AI in a Language School: How to Run a 30-Day Pilot Without Losing Pedagogical Control

Adopting artificial intelligence at a school does not start with choosing a tool. It starts by identifying a recurring learning or operational problem and deciding how to tell whether a solution improved it without creating greater risks.

That distinction matters. UNESCO’s Guidance for generative AI in education, updated in January 2026, advocates a human-centred approach that protects data, validates pedagogical use, and preserves human agency. The British Council applies the same logic to language teaching: technology should support pedagogical goals, not set them.

For a small or midsize school, this does not require an AI committee or a long transformation project. A 30-day pilot with a narrow scope and criteria set before the test can help distinguish real value from novelty.

Choose a problem, not a technology showcase

Start with a frequent, bounded, low-risk task. For example: draft several versions of a customer-service dialogue for B1 learners using an objective and vocabulary already approved by the teacher. AI produces a draft; the teacher checks accuracy, level, and fit before using it.

Avoid starting with high-impact decisions such as passing or failing students, assigning final grades, or replacing a teacher’s assessment. Automatic content generation is not evidence of learning either: a well-written activity still has to serve the course objective.

Better first use case	Why it is more controllable	What still needs human review
Draft dialogues and questions from a lesson plan with no identifiable learner data	The task has clear context and observable output	Language accuracy, level, inclusion, and alignment with the objective
Adapt an approved text for different proficiency levels	The source material and constraints are known	Whether the adaptation preserves meaning without oversimplifying
Suggest formative feedback on an anonymised response	It can support a teacher without deciding the outcome	Tone, accuracy, usefulness, and any student assessment
Automatically assign final grades or progression	High impact and difficult to challenge	Not a good scope for a first pilot

UNESCO’s teacher AI competency framework spans five dimensions, including ethics, foundations, pedagogy, and professional development. The operational point is simple: a tool does not remove the need to prepare the people who select, use, and review it.

The 30-day plan

Days 1–5: establish a baseline

Write the hypothesis in one sentence: “If we use AI to draft B1 role-plays, teachers will spend less time preparing them without lowering pedagogical quality.” This is not a claim that the outcome will happen; it defines what the test must check.

Record the current process before changing anything:

• minutes a teacher spends preparing the activity;
• how many revisions are needed before it can be used;
• the teacher’s assessment of alignment, accuracy, and level appropriateness;
• questions or problems raised by learners and teachers.

Choose a small sample the team can actually monitor. The goal is to keep the activity type and review criteria comparable, not to recruit a large number of participants.

Days 6–10: set boundaries and prepare the team

Before writing the first prompt, document four decisions:

1. Purpose: which task the tool is approved for and which uses are out of scope.
2. Data: what may be entered, what must be removed, and who can access the output.
3. Review: who checks content, level, bias, cultural appropriateness, and accuracy before use.
4. Incidents: how to stop use, record a problem, and notify the responsible person.

As a prudent default, do not enter student names, contact details, recordings, grades, or identifiable stories into public tools until the tool and its processing terms have been reviewed. UNESCO’s guidance calls for privacy protection and ethical, age-appropriate validation. For minors, schools should assess applicable law and the learner’s best interests rather than assume one universal consent rule.

Train the team to write and review prompts. A starting template:

Learning objective: [objective]. Level: [level]. Task: draft [activity] for a [context] lesson. Use only the approved content below. Do not invent facts, include personal data, or hide uncertainty. Expected output: [format].

Days 11–24: run the pilot with human review

Use the tool only for the approved use case. Keep the draft and the teacher’s edits so the school can see where AI helps and where it creates rework. Do not publish or send an unreviewed output directly to learners.

A short review rubric, scored from 1 to 5, can keep decisions consistent:

Criterion	Review question
Objective	Does the activity practise what the lesson is meant to teach?
Language	Are vocabulary, grammar, and examples accurate for the level?
Learning	Must learners think, interact, and produce, or can they just accept the AI’s answer?
Inclusion	Are there stereotypes, exclusionary references, or unnecessary cultural assumptions?
Privacy	Does the material expose data that should not have been entered?

Also record total time, including corrections. A tool may generate in seconds but require enough editing that the productivity gain is smaller than it first appears.

Days 25–30: decide using evidence

Compare the pilot with the baseline. Do not rely only on “the team liked it,” and do not present a small test as causal proof of improved learning. Separate the signals:

• Efficiency: did net preparation time fall after review?
• Quality: did materials meet the school’s rubric?
• Adoption: could teachers use the workflow without depending on one AI specialist?
• Safety: was data sent improperly, or did biased or materially inaccurate output appear?
• Experience: did teachers and learners understand when AI was used and how to request a correction?

Set minimum continuation criteria before starting. Continue if there is observable benefit, quality is maintained, and controls are workable. Adjust if the use case is useful but prompts, training, or tooling still create rework. Stop if risk is uncontrolled, pedagogical failures recur, or no measurable benefit appears.

A decision record for the next team meeting

Finish with a one-page record: use case; owner; tool and configuration; permitted data; baseline; rubric; results; incidents; decision; and review date. This turns an experiment into operational memory and prevents every teacher from improvising a different rule.

The best pilot is not the one that uses the most AI. It is the one that gives the school enough evidence to decide whether a specific task became better, safer, or more efficient. An integrated platform can help organise lessons, activities, and progress tracking; pedagogical accountability and data rules remain with the school and its educators.

Sources

• UNESCO — Guidance for generative AI in education and research
• UNESCO — AI competency framework for teachers
• British Council — Human-centred AI: lessons for English learning and assessment
• Brazilian General Data Protection Law (LGPD), Law No. 13,709/2018
• Brazil’s ANPD — Personal data processing for children and adolescents

This article is informational and does not replace legal advice for a specific data-processing activity.