Why evidence-based answers beat confidence scores
Confidence scores hide reasoning. Evidence exposes it. Why "shown on line 12 of the CV" wins arguments that "85% match" never will.
Evidence-based answers show you the exact source behind every claim about a candidate. Confidence scores hide it. When a screening tool says a candidate is an "85% match", no one can explain what that number means. When it says "mentioned on line 12 of the CV, click to see it", anyone can check it in one second. That difference decides whether you can defend your shortlist.
The Monday morning test
Picture the conversation every recruiter knows. You send a shortlist to the hiring manager. They pick one name and ask, "Why her?"
If your tool gave you a score, your honest answer is "the AI said 85%." That is not an answer. It is an appeal to authority, and the hiring manager knows it. You cannot say what the 85 covers, what the missing 15 is, or whether the number would survive a second look.
If your tool gave you evidence, your answer is different. "She led payment infrastructure at a fintech for three years. It is on line 12 of her CV. Click here, I highlighted it." The conversation moves from "do we trust the tool" to "do we want this experience." That is the conversation you were hired to have.
A confidence score is a compressed opinion
A confidence score takes everything the model considered and crushes it into one number. Whatever reasoning existed is gone by the time it reaches you.
That creates three problems.
You cannot audit it. There is no way to check whether 85% is right, because there is nothing to check it against. The number is the whole output.
Two identical scores can mean opposite things. One candidate scores 85% because of deep, relevant experience. Another scores 85% because of keyword overlap and a well-formatted CV. The score treats them as the same candidate. They are not.
It trains you to stop reading. Scores invite ranking by number and cutting at a threshold. The strongest candidate with an unconventional CV falls below the line, and nobody ever finds out why.
Evidence works the other way around
An evidence-based answer has three parts, the answer itself, the source it came from, and a click-through to the exact line in that source.
Ask "Has this candidate managed a team of five or more?" and you get a plain answer, a citation, and a highlighted passage in the actual CV when you click. If the answer draws on other sources, a public profile or your own company data, it says so.
Nothing is compressed. The reasoning is not hidden inside a model. It is sitting on the screen, one click away.
Why this matters more, not less, in the age of AI
AI screening is only useful if you can trust it, and trust is not built by bigger numbers. It is built by verification.
Evidence makes verification cheap. You do not re-read 250 CVs to check the AI's work. You spot-check. Click five answers, see five highlighted lines, move on. When every claim is traceable to its source, a wrong answer has nowhere to hide. Click it, see that nothing supports it, discard it. With a confidence score, a wrong answer looks exactly like a right one.
This is also where compliance lives. If a rejected candidate or a regulator asks why a decision went the way it did, "the model scored them 62%" is hard to stand behind. A trail of sourced, human-verifiable answers is easy. You can show what every answer was based on. Evidence gives every screening decision a traceable record by default.
The practical payoff
Teams screening this way get through around 250 applications in about an hour. Not because they read faster, but because they stopped doing verification the slow way. Ask one question, get an answer for every applicant at once, click into anything that matters, and spend the saved hours talking to the top of the list.
The hiring manager gets sourced claims instead of scores. The recruiter gets speed without giving up judgment. The candidate gets evaluated on what their CV actually says.
Three questions to ask any screening tool
If you are evaluating AI screening software, these three questions separate evidence from scores.
- 1. Can I click any answer and see the exact line in the source? Not a summary of sources. The line itself, highlighted.
- 2. Does every answer name where it came from? CV, public profile, or your own data. If the tool cannot say, you cannot defend it.
- 3. What happens when it does not know? The right answer is "it says the CV does not mention it." A score never admits that.
If a tool fails these, its number is asking you to trust it blind. Your hiring manager will not, and neither should you.
The bottom line
Confidence scores hide reasoning. Evidence exposes it. You can defend "mentioned on line 12 of the CV, click to see it" in any room in your company. You cannot defend "85% match" in any of them.
10xTable's Screening was built on this principle, with every answer, for every applicant, clickable to the exact line in the source. See how the method works on our trust page, or try it on your next role.