Ask the vendor what the model weighs and what data it learned from; if they cannot tell you, that is your answer. What good looks like: Probing and challenging the tool keeps you the decision-maker. Keep asking what it optimises for; a screening model quietly rewards whatever the last generation of hires had in common.

Candidates get a real experience

  1. 1Fully automated, cold
  2. 2Efficient, impersonal
  3. 3Mixed
  4. 4Human touches kept
  5. 5Warm and human-led

At the low end: A cold automated process wins efficiency and loses the candidates who have other offers. Add a human touchpoint where it matters most: the first real conversation and the rejection. What good looks like: A warm, human-led process is a competitive advantage in a market where everyone else automated. Keep the AI in the back office and the people at the front.

You watch for learned bias

  1. 1Never considered it
  2. 2Assume tools are neutral
  3. 3Aware, no action
  4. 4Watch the outcomes
  5. 5Monitor and correct

At the low end: A tool that learned from biased history will reproduce it faster than any human could. Start by asking whether your screened pool looks different from your applicant pool, and why. What good looks like: Monitoring outcomes and correcting for skew is what makes automated screening defensible. Keep doing it; bias is not a bug you fix once, it is drift you watch for.

A human makes the call

  1. 1Tool decides alone
  2. 2Rubber-stamp the tool
  3. 3Human on close calls
  4. 4Human owns key gates
  5. 5Human owns, tool advises

At the low end: Letting the tool make the call at decisive gates is the exact use most likely to be unlawful and unfair. Put a human in charge of who advances and who is rejected, starting now. What good looks like: A human owning the decisions with the tool advising is the right shape for high-stakes hiring. Keep the override real; a checkpoint nobody ever uses is not a checkpoint.

Rejections have real reasons

  1. 1The tool decided
  2. 2Score, no reason
  3. 3Vague rationale
  4. 4Reason on request
  5. 5Documented, defensible

At the low end: "The tool decided" is not a reason a candidate or a tribunal will accept. Make sure every rejection traces to a human-legible criterion, not just a number. What good looks like: Documented, defensible rejection reasons protect the candidate and you. This is the paper trail that turns a contested decision into a settled one.