In a public administration, a university, a language centre or a state-funded training programme, language assessment drives decisions: placing learners, grouping them, tracking progress, justifying funding. When AI enters that process, the right question is not "is AI reliable?" in general, but: which type of assessment do you need, and what guarantees should you require from the vendor?
This guide gives programme managers and decision-makers a checklist for choosing an AI language proficiency testing tool. It is not legal advice: for a specific case, consult your legal department or your privacy officer.
Which Type of Assessment Do You Need?
The same word, "test", covers different needs. Before comparing tools, pin down yours:
| Type of assessment | What it is for | Typical moment |
|---|---|---|
| Placement | Putting each learner in the right group or course | Start of training |
| Diagnostic | Identifying strengths and weaknesses, skill by skill | Before a training plan |
| Progress | Measuring how a learner or a group evolves | At regular intervals |
| Proficiency | Establishing the overall level, whatever course was taken | Hiring, mobility |
| Certification | Issuing a recognized official result | A process that requires an official exam |
An AI assessment such as CEFRHub's serves placement, diagnostic and progress. It does not issue a certification: that remains the role of official exams (IELTS, TOEFL, Cambridge English, ACTFL assessments, DELF, DALF, TEF…).
What the CEFR Requires from a Tool
A level is justified by descriptors
A tool that returns a level between A1 and C2 must be able to say what it relies on: the descriptors of the CEFR and its Companion Volume (Council of Europe, 2020), which describe what a person can do at each level, including in mediation (for example, the descriptors for mediating a text, from B2 to C1). A level with no link to those descriptors borrows the CEFR's vocabulary without following its method. For an overview of the levels, see our guide to CEFR language levels.
Assessing is not certifying
An assessment appraises skills within a learning programme (placement, monitoring, review). A certification issues an official credential, following the procedures of an examination body. An AI assessment tool can produce a documented, reproducible assessment; it does not replace an official exam when a process requires one.
The Questions to Ask the Vendor
- Which type of assessment does the tool provide: placement, diagnostic, progress?
- How does the level returned relate to the descriptors of the CEFR and its Companion Volume?
- Which skills are assessed: written production, spoken production, comprehension?
- What validation was carried out, on which productions, with which assessors? Ask for the documentation.
- Is the result a decision aid or an automated decision? Who keeps the decision?
- Where is the data stored and processed, and how long is it kept?
- How does the tool handle accents, language varieties and diverse learner profiles?
- What limitations does the vendor publish?
A missing documented answer to any of these questions is a warning sign.
What the Texts Say
Here are the main texts an institution consults, cited with their source (this is not legal advice):
- European Union — AI Act. Regulation (EU) 2024/1689 entered into force on 1 August 2024 and applies in stages (EUR-Lex). How a tool is classified depends on its use; CEFRHub's position is set out on its AI Transparency page.
- European Union — GDPR, Article 22. "The data subject shall have the right not to be subject to a decision based solely on automated processing, including profiling, which produces legal effects concerning him or her or similarly significantly affects him or her" (EUR-Lex).
- Canada — federal laws. "The Privacy Act only applies to federal government institutions listed in the Privacy Act Schedule of Institutions"; PIPEDA covers commercial activities, and "Municipalities, universities, schools, and hospitals are generally covered by provincial laws" (Office of the Privacy Commissioner of Canada).
- Quebec — public bodies. The Access Act (A-2.1), amended by Law 25, requires a privacy impact assessment "for any project to acquire, develop or overhaul an information system" involving personal information (section 63.5), and requires a public body that renders "a decision based exclusively on an automated processing" of personal information to inform the person (section 65.2) (LegisQuébec).
- International standards. ISO presents ISO/IEC 42001 as "the world's first AI management system standard". The OECD AI Principles, adopted in 2019 and updated in May 2024, have 47 adherents.
- Canada's federal public service. Second-language proficiency is rated at levels A, B and C of the Treasury Board standard; the choice of assessment methods rests with the Public Service Commission. A tool such as CEFRHub does not replace it.
The Hybrid Model
AI does not replace the assessor; it saves time where the assessor loses the most. A common institutional setup:
- Initial placement by AI, for all learners.
- Targeted human review: borderline cases, levels C1 and C2, high-stakes decisions.
- Regular progress tracking, with the same rubric from one assessment to the next.
- An official exam when a process requires a recognized result.
Our comparison AI vs manual assessment details this division of roles, and our methodology describes what CEFRHub analyses.
CEFRHub for Institutions
- Written and spoken production, in English and French, from real productions.
- CEFR level from A1 to C2, with sub-levels, and a report based on the descriptors of the CEFR and its Companion Volume, in under 3 minutes.
- Indicative results: generated automatically by AI, they are intended to support, not replace, the judgment of a teacher or assessor; the decision stays with the institution.
- Data stored and analyzed in Canada.
- Educational assessment tool, not a certifying test.
- Organization plans: Team (CA$169/month) and Business (CA$499/month), Enterprise on quote; free trial, no credit card. See pricing.
Frequently Asked Questions
What are the types of language tests?
Assessments are usually grouped into placement (putting learners in a group), diagnostic (identifying strengths and weaknesses), progress (measuring change), proficiency (establishing the overall level) and certification (issuing an official result). An AI assessment such as CEFRHub's serves the first three.
Can AI assessment replace an official exam?
No. When a process requires an official result, you need an official exam (IELTS, TOEFL, Cambridge English, DELF, DALF, TEF…). AI assessment serves placement, diagnostic and monitoring.
Can a Quebec public body use an AI assessment tool?
The Access Act (A-2.1) requires it to conduct a privacy impact assessment before acquiring an information system that handles personal information (section 63.5), and to inform the person of a decision based exclusively on automated processing (section 65.2) (LegisQuébec). This is not legal advice.
Does the AI Act apply to language assessment tools?
Regulation (EU) 2024/1689 entered into force on 1 August 2024 and applies in stages; how a tool is classified depends on its use. CEFRHub's position is set out on its AI Transparency page.
Are Canadian universities covered by PIPEDA?
Generally not: the Office of the Privacy Commissioner states that municipalities, universities, schools and hospitals are generally covered by provincial laws (Office of the Privacy Commissioner).
Does CEFRHub replace the federal public service second language evaluation?
No. The choice of methods to assess levels A, B and C rests with the Public Service Commission; the Treasury Board standard gives no correspondence with the CEFR.
Does AI replace the assessor?
No. Results are generated automatically by AI; they are indicative and intended to support, not replace, the judgment of a teacher or assessor. The decision stays with the institution.
Does CEFRHub use the ACTFL scale?
No: CEFRHub places each production on the CEFR scale, from A1 to C2. For the correspondence between ACTFL and the CEFR, see our page on ACTFL and the CEFR.