{"course":{"slug":"ai-for-teachers","title":"AI for Teachers","url":"https://addaly.com/learn/ai-for-teachers","level":"beginner","track":"work"},"lesson":{"slug":"the-second-message-matters-more","title":"The second message matters more than the first","url":"https://addaly.com/learn/ai-for-teachers/the-second-message-matters-more","position":11,"of":57,"minutes":8,"answer":"Treat the first output as a draft and spend your effort on the critique (specific, pointed at one thing, with the reason) because a model steered by feedback improves fast.","keyIdea":"Treat the first output as a draft and spend your effort on the critique (specific, pointed at one thing, with the reason) because a model steered by feedback improves fast, while a first prompt rewritten five times mostly does not."},"body":{"format":"markdown","text":"## Where teachers spend their effort, and where it pays\n\nMost people meeting these tools put all their effort into the first prompt. They write it, read the output, feel disappointed, and rewrite the prompt from scratch. Three rounds of that and they conclude the tool is not much use.\n\nThe effort is in the wrong place. The first output is almost always a draft. The fastest route from draft to usable is not a better first prompt; it is a pointed second message. A model handed a specific critique of its own output improves markedly, because the critique tells it exactly which part of the space to move to. A rewritten first prompt starts the guess again from nothing.\n\n## What a useful second message looks like\n\nThree parts: what is wrong, where, and why.\n\n> Figure: Two ways to spend ten messages. Line chart of % of the output you would use as it is against Messages spent on one task: Pointed critique of the draft (1, 35) (2, 62) (3, 78) (4, 85) (6, 88) (8, 84) (10, 79); First prompt rewritten from scratch (1, 35) (2, 38) (3, 44) (4, 46) (6, 49) (8, 50) (10, 50). Critique moves the output because it names which part of the space to move to. A rewritten first prompt starts the guess again from nothing. Note the turn after eight messages: the conversation is long, early constraints are drifting out of the window, and the right move is a fresh chat with the best version pasted in as the example.\n\nWeak: \"Make it better.\"\n\nWeak: \"These are too easy.\"\n\nUseful: \"Questions 1 to 6 can all be answered by copying a phrase from the passage. I want questions where the answer is not in the text and the student has to infer it. Question 4 is close: do the rest like question 4.\"\n\nThe last version names the fault, points at the location, gives the reason, and hands over an example of the target from the model's own output. That last move is the strongest available: you are showing it one of its own, which the previous lesson explained is the densest steer there is.\n\n## Why \"better\" fails mechanically\n\n\"Better\" has no direction. The model resolves it to the most common meaning of better in the text it learned from, which for generated content is nearly always \"more\": more questions, longer explanations, extra sections. You asked for a change of kind and got a change of quantity. This is not the model misunderstanding. It is the model correctly guessing what an unspecified \"better\" usually means.\n\nEvery vague critique has a default the model falls into. \"More engaging\" produces exclamation marks and a game. \"Simpler\" produces shorter sentences with the same concepts. \"More rigorous\" produces longer sentences with the same concepts. If you find the same disappointing pattern each time, the pattern is the default, and the fix is to say what you actually mean.\n\n## One thing at a time\n\nA critique that lists six problems gets six half-fixes. The model addresses each, briefly, and the output becomes a compromise. Fix the biggest problem first, look at the result, then fix the next. Three focused rounds beat one comprehensive complaint, and they take less time in total because each output is easier to read.\n\n## When to stop and start again\n\nIteration has a limit, and the memory-limit lesson explained it. After eight or ten rounds the conversation is long, the early constraints are drifting out of the window, and the model is carrying the residue of every abandoned version. Signs that you have reached it: the output starts reintroducing things you removed three messages ago; the model apologises and produces the same thing again; the tone shifts.\n\nAt that point, take the best version, open a new chat, paste the best version as your example, and write a clean prompt around it. You lose nothing (the good output travels with you) and you get a model with a clear window.\n\n## The critique is teaching you too\n\nThere is a side benefit that experienced teachers notice quickly. Writing \"these are all recall, I want inference\" forces you to name what you wanted, in words. Teachers who spend a term critiquing generated questions become sharper at seeing the difference between recall and inference in their own questions, because they have had to articulate it two hundred times. The tool did not teach them that. The act of correcting it did.\n\n## A pattern to copy\n\nFor any generated resource, this three-message shape covers most cases:\n\n1. The five-things prompt.\n2. \"Questions [numbers] have problem [X] because [reason]. Do them like question [Y].\"\n3. \"Now check every question against the passage and tell me any whose answer is not actually in it, or is ambiguous.\"\n\nThe third message is a self-check. It is not reliable (the model checking its own work misses things a person would catch) but it catches some, cheaply, before you read. Then you read.\n\n## The honest limit\n\nA model steered by critique can circle. Ask it to make questions harder and then more accessible and then harder again and you will get an oscillation rather than a synthesis, because each message pulls toward one pole. When two goals pull against each other (rigour and accessibility, brevity and completeness) state both in the same message and let it find the balance in one go. Sequential pulls do not converge."},"check":{"question":"A teacher's first output is decent but the questions are all recall. She replies: \"Make it better.\" The next version is longer, with more questions, still all recall. What went wrong?","options":[{"text":"The model needs to be told explicitly that recall questions are bad before it will stop producing them, so the fix is to add that sentence.","correct":false,"tempting":"Close but misplaced; it is not that recall must be labelled bad, it is that \"better\" carries no direction at all, so the model guessed at the most common meaning of better: more."},{"text":"\"Better\" has no direction, so the model improved along its default axis of more; she needed to name the one thing to change.","correct":true},{"text":"Follow-up messages cannot change the type of question once a set has been produced; she must start a new chat with a rewritten first prompt.","correct":false,"tempting":"Follow-ups are the most effective steering available; the failure was the vagueness of the follow-up, not the use of one."},{"text":"The model had already produced its best attempt on the first pass, and any further request can only add volume to what is there.","correct":false,"tempting":"There is no \"best attempt\" state; a specific critique routinely produces a markedly different and better output from the same model."}]},"prev":{"slug":"show-it-one-of-yours","title":"Show it one of yours","url":"https://addaly.com/learn/ai-for-teachers/show-it-one-of-yours"},"next":{"slug":"set-the-voice-and-the-level","title":"Set the voice and the reading level","url":"https://addaly.com/learn/ai-for-teachers/set-the-voice-and-the-level"},"authorship":"Founder and series editor: Dr Mustafa Mun. Written by AI models to a written standard. Exam keys checked blind by a second model: 3,665 of 3,666 agreed. Academic review invited.","review":{"status":"Draft, open for academic review"},"cite":"Addaly (2026) 'The second message matters more than the first', in AI for Teachers. Edited by M. Mun. Addaly. Available at: https://addaly.com/learn/ai-for-teachers/the-second-message-matters-more (Accessed: 13 October 2026).","licence":{"name":"CC BY-NC-SA 4.0","url":"https://creativecommons.org/licenses/by-nc-sa/4.0/"},"reader":{"schema":1,"title":"The second message matters more than the first","courseTitle":"AI for Teachers","module":{"index":2,"slug":"prompting-as-a-teacher","title":"Prompting as a teacher"},"answer":null,"keyIdea":"Treat the first output as a draft and spend your effort on the critique (specific, pointed at one thing, with the reason) because a model steered by feedback improves fast, while a first prompt rewritten five times mostly does not.","hero":{"kind":"figure","figure":{"t":"figure","label":"Figure","figure":{"type":"line","title":"Two ways to spend ten messages","series":[{"label":"Pointed critique of the draft","points":[[1,35],[2,62],[3,78],[4,85],[6,88],[8,84],[10,79]]},{"label":"First prompt rewritten from scratch","points":[[1,35],[2,38],[3,44],[4,46],[6,49],[8,50],[10,50]]}],"xLabel":"Messages spent on one task","yLabel":"% of the output you would use as it is","caption":"Critique moves the output because it names which part of the space to move to. A rewritten first prompt starts the guess again from nothing. Note the turn after eight messages: the conversation is long, early constraints are drifting out of the window, and the right move is a fresh chat with the best version pasted in as the example."},"alt":"Figure: Two ways to spend ten messages. Line chart of % of the output you would use as it is against Messages spent on one task: Pointed critique of the draft (1, 35) (2, 62) (3, 78) (4, 85) (6, 88) (8, 84) (10, 79); First prompt rewritten from scratch (1, 35) (2, 38) (3, 44) (4, 46) (6, 49) (8, 50) (10, 50). Critique moves the output because it names which part of the space to move to. A rewritten first prompt starts the guess again from nothing. Note the turn after eight messages: the conversation is long, early constraints are drifting out of the window, and the right move is a fresh chat with the best version pasted in as the example."}},"blocks":[{"t":"h2","id":"where-teachers-spend-their-effort-and-where-it-pays","runs":[{"text":"Where teachers spend their effort, and where it pays"}]},{"t":"p","runs":[{"text":"Most people meeting these tools put all their effort into the first prompt. They write it, read the output, feel disappointed, and rewrite the prompt from scratch. Three rounds of that and they conclude the tool is not much use."}]},{"t":"p","runs":[{"text":"The effort is in the wrong place. The first output is almost always a draft. The fastest route from draft to usable is not a better first prompt; it is a pointed second message. A model handed a specific critique of its own output improves markedly, because the critique tells it exactly which part of the space to move to. A rewritten first prompt starts the guess again from nothing."}]},{"t":"h2","id":"what-a-useful-second-message-looks-like","runs":[{"text":"What a useful second message looks like"}]},{"t":"p","runs":[{"text":"Three parts: what is wrong, where, and why."}]},{"t":"p","runs":[{"text":"Weak: “Make it better.”"}]},{"t":"p","runs":[{"text":"Weak: “These are too easy.”"}]},{"t":"p","runs":[{"text":"Useful: “Questions 1 to 6 can all be answered by copying a phrase from the passage. I want questions where the answer is not in the text and the student has to infer it. Question 4 is close: do the rest like question 4.”"}]},{"t":"p","runs":[{"text":"The last version names the fault, points at the location, gives the reason, and hands over an example of the target from the model’s own output. That last move is the strongest available: you are showing it one of its own, which the previous lesson explained is the densest steer there is."}]},{"t":"h2","id":"why-better-fails-mechanically","runs":[{"text":"Why “better” fails mechanically"}]},{"t":"p","runs":[{"text":"“Better” has no direction. The model resolves it to the most common meaning of better in the text it learned from, which for generated content is nearly always “more”: more questions, longer explanations, extra sections. You asked for a change of kind and got a change of quantity. This is not the model misunderstanding. It is the model correctly guessing what an unspecified “better” usually means."}]},{"t":"p","runs":[{"text":"Every vague critique has a default the model falls into. “More engaging” produces exclamation marks and a game. “Simpler” produces shorter sentences with the same concepts. “More rigorous” produces longer sentences with the same concepts. If you find the same disappointing pattern each time, the pattern is the default, and the fix is to say what you actually mean."}]},{"t":"h2","id":"one-thing-at-a-time","runs":[{"text":"One thing at a time"}]},{"t":"p","runs":[{"text":"A critique that lists six problems gets six half-fixes. The model addresses each, briefly, and the output becomes a compromise. Fix the biggest problem first, look at the result, then fix the next. Three focused rounds beat one comprehensive complaint, and they take less time in total because each output is easier to read."}]},{"t":"h2","id":"when-to-stop-and-start-again","runs":[{"text":"When to stop and start again"}]},{"t":"p","runs":[{"text":"Iteration has a limit, and the memory-limit lesson explained it. After eight or ten rounds the conversation is long, the early constraints are drifting out of the window, and the model is carrying the residue of every abandoned version. Signs that you have reached it: the output starts reintroducing things you removed three messages ago; the model apologises and produces the same thing again; the tone shifts."}]},{"t":"p","runs":[{"text":"At that point, take the best version, open a new chat, paste the best version as your example, and write a clean prompt around it. You lose nothing (the good output travels with you) and you get a model with a clear window."}]},{"t":"h2","id":"the-critique-is-teaching-you-too","runs":[{"text":"The critique is teaching you too"}]},{"t":"p","runs":[{"text":"There is a side benefit that experienced teachers notice quickly. Writing “these are all recall, I want inference” forces you to name what you wanted, in words. Teachers who spend a term critiquing generated questions become sharper at seeing the difference between recall and inference in their own questions, because they have had to articulate it two hundred times. The tool did not teach them that. The act of correcting it did."}]},{"t":"h2","id":"a-pattern-to-copy","runs":[{"text":"A pattern to copy"}]},{"t":"p","runs":[{"text":"For any generated resource, this three-message shape covers most cases:"}]},{"t":"ol","start":1,"items":[[{"text":"The five-things prompt."}],[{"text":"“Questions [numbers] have problem [X] because [reason]. Do them like question [Y].”"}],[{"text":"“Now check every question against the passage and tell me any whose answer is not actually in it, or is ambiguous.”"}]]},{"t":"p","runs":[{"text":"The third message is a self-check. It is not reliable (the model checking its own work misses things a person would catch) but it catches some, cheaply, before you read. Then you read."}]},{"t":"h2","id":"the-honest-limit","runs":[{"text":"The honest limit"}],"role":"limit"},{"t":"p","runs":[{"text":"A model steered by critique can circle. Ask it to make questions harder and then more accessible and then harder again and you will get an oscillation rather than a synthesis, because each message pulls toward one pole. When two goals pull against each other (rigour and accessibility, brevity and completeness) state both in the same message and let it find the balance in one go. Sequential pulls do not converge."}]}],"check":{"question":[{"text":"A teacher’s first output is decent but the questions are all recall. She replies: “Make it better.” The next version is longer, with more questions, still all recall. What went wrong?"}],"options":[{"text":[{"text":"The model needs to be told explicitly that recall questions are bad before it will stop producing them, so the fix is to add that sentence."}],"correct":false,"tempting":[{"text":"Close but misplaced; it is not that recall must be labelled bad, it is that “better” carries no direction at all, so the model guessed at the most common meaning of better: more."}]},{"text":[{"text":"“Better” has no direction, so the model improved along its default axis of more; she needed to name the one thing to change."}],"correct":true,"tempting":null},{"text":[{"text":"Follow-up messages cannot change the type of question once a set has been produced; she must start a new chat with a rewritten first prompt."}],"correct":false,"tempting":[{"text":"Follow-ups are the most effective steering available; the failure was the vagueness of the follow-up, not the use of one."}]},{"text":[{"text":"The model had already produced its best attempt on the first pass, and any further request can only add volume to what is there."}],"correct":false,"tempting":[{"text":"There is no “best attempt” state; a specific critique routinely produces a markedly different and better output from the same model."}]}],"order":[3,1,2,0],"why":null},"prev":{"slug":"show-it-one-of-yours","title":"Show it one of yours","minutes":8},"next":{"slug":"set-the-voice-and-the-level","title":"Set the voice and the reading level","minutes":9}}}