AI can draft a lesson, but the teacher must still make the decisions

AI is useful when it removes repetitive preparation, suggests examples or helps a teacher organise a first draft. It is not a curriculum authority and it does not know the learners sitting in a particular classroom.

Before using a generated lesson, check five things: the learning objective, factual accuracy, class difficulty, available materials and the time required. A polished table can still contain the wrong fact, an activity that assumes internet access or too much work for one period.

Localisation also needs judgement. A Cameroon name does not automatically make an example accurate. Verify history, geography and cultural claims, use places naturally and avoid stereotypes. Universal knowledge should remain universal when a local reference adds no learning value.

Never place confidential learner information into a general AI prompt. Describe the support need without a child’s full name, address, medical details or identifying family information. Teachers should also avoid using AI to diagnose a learner.

The strongest workflow is draft, check, adapt and teach. AI saves time at the draft stage; the teacher protects quality throughout the rest of the process.

A strong prompt gives the AI boundaries it can use. State the Cameroon subsystem, level, class, subject, exact topic, period length, available materials and the form of output required. Describe the desired learning evidence and request low-resource alternatives where necessary. Avoid asking for a perfect lesson without context. Even a detailed prompt does not guarantee accuracy, but it reduces generic content and makes weaknesses easier for the teacher to identify during review.

Fact-checking should follow risk. A simple set of addition questions needs arithmetic verification; a history lesson needs dates, names, sequence and source review; a Science activity needs both factual and safety checks. Generated quotations, policies and curriculum references are especially likely to sound convincing when wrong. Use official ministry material, reputable textbooks and reliable subject sources. If a claim cannot be checked before the lesson, remove it or present it clearly as a question for investigation rather than a fact.

Teachers should also examine bias and representation. Generated examples may repeatedly assign leadership to men, care roles to women, urban life to wealthy families or disability to helpless characters. A quick review can vary names, regions, occupations and learner roles without forcing diversity into every sentence. Cultural detail should be specific and verified. The aim is not to make every resource identical; it is to avoid teaching narrow assumptions through repeated patterns that have nothing to do with the objective.

AI-generated assessment requires particular caution. Check that each question can be answered from taught content, that the language does not make the task harder than intended and that marks match the work required. Solve every numerical item and review the marking guide for alternative valid answers. A generated answer key should never overrule a learner who gives a correct explanation the system did not predict. Professional judgement remains necessary during both design and marking.

Schools can create a simple responsible-use policy before adoption spreads informally. Define approved tools, prohibited personal data, review responsibilities, copyright expectations and how generated material is labelled. Encourage teachers to share useful prompts and corrected resources so quality improves collectively. AI should reduce repetitive drafting and widen possibilities while leaving curriculum interpretation, relationships, safeguarding and final decisions with educators who understand the learners and are accountable for what happens in the classroom.

The teacher should keep the final corrected version rather than the unreviewed generation. Useful corrections can become part of a school prompt guide and resource bank. Over time, this reduces repeated errors and helps colleagues understand the standards expected from AI-assisted work. Responsible use is not measured by how often a tool is used. It is measured by whether the resulting lesson is accurate, safe, inclusive, teachable and genuinely helpful to the learners in front of the teacher.