AI in the classroom is no longer a Friday curiosity. By the spring of 2026, most teachers in most schools have access to a chat model and a brief from somewhere upstairs to do something with it. The question is no longer whether to use AI. The question is what use looks like when it works. When I place the AI lessons I see on the PICRAT grid, the answer is sobering: a lot of them sit in the bottom-left corner. Passive Replace. AI does the work, students consume the output. The lesson would teach roughly the same thing without any of it.
That is not because the teachers are wrong about AI. It is because the path of least resistance with a chat model is to ask it for a paragraph and put the paragraph on a slide. There is one move that breaks the habit. It is small, it is repeatable and it works on almost any AI lesson you have already planned.
Why AI lessons so often default to PR
A teacher with thirty minutes on a Tuesday evening and an AI tab open will reach for what AI is best at: producing fluent text quickly. Generate a summary. Generate an explanation. Generate a set of worked examples. The output gets pasted into the lesson and the lesson is built around that output. Students arrive. Students read. Students copy it down. Done.
What just happened is that the AI did the cognitive work and the student watched. The technology has not changed what the lesson asks of the student; it has just changed who produced the text on the slide. On the activity axis of PICRAT, that is Passive. On the technology axis, the AI only carried the text: a textbook paragraph or a teacher-written summary would have asked the same of students. Replace. PR. The AI lesson I see most often right now is the one where the AI is doing what the teacher used to do, and the student is doing exactly what they always did.
It is worth saying that the alternative is not to ask AI for a better paragraph. The alternative is to ask a different question of the AI. Tools like Generate exist for exactly this reason: pick a subject and an age range and you get a 3x3 matrix of nine lesson ideas, one per PICRAT cell, with the cognitive load on the student instead of on the screen. That is a different kind of help than "write me a slide."
Three AI lessons in PR
Three patterns that appear most often when I look at lessons in our schools. Each one is well-meaning. Each one is PR. Same shape every time.
Grade 7 (Year 8) history, the Industrial Revolution. The teacher pastes a long primary-source extract into ChatGPT and asks for a five-bullet summary. The summary goes on a slide. Students copy it into their books.
Grade 8 (Year 9) biology, mitosis. The teacher prompts an AI for a one-paragraph explanation of the four stages. The paragraph goes on a slide. The class reads it together. Students copy the four stages into their books.
Grade 5 (Year 6) maths, fractions. The teacher uses an AI to spit out ten worked examples of adding fractions with unlike denominators. The examples go on slides. The class reads through them one by one.
The student in each of these lessons read or copied what the AI wrote, and went home. The AI was the source of the lesson's content. The student was the audience.
The one move
The shift sounds small in a sentence and is large in a classroom.
Stop using AI to give students answers. Use AI to give students something to argue with.
The AI becomes a draft, not the answer. Generate the summary, then ask the students to find what is wrong with it. Generate the explanation, then ask the students to write a better one. Generate the worked examples, then ask the students to spot the one with a mistake in it. The student's task moves from read what AI made to fix, judge, sharpen or argue with what AI made. That is no longer Passive. That is Interactive.
The pedagogy under it is older than AI. Students learn more by critiquing a model answer than by reading one. They learn more by spotting an error than by repeating a correct fact. AI is unusually good at producing the right shape of artefact for this kind of work, at speed, in volume, and tuned to the year group. The teacher who treats the AI output as the final word writes a PR lesson. The teacher who treats it as a draft writes an IA lesson.
The same three lessons, reworked
Here are the same three lessons rewritten with the move applied. Same time budget, same AI tool, same content domain. The only thing that changes is what the student does with the AI output.
Grade 7 (Year 8) history. The AI summary still goes on the slide, but with the original primary source on the desk in front of every pair. The pair's job: highlight one piece of evidence from the source that the AI summary missed and one piece it got slightly wrong, and rewrite the bullet that needs work. Five minutes. They post their rewritten bullet to the class Padlet, where every pair's fix sits side by side.
Grade 8 (Year 9) biology. The teacher generates two AI explanations of mitosis, one accurate and one with a deliberate error introduced by tightening the prompt. Pairs get both in a shared document, are told one is wrong, and have to find the error, justify their choice in a comment, and write a corrected sentence. The closing slide shows the original prompts so students see what produced each one.
Grade 5 (Year 6) maths. Each table gets the AI-generated worked examples in a shared document, but with a brief: one of the ten has a mistake in it, two skip a step and one could be done a quicker way. Tables work in pairs to flag which is which in the comments and explain why, then swap with another table to check each other's flags.
None of these lessons takes longer to plan. The AI did most of the heavy lifting. The work the teacher does is the same work she would have done anyway: pick a topic, choose a question. The shift is in what the student is asked to do with what the AI produced.
A note on Creative Transform
You can write Creative Transform AI lessons. A Grade 9 (Year 10) group designing a chatbot that explains a Shakespeare scene to a Grade 6 (Year 7) audience, iterating on its prompts, and presenting it to the younger class is firmly in CT. So is a Grade 11 (Year 12) economics group using AI to model alternative budget assumptions and writing a position paper that defends one. Both are good lessons. Both take longer, manage harder, and go wrong more often. CT is worth the cost when the output of the lesson is genuinely a new artefact and the AI is one of several instruments the student is wielding. For most weeks, in most classrooms, the useful nudge is the one above: from PR into IA. Stay there until the move feels routine, then decide whether a unit has a lesson that asks for CT.
What to do tomorrow
Take the AI lesson you most recently taught or are about to teach. Write down, in one sentence, what your students did with the AI output. If the sentence is "they read it" or "they copied it" or "they answered questions about it in their books", you are in PR. The fix is to add a sentence that begins with a verb of judgement: criticise, fix, rank, improve, defend, refute, rewrite. Whatever the next lesson is, the AI part of it earns its keep when the student is the editor of what the AI produced, not the audience for it.
If you want to test where a draft sits, drop it into Analyse. It walks you through a guided decision tree about your lesson and helps you place it on the grid yourself. The first time most teachers do this with an AI lesson they thought was Interactive, the answer surprises them. That surprise is the point.
Andy Perryer is Director of Education Innovation at a group of international schools and the creator of PICRAT Suite. The PICRAT framework was developed by Royce Kimmons, Charles Graham and Richard West in their 2020 paper in the CITE Journal.