Make AI argue against your lesson plan before you teach it

Make AI argue against your lesson plan before you teach it

Feed a lesson plan to any AI on a Monday morning and ask for feedback. What comes back is a compliment sandwich, three "you might consider" suggestions, and a closing line about how engaging your do-now sounds. Useless. The model has been trained to be pleasant, and pleasant is the opposite of what a lesson plan needs at 8:15am. What you need is the colleague on your corridor who has been teaching your subject for twenty years, who has walked in during your Period 3 exactly once, and who will now tell you the one thing that is going to go wrong. This prompt asks the model to be that colleague. Not to improve the lesson. To attack it.

The prompt

You are running a pre-mortem on my lesson plan for tomorrow. Your job is NOT to improve it, praise it, or tell me what is good about it. Your job is to argue, as convincingly as you can, that this lesson will FAIL to achieve the learning intention it claims. Steelman the failure case.

You are playing the role of an experienced [SUBJECT] head of learning area in an Australian [PRIMARY / SECONDARY / VCE] school who has taught this year level for at least ten years, has coached three colleagues out of exactly this lesson going wrong, and would rather be blunt with me on Sunday than diplomatic on Wednesday when the class arrives at the next lesson still not knowing the concept.

You are working in the tradition of Rob Coe on the difference between engagement and learning, Dylan Wiliam on the fact that no plan survives contact with a real class without evidence, and the honest teacher's rule that a lesson plan that "feels right" is the one that most often fails, because feel is not a diagnostic.

DO NOT hedge. Do not open with "this is a solid plan, however". Do not soften with "you might consider". Do not use the words "great", "engaging", "creative" or "innovative" about my plan at any point. If the plan is genuinely fine, say "no material failure predicted" and stop, but assume by default that any plan a busy teacher wrote in fifteen minutes has at least three points of failure worth naming.

CONTEXT I WILL PROVIDE

- Year level and subject: [E.G. Year 9 English, or VCE Unit 3 Biology]
- The single learning intention (LI): [ONE SENTENCE. WHAT STUDENTS SHOULD BE ABLE TO DO BY THE END OF THE LESSON.]
- Success criteria the LI will be judged against: [2 TO 4 CRITERIA IN STUDENT-FACING LANGUAGE]
- The class as you would describe it to a coach: [3 TO 5 LINES. INCLUDE THE 2 OR 3 STUDENTS MOST LIKELY TO STRUGGLE AND WHY, AND THE 1 OR 2 WHO WILL COAST IF NOT STRETCHED.]
- The lesson plan itself, in whatever form I have it: [PASTE THE PLAN. DOT POINTS FINE, SLIDE-BY-SLIDE FINE, A PARAGRAPH FINE. INCLUDE TIMINGS IF I HAVE THEM.]
- Anything I already suspect is weak but wrote in anyway because I ran out of time: [OPTIONAL. IF I LEAVE THIS BLANK, DO NOT ASSUME I KNOW.]

RETURN, IN THIS EXACT ORDER

1. THE LI GAP
One paragraph, three to five sentences. Does this plan actually teach what its LI claims, or does it teach an adjacent thing that will feel like the LI in the room but not evidence it on the exit ticket? Name the specific mismatch. If the plan drifts into "cover the content" or "engage the class" without producing evidence a student can now do the LI, say so plainly.

2. THE COGNITIVE-LOAD RISK
Two paragraphs. In paragraph one, name the ONE moment in the plan a novice in this year level will be overloaded (too many new terms at once, unlabelled prior knowledge assumed, a stimulus that carries more inference than the instruction budgets for). In paragraph two, name the ONE moment a student who has already met this concept will disengage because the plan is not stretching them. Give me the minute-mark for each, based on the timings I gave you.

3. THE STUDENT WHO WILL SILENTLY DISENGAGE
Pick one of the students I named in the cohort context, or, if I named none, invent a plausible archetype and say so. Write two to three sentences from their point of view (not from mine) describing the exact minute of this lesson they check out, what they will appear to be doing while checked out, and why the plan as written will not surface their disengagement to me before the bell.

4. THE THREE MOMENTS OF INVISIBLE LEARNING
List three moments in the plan where the class could nod along, look busy, or produce output that looks correct, but where I have NO diagnostic evidence that the LI has actually landed. For each moment, name (a) the moment, (b) why it will feel like learning is happening, (c) the one small check I could insert (a mini-whiteboard show, a targeted cold call, a 30-second exit sentence) to convert the moment from feel to evidence.

5. THE ONE THING TO CHANGE BEFORE CLASS
One sentence. The single highest-leverage change to this plan, given everything above. Not a redesign. Not five changes. The one move that, if I make it in the twenty minutes I have before the bell, will most reduce the chance that the LI is not met. If the honest answer is "reteach yesterday's prerequisite first, this LI is not reachable in one lesson from where the class is", say that.

6. THE ONE THING THE PLAN GETS RIGHT (STRICTLY OPTIONAL)
Only include this section if there is a specific choice in the plan that a coach would genuinely commend. If yes, one sentence, naming the choice and why it matters. If no, omit the section entirely. Do NOT manufacture a compliment to soften sections 1 to 5. The absence of this section is information; it tells me the plan is closer to broken than to fine.

RULES

- Australian English throughout. Realise, colour, behaviour, analyse. Year 9, not 9th grade.
- No em-dashes and no en-dashes anywhere. Commas, full stops, parentheses, semicolons only. Regular hyphens in compound words are fine.
- No sycophancy. No "however", no "that said", no "on balance". If a section requires "and yet", write two sentences instead.
- No emoji. No exclamation marks. This is a coaching document, not a chat.
- No generic pedagogy advice ("use more formative assessment"). Every criticism must anchor to a specific line, moment or timing in the plan I pasted. If a criticism could apply to any lesson plan on any topic, cut it.
- No corporate hedge words: "opportunity to consider", "you may wish to reflect on", "there is scope for". If the failure mode is real, name it.

If any of the following are true, STOP and tell me before writing:
- I have not given you the LI, or the LI is genuinely unclear (e.g. "explore fractions", "engage with the text"). Ask me to sharpen the LI to a single testable proposition before you attack a plan whose target is undefined.
- The lesson plan I pasted is a slide deck without instructions, so you cannot see what I intend to say or do in the room. Ask for the teaching moves, not just the slides.
- The plan is for a summative assessment (SAC, NAPLAN practice under conditions, a graded piece of work). Tell me a pre-mortem on the design belongs with the head of learning area and a moderation partner, not here.

What it does well

The role instruction and the ban on hedge language are what stop the output being the compliment sandwich. Asking for the specific minute a student will disengage forces the model past "some students may struggle" into "the third student on the far side of the room will check out at minute 22 when the slide of six technical terms goes up". The "one thing to change" line is the piece that earns its keep on a Monday morning, because it is what fits into the twenty minutes you actually have before the bell.

Where it falls over

Two honest limits. First, the model has never watched your class, and its "student who will disengage" archetype is drawn from the cohort context you paste in. If you type "mixed ability", you will get a generic archetype back. Feed it the two or three students you already worry about, by first initial and one sentence each. Second, an adversarial model with no compliment section can occasionally read as too harsh, especially on a lesson plan you are quietly proud of. Read it once, take a breath, then decide which of its criticisms are right. The plan is not the model's opinion. The lesson is.

Try it like this

For a Year 9 English lesson introducing symbolism in a set text: paste the LI ("students can identify one symbol in the chapter and explain what it represents"), the plan (do-now, mini-lecture, group annotation, share-out), and the two students you know will drift on group work. The pre-mortem will name the exact minute of the group-annotation phase where the two of them stop reading and start off-task, and it will name the check to insert (a two-minute individual write before the group opens) that keeps them accountable without changing the lesson shape.

For a VCE Unit 3 Chemistry lesson on Le Chatelier's principle: paste the LI, the study-design outcome, and the sequence of worked examples you plan to run. The pre-mortem will flag whether your first example is genuinely a stress test of the principle or a rehearsal of the previous lesson's equilibrium content, and it will name the minute-mark at which your strongest students will already have generalised to the wrong rule (rate versus position of equilibrium is the classic).

For a Year 5 numeracy lesson on equivalent fractions: paste the plan and the two students on adjustments. The pre-mortem will call out any moment where the concrete-representational-abstract sequence has been collapsed for time, and it will predict the exact question at which the class will split into three groups, so you know where to put your circulation focus.

If you only have five minutes: paste the LI and the plan, skip the cohort context, and read only section 1 (the LI gap) and section 5 (the one thing to change). That is the smallest defensible version of the pre-mortem, and on a real Monday morning it is often all you need.

Run the pre-mortem before the coffee, not after the lesson. A plan you attack yourself is a plan the class will not have to attack for you.