evidoria

← Back to browse

Good practice Imported

Ed.ai — Île-de-France Pilots AI-Assisted Essay Correction Across 20 Lycées

France · Paris · See the France profile · See the Paris profile

Evidence: Descriptive / self-reported Top 80% 33/100 · Ask Evidence Copilot about this practice

The Île-de-France region piloted the Ed.ai tool in 20 high schools, proposing AI-generated grades and remediation exercises for 7,600 exam copies across French and maths. A parallel Lyon trial found handwriting recognition still imperfect; teachers retain final grading authority.

7600 copies
Exam copies processed during the pilot (2026 (3-month pilot))
~150 teachers
Teachers involved in the pilot
~1250 students
Students covered by the pilot
~400
AI-generated evaluations produced
125
Targeted remediation activities generated
20 schools
Lycees participating in the pilot
Ed.ai — Île-de-France Pilots AI-Assisted Essay Correction Across 20 Lycées

Details

Maturity
Pilot
Promoter
Région Île-de-France
Period
June 2026 – ongoing (expansion planned 2026/2027)
Region (NUTS)
FR10
Keywords
education, assessment technology, generative AI, exam grading

Context

During a three-month pilot ending before the September 2026 return to school, the Ile-de-France regional council tested the AI grading assistant Ed.ai in 20 public lycees, where teachers uploaded scanned exam copies together with their grading rubric and the tool proposed a grade, annotations and comments for each answer before teachers reviewed and validated it.

Objectives

The pilot aimed to test whether an AI grading assistant could support teachers in French and mathematics by proposing grades, annotations and targeted remediation activities based on recurring class-wide difficulties, while leaving final grading authority with teachers.

Activities

Across the pilot, Ed.ai processed 7,600 exam copies involving roughly 150 teachers and 1,250 students in French and mathematics, generating around 400 evaluations and 125 targeted remediation activities, while a parallel, independently run experimentation at the Academie de Lyon flagged that handwriting recognition remained imperfect, particularly for atypical answer formats such as geometric diagrams and cartography.

Results

The Ile-de-France Region announced the tool would be renewed and expanded to ten subjects for the 2026-2027 school year, covering history-geography, life and earth sciences, physics-chemistry, economic and social sciences and philosophy alongside French and mathematics, but regional reporting noted that no public data yet existed on how many AI-proposed grades required teacher correction or on actual time savings.

Conclusions

The tool's real classroom impact therefore remains unverified even as its scope expands, with teachers retaining final grading authority over every AI-proposed grade.

Implementation
Indicative cost
Medium (€50k–€500k) — No public budget figure disclosed; cost implied by regional procurement of an AI grading tool across 20 schools with planned expansion to ten subjects.
Time to results
Short (< 1 year) — Three-month pilot ending before September 2026; expansion to ten subjects planned for the 2026-2027 school year.
Staffing & skills
Teachers uploading scanned copies and validating AI-proposed grades, Ile-de-France regional council administering the pilot

Conditions for success

  • Teachers retain final grading authority over every AI-proposed grade
  • Region publicly disclosed a known technical limitation (handwriting recognition) rather than concealing it

Common failure modes

  • Handwriting recognition remains imperfect for atypical answer formats such as diagrams and cartography (flagged in the parallel Lyon trial)
  • No public data yet on correction rates or actual time savings, leaving classroom impact unverified even as scope expands

Where it fits

Governance type
regional government-led public pilot
Scale
20 lycees, single French region
Income level
high-income

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful