If your roles have an English language requirement, Talent Insights can show an approximate CEFR level for each candidate who answers in English. It sits next to the Language Proficiency score and gives you a quick read of written English on a scale many businesses already use.
At a glance
Products: Classic and Pro.
Language: English responses only.
Where: Talent Insights reports, in the Language Proficiency section.
Set-up: off by default. We switch it on for your account when you ask.
What is CEFR?
CEFR stands for the Common European Framework of Reference for Languages. Published by the Council of Europe, it's an international standard for describing language ability across six levels:
A1 and A2: basic user
B1 and B2: independent user
C1 and C2: proficient user
A1 describes a beginner. C2 describes someone who has mastered the language.
Sapia.ai's approximate CEFR level is an estimate of a candidate's written English, based on our analysis of their interview responses. It takes our existing language analysis and expresses it on the CEFR scale, so you can compare it with the level your role asks for.
Step 1: Ask for it to be switched on
What to do
Get in touch with your Customer Success Manager, or reply to the Monthly Drop email.
We'll switch it on for your account. You'll then see approximate CEFR levels in Talent Insights for candidates who answered in English.
Step 2: Find the level in Talent Insights
What to do
Open a candidate's Talent Insights report.
Go to the Language Proficiency section. You'll see the candidate's score out of 5 and, next to it, a label such as CEFR ≈ C2*.
To read a written summary of the candidate's language, expand the Language Proficiency section.
(1) The approximate CEFR level sits next to the Language Proficiency score.
Step 3: Open the level details
What to do
Select the CEFR label to open Approximate CEFR level. You'll see:
The candidate's level and its name, for example C2 Mastery.
Where the level sits on the scale: A1/A2, B1, B2, C1 and C2.
A short description of what someone at that level can typically do.
A note explaining that the level is approximate.
The numbers match the list above.
Every CEFR level carries an asterisk and this note: "Approximate CEFR level. This level is estimated from Sapia AI's analysis of the candidate's interview responses. It is not an official CEFR certification and is not equivalent to a standardised language test."
How to make good decisions here
Use it as one signal about written English. Read it alongside the rest of the report. It isn't an official CEFR certification, it isn't equivalent to a standardised test such as IELTS, TOEFL or PTE, and it doesn't assess speaking, listening or reading. If you need an official language result, for example for compliance or visa purposes, ask the candidate to take a formal language test.
Read it as a band. In our research, the approximate level matched an expert assessor's level exactly about 6 times in 10, and was within one level almost every time. If a candidate sits close to your requirement, look at the rest of the report before you decide.
Take extra care at the lower end of the scale. The research behind our levels didn't include any candidates at A1 or A2, so our evidence covers B1 to C2. Treat an A1/A2 or B1 result as a prompt to look more closely. If your requirement is B1 or below, confirm with a formal language test before you rely on the level.
It reflects this interview only. It tells you how a candidate wrote in their answers.
It doesn't change scoring. The CEFR level comes from the same language analysis as the Language Proficiency score. Switching it on only changes what you see in Talent Insights. It doesn't change how candidates are scored, ranked or recommended.
The research behind it
Sapia.ai's science team published the method and results in a research paper, Approximating CEFR Levels with Text-based Structured Interview Responses, by Madhura Jayaratne, Ph.D. and Buddhi Jayatilleke, Ph.D.
How the levels were set. A qualified, experienced CEFR assessor read the written interview answers of 225 candidates and gave each one an overall CEFR level. The team then set cut-off points on our language fluency score to predict each level, and compared the predictions with the assessor's levels.
How closely they agreed.
Exact match: our level matched the assessor's for 139 of 225 candidates (61.8%).
Within one level: our level matched the assessor's, or was one level either side, for 219 of 225 candidates (97.3%).
Read together, those results mean the approximate level usually lands on the right level or the one next to it. That's why we describe it as approximate. For context, the paper notes that a 2004 study of an established automated essay-scoring system reported exact agreement of 50% to 58%.
Who was in the sample. All 225 candidates had applied for roles where English was the working language. The assessor placed 6 at B1, 39 at B2, 99 at C1 and 81 at C2, and none at A1 or A2. That's why our evidence covers B1 to C2.
Consistency over time. Across 6,605 interview retakes taken at least 30 days apart, the correlation between the two CEFR levels was 0.60.
Differences between groups. The team compared predicted levels across groups. There was no statistically significant difference between genders. There was a statistically significant but small difference across EEOC ethnicity and race groups (p = 0.04, Cramér's V = 0.18). The full results are in the paper.


