Screening basics
How AI interview scoring works.
From transcript to skill scores to a place in your review list, with a worked example and an honest look at where scoring goes wrong.
7 min read · Updated September 24, 2026 · The LessRounds team
An AI interview score is only as useful as your understanding of where it came from. If you know what the model read, what it compared the answers against, and how the pieces add up, you can tell a trustworthy score from a shaky one in seconds. This guide walks through each step, using LessRounds’s rules as the worked example because those are the ones we can state exactly.
Step one: the transcript
Scoring starts with words. The interview is recorded and transcribed, and the transcript is what gets assessed: what the candidate said, in answer to what question.
That boundary matters. A well-designed system does not score faces, expressions, clothing, rooms, voices or emotions. Video, if you use it, is there so your team can watch it. It isn’t an input to the score.
It also means language and grammar should never count against anyone by accident. The scoring we describe here is instructed never to mark a candidate down for which language they use, for mixing languages, for script or for grammar. If clear English genuinely matters for the job, add it as a skill so it’s scored openly, where you can see it.
Answers to screening form questions, such as yes/no, multiple choice and short text, are shown to your team but aren’t scored. They never reject anyone on their own.
Skills and expected levels
You tell the system what to look for as a short list of skills, each with the level you expect for this role: beginner, intermediate, advanced or expert.
The level is what keeps scoring fair across roles. Take two candidates who both say, “I’d check the order history, apologize, and offer a replacement.” For a first support job with “Handling complaints” set to beginner, that’s a solid answer. For a support team lead with the same skill set to advanced, the same words are thin, because you’d expect them to mention patterns, policy exceptions or coaching the team.
Each skill gets a score from 0 to 100 against that level, along with:
- Reasoning: a few sentences on why the score is what it is.
- A quote: the line from the transcript the score rests on most.
Weights and the overall score
Not every skill matters equally, so each one has a relative weight. The overall score is the weighted average of the skill scores.
A worked example for a customer support role:
| Skill | Level | Weight | Score | Contribution |
|---|---|---|---|---|
| Handling complaints | Intermediate | 40 | 78 | 31.2 |
| Clear explanation | Intermediate | 30 | 64 | 19.2 |
| Troubleshooting | Beginner | 30 | 55 | 16.5 |
| Overall | about 67 |
Weights are relative, so 40, 30 and 30 behave exactly like 4, 3 and 3. Two practical points follow. First, a skill with a small weight barely moves the total, so if something is a true must-have, give it real weight. Second, look past the total. The candidate above is strong on complaints and weaker on troubleshooting, which the overall number hides and the skill rows show.
Evidence: highlights and conduct flags
Beyond skill scores, a useful write-up points you to the moments worth your time.
- Highlights: three to five timestamped moments, each marked as a strength, a concern or an insight, so you can jump to them in the recording.
- Conduct flags: answers that were evasive, generic, very short, scripted or off topic. A flag isn’t a verdict. A “very short” flag on a nervous graduate is a reason to listen, not to reject.
Some behavior ends an interview. After two conduct warnings, for example for abuse or repeatedly going off topic, the interview stops and is set aside, with the reason shown on the candidate’s page.
Pass marks and what happens next
A pass mark is the line that sorts completed interviews. The rules we use are simple, and worth knowing exactly:
| With a pass mark | Without one | |
|---|---|---|
| At or above the line | To review | 50 and above: to review |
| Below the line | Set aside in Rejected | Under 50: set aside in Rejected |
| Is the candidate told? | No | No |
| Can your team move them? | Yes, any time | Yes, any time |
With a pass mark of 60, the candidate in the example above, at about 67, goes to review. Nobody is emailed about the outcome in either direction, and candidates never see their scores. Anyone set aside can be moved back to review, shortlisted or put on hold.
Choosing the number is its own decision. Setting a pass mark covers how to pick one and when to move it.
Where scoring goes wrong
Every scoring system has limits. These are the ones to watch:
- Transcription errors. Background noise, a weak microphone or a patchy connection can drop words, and a missing word can lower a score. If a quote looks garbled, watch that part of the recording.
- Short answers. A strong candidate who says little gives the model little to work with. Follow-up questions help, but a very short interview is thin evidence either way.
- Skills that don’t fit the questions. If no question gives a candidate a chance to show a skill, its score will be low for everyone. Check that each skill has at least one question behind it.
- The wrong level. Setting “expert” for a junior role will push good candidates below your pass mark.
- Reading the total only. The overall number compresses the story. The skill rows and quotes are where the judgment is.
- A quote that doesn’t support the score. Occasionally the reasoning and the evidence won’t line up. Treat that as a signal to read the transcript yourself, and trust what you find there over the number.
The habit that catches most of these: for your first few interviews, read a sample of candidates who were set aside. If good people are landing there, lower the pass mark or adjust the skills and levels. The scoring page shows what a full write-up looks like.