Deploy at the school or district level

From pilot to school-wide project

Getting ChatGPT working in a single motivated classroom is one thing; turning it into a school-wide project, or rolling it out across a network of schools, is another. The challenge is organizational as much as technical: you must convince, train, equip, and reassure people who did not all ask to change their practices. It is also legal: as soon as personal data is involved, the project falls under the authority of the principal in secondary schools or the academic director in primary schools, working with the data protection officer (DPO) and relevant governing bodies. The roadmap unfolds in four phases, in a non-negotiable order: each phase creates the conditions for the next.

Phase 1: The pilot

Everything starts with a pilot, and its value depends on the quality of the sample. Resist the temptation to select only the best-equipped classes and the most tech-savvy teachers: you would get brilliant but non-reproducible results. Aim for a varied sample—multiple grade levels, multiple subjects, volunteer teachers—while respecting the usage framework rules: student use begins only in 8th grade (4e). A scoping note that names the selected classes and justifies each choice will save you from having to defend the sample after the fact.

Before launching anything, define what you will measure. A pilot without indicators produces an anecdote, not an experiment:

During the pilot, hold regular check-ins with participating teachers. Their feedback—such as a setting that blocks workflow or an unexpected use case that works well—will allow you to adjust before scaling up.

Phase 2: Team training

Deployments rarely fail for technical reasons; they often fail because teachers were not supported. The training plan should be progressive:

1. Session 1 (2 hours): Discovering ChatGPT and the usage framework, demonstration
2. Session 2 (2 hours): Hands-on practice, first pedagogical prompts, data settings
3. Session 3 (2 hours): Projects and reusable prompts for your subject
4. Session 4 (2 hours): Assessment and differentiation with AI
5. Follow-up: A monthly office hour to answer questions

The office hour is not an optional extra: it is what turns a one-off training session into a lasting change in practice, because real questions arise after three weeks of actual use. Also point teams toward ministry resources presented by éduscol: the CREIA community on Magistère (open for self-enrollment), the Magistère course “Using Generative AI in Education in an Informed and Responsible Way,” and the one-hour module “Discovering Generative AI” on the Mentor platform.

Identify one or two leads in each school, trained more deeply: help then comes from a peer who knows the context. And do not forget families. Inform them of what is being done, why, and with what data, as required by the CNIL (France’s data protection authority); if students use the tool, collect parental consent as required by OpenAI’s terms for minors; keep a channel open for questions. An informed parent becomes an ally; a parent who discovers AI through rumors prepares a crisis.

Phase 3: Scaling in waves

Never open everything at once. Proceed in waves: first the trained volunteers, who become ambassadors; then a targeted expansion with their support; finally, all relevant teams, once support is in place and best practices are documented.

On the infrastructure side, tasks depend on the access method chosen. For use without accounts, test every room: access can vary by network, and school filtering may block the service. For a managed workspace, plan for login with school accounts, the contract and its data addendum, registration in the processing registry, and, if necessary, an impact assessment. In all cases, verify bandwidth, name a contact for incidents, and write short documentation accessible at any time.

Phase 4: Sustainability

A successful project can die out in two years if no one ensures its governance. Have it overseen by an existing body—the pedagogical council, or a committee bringing together teachers from each level, administration, the lead, parent representatives, and the data protection officer—which arbitrates changes and handles incidents.

Finally, anticipate the recurring budget: a managed workspace is billed according to the contract, at headquarters or via credits; training must continue for new arrivals and new functions; lead time has a cost. Also reserve time for evaluation: this is what will feed the impact measurement, the subject of the next lesson.

Your turn

Write the pilot scoping note.

1. Name the selected classes and justify each choice, respecting the grade-level framework
2. Choose four indicators and their measurement methods
3. Plan the four training sessions and the monthly office hour
4. Draft the letter to families: what is being done, why, and with what data
5. Have the note validated by administration and the data protection officer

Official sources: Artificial Intelligence and Its Uses in Education (éduscol) · Data Controller: How to Implement AI Systems in Education? · Teacher: How to Use an AI System in the Course of Your Duties? (CNIL) · The ChatGPT home page · Managing members, workspace roles, and seats in ChatGPT Enterprise and Edu · Workspace analytics for ChatGPT Enterprise and Edu (OpenAI Help Center) · Europe Terms of Use (OpenAI)

Key takeaways

• A project involving personal data falls under the principal’s authority, working with the data protection officer
• The pilot must be measured: four indicators defined before starting
• Progressive training and monthly office hours drive changes in practice
• Families are informed, and their consent collected when students use the tool
• A governance body ensures the project’s longevity