A screening does not judge a candidate. It reports, line by line, whether their CV gives evidence for each requirement. No evidence in the CV means the CV does not mention it, not that the person cannot do it.
Run a screening
Screening never starts by itself, including for people who apply through the careers page. Their row says Not screened yet until someone runs it.
- Open the role. Go to the Roles tab and open the role to see its candidates.
- Press Screen on the candidate's row. While it runs, the row says “Reading the CV against your requirements…”.
- Read the result. What the AI found opens under the row, and the row now shows Match with a percentage. Later, use Show what the AI found to open it again.
You can screen with HR admin or Recruiter access, or as the Hiring manager named on that role. Someone with Interviewer access cannot run a screening. See access for what each one covers.
What the AI reports
The AI reads two things: the role's requirements and the CV text. It returns one answer per requirement, and the page shows each line like this:
| On each line | What it means |
|---|---|
| The requirement and its weight | Your own line from the role, with Required, Preferred or Nice to have. |
| Evidence found | The CV supports this. The words the AI relied on are quoted underneath. |
| No evidence in the CV | The AI found nothing that shows it, and the line reads “Nothing in the CV speaks to this.” |
The AI works to these instructions:
- Judge only what the CV says. It is told not to infer ability or fit from a name, age, gender, nationality, school or how well known an employer is. The name is still in the CV it reads, so this is an instruction and not anonymization.
- A requirement is a minimum. Eight years meets “5+ years”, and native Japanese meets “business-level Japanese”.
- Quote, do not paraphrase. Evidence should be the CV's own words, and it appears only on a match.
- Say nothing rather than speculate. With no evidence, the answer is No evidence in the CV and no quote.
- Do not recommend. It is told not to rank, score or suggest an outcome.
Treat a quote as a pointer and not as proof. AB Talent does not show the full CV text on any screen today, so these quotes are the part of it you will see, and an interview is the place to confirm them.
How the Match score is calculated
The AI does not produce the number. It answers yes or no per requirement, and AB Talent turns those answers into a percentage with fixed arithmetic. The same answers always give the same score.
| Weight | Points |
|---|---|
| Required | 3 |
| Preferred | 2 |
| Nice to have | 1 |
Match = points on lines with evidence ÷ points on all lines × 100, rounded to the nearest whole number.
Example
A role has four requirements: two Required, one Preferred, one Nice to have. That is 3 + 3 + 2 + 1 = 9 points.
The CV gives evidence for everything except one Required line, so it earns 6 of 9 points. 6 ÷ 9 = 66.7%, shown as Match 67%.
The percentage is only that arithmetic. It is not a probability, and the candidate list is not re-ordered by it. Always read the lines behind it: a sparse CV from a strong person scores low, and two people on the same score can have met different requirements.
Screen again
Once a candidate has a score, the button becomes Screen again. It runs a completely new screening of the same CV against the role's requirements. It does not average with the last one.
- The new result replaces what the row and the page show, and its heading reads Screening 2, then Screening 3, and so on.
- Earlier screenings stay on record, but the page shows only the newest.
- Because an AI reads the text each time, a new run may not read every line the same way.
- Each screening that completes is counted.
Beside the heading, the page also names the AI model and prompt version behind that screening, so a decision questioned later can be traced to what produced it.
When a screening cannot run
A screening either completes or records nothing at all. If anything goes wrong, you never get a low score that looks like a judgement about a person. The AI's reply is thrown away if it fails, cannot be read, leaves out a requirement, or marks a requirement as not met while quoting evidence for it.
| Message on the row | What happened | What to do |
|---|---|---|
| “Couldn't screen this application, so nothing was recorded. The CV and the requirements are unchanged.” | The AI failed, or its reply was discarded. No score was written, any earlier score stays, and the attempt is not counted. | Press the button again. If it keeps happening, see troubleshooting. |
| “Add the CV text to this application before screening it.” | The candidate was added with no CV, so there is nothing to read. | A CV cannot be added afterwards. Add the person again with New candidate and paste the CV. |
Decide
When you have read the lines, choose Advance or Reject and fill in Why?. Write your own reasoning there, because the reason is yours and not the AI's. Roles and the careers page explains the pipeline, and scorecards rate the same requirements after an interview, so the panel's view can sit beside the AI's.
The AI shows you what a CV says about each requirement, and every decision that follows is made by a person.