Paste a CV into most AI screening tools and you get a number. 82% fit. Rank 3 of 11. Strong match. It feels like progress, because a number is easier to act on than four pages of prose. The trouble is what the number is made of: a document the candidate wrote about themselves, read against a requirement list that was itself a rough approximation, compressed into a single figure that hides every gap in the source material.
The output looks like measurement. It is closer to a summary with the uncertainty removed.
Should recruiters use AI to score candidates?
Answer-first: use AI to organise what the material establishes, not to decide who is best. Scoring collapses two different things, what the evidence shows and what it is worth, into one number, and only one of those is a recruiter's job to own.
There is a compliance dimension too. Where an automated tool substantially assists a hiring decision, several jurisdictions attach obligations to it, including New York City's Local Law 144 bias audit and notice requirements, and the EU AI Act, which classifies recruitment and selection systems as high risk under Annex III. A tool that maps evidence and leaves the decision with a person sits in a very different place to one that ranks people.
But the practical argument matters more day to day. A ranking tells you nothing you can take to a client. "Third of eleven" does not survive the question "why?". A mapped set of evidence answers it in one sentence per requirement.
What is candidate evidence?
Candidate evidence is what the supplied material actually establishes about a candidate against a specific requirement, along with where in the material it came from. It is not a judgment about the candidate. It is a description of what you currently have in front of you.
That distinction only becomes useful when you stop treating evidence as binary. Four states cover almost every real case.
| State | What it means | What to do next |
|---|---|---|
| Evidenced | The material describes specific work that meets the requirement, with enough detail to discuss. | Use it. Quote the detail in your submission. |
| Partly evidenced | Related work is described, but the scale, recency or exact scope does not fully meet the requirement. | Ask one targeted question to close the gap. |
| Claimed | The candidate asserts it, with no supporting detail. A skills list is a claim. | Treat as unverified. Probe it in the screen. |
| Not established | Nothing in the supplied material speaks to this requirement either way. | Ask. Do not infer, in either direction. |
Not established is not the same as cannot do.
This is the point recruiters most often need to defend internally. A candidate who never mentioned mentoring in a two page CV has not failed a mentoring requirement. The CV simply does not address it. A scoring model marks that absence down. A recruiter reads it as a question for the call. Those two behaviors produce different shortlists, and the second one is the defensible one.
Evidence, then judgment
- Supplied materialCV, screening call notes, work history
- Read against the RoleEach requirement, one at a time
- Classified with sourcesEvidenced, partly evidenced, claimed, not established
Why separating evidence from judgment changes the conversation
Three things get better immediately.
- Your screening calls get sharper, because you go in with the three specific gaps rather than a general plan to "cover the requirements".
- Your client conversations get concrete. "Four years of production Python, evidenced in the payments work at Northwind; team leadership is claimed but not detailed, which is my first question on the call" is a different level of professional than "strong candidate, 85% match".
- Your own bias gets a check. When you have to say which requirement a decision rests on and where the evidence for it came from, gut-feel rejections become visible.
Where AI genuinely helps
Reading twenty pages of material against twenty-four requirements and tracking which line supports which claim is slow, repetitive and easy to do inconsistently at five o'clock. That is a good use of a model. Deciding whether the gap in scale matters for this particular client is not.
Where it should be kept out
- Inferring protected characteristics, or anything that proxies for them, from names, dates, schools or gaps.
- Inferring personality or culture fit from writing style.
- Automatic rejection of anyone, at any threshold.
- Producing a single overall fit figure that hides which requirements it is made of.
How Creo Access implements this
Creo Evidence brings a candidate's material together on a Role, pasted CV text or an uploaded PDF or DOCX, along with conversations captured from screening calls, and reads it against the requirements you calibrated. Every requirement comes back as evidenced, partly evidenced, claimed only or not established, with the exact lines in the source material it came from, plus conflicts and the questions worth asking next.
Evidence mapped against the Role’s requirements
Select a requirement to see its citations and the question to ask next.
Each requirement carries an evidence state and the lines that support it. Creo does not score or rank candidates.
Read this figure as text
- Evidenced: Owned a month-end close in a group finance function. CV, Wexmoor Group (2021–2025), Owned group close for eleven entities, reduced to five days.
- Partly evidenced: Worked on an ERP migration end to end. CV, Wexmoor Group, Finance workstream lead on ERP replacement.
- Claimed: Comfortable presenting to auditors. Candidate screen, 08:15, I handled the auditors every year.
- Not established: Built reporting for a plc board. Reading, Absence in the material is not absence of the capability, so it stays a gap rather than a negative.
- Conflict: Managed a finance team. CV, Wexmoor Group, Led a team of four analysts.
The counts at the top are deliberately counts and not a percentage. There is no ranking across candidates, no recommendation and no reject action, because those are decisions and they belong to the recruiter. What the workspace gives you is the part that is tedious to do by hand and easy to do inconsistently.
From there, a Candidate Case can become a client-facing submission that carries the same discipline: what the evidence shows, what is still open, and nothing dressed up as certainty it does not have.
Try it on a real candidate
See what a CV actually establishes against your requirements
Map one candidate against a Role and read the evidence with sources. Start free for 14 days. Then choose monthly or annual billing. Cancel anytime.
Start 14-day free trial