AI agrees with everything I say
The AI praises weak ideas, caves when you push back and tells you what you want to hear. For decisions, plans and drafts, that makes its feedback close to useless.
Open the fixAI won't follow our document template
Last verified: Claude Fable 5 / GPT-5 era models ยท July 2026
You have a house structure: proposals open a certain way, reports run in a set order, every SOW carries the same boilerplate. The AI knows none of that, so it reorders your sections, quietly drops the bits it finds boring, and invents its own headings. This page gives you a prompt that turns your house structure into a reusable block the AI follows every time, plus where to keep it so you never re-explain it.
Part 1 · The generator prompt
Paste this into a fresh Claude or ChatGPT chat. It interviews you about each document type and its required shape, then hands back a structure block you paste at the top of any drafting chat so the AI stops improvising your layout.
Generator prompt
You're going to build my company's DOCUMENT STRUCTURE BLOCK: a reusable instruction block I'll paste into any AI chat that drafts our documents, so the AI follows our house structure without being reminded. Interview me first, one question at a time. Where I don't know, propose a sensible default and move on: 1. Which 3 to 5 document types do we produce most, and what is each one for (proposals, client reports, SOWs, board memos)? 2. For each type, what sections do readers expect, in what exact order? Paste an example document if you have one and infer the skeleton from it. 3. How does each document open: straight in, or a summary first? What do we call that opening section? 4. What is mandatory in every document of that type (a scope note, pricing terms, a legal line, next steps) and what is optional? 5. What does the AI most often get wrong about our structure today: reorders sections, drops parts, invents headings, over-explains? Then output my DOCUMENT STRUCTURE BLOCK as one copy-paste block containing: (a) one line of company context, (b) for each document type, its ordered skeleton with the exact section names, (c) the opening convention and the mandatory elements per type, (d) hard rules: never reorder sections, never invent a heading, never drop a mandatory section, and ask if a required input is missing rather than inventing it, (e) a closing self-audit that checks the draft against the correct skeleton before output and appends "Structure check: pass" or lists what is out of order or missing. Keep it under 400 words so it fits any project-instructions field. After you output it, name the three ways the AI is most likely to still drift from our structure and how the block prevents each.
Part 2 · Where to keep it so you stop repeating yourself
The block fixes what the AI produces. This fixes having to paste it every time.
- Save the structure block once, in your AI tool's project or custom-instructions field (a Claude Project, a ChatGPT project or custom GPT), not in a note you paste each session.
- Keep one skeleton document per type: an empty proposal, report and SOW with the section headings already in order. Start every new document from the matching skeleton so the shape is right before a word goes in.
- When you brief the AI, name the document type ("draft this as one of our client reports"). The block maps that type to the correct skeleton automatically.
- If the AI is missing an input a mandatory section needs, the block tells it to ask rather than invent. Answer the question instead of fixing the gap afterwards.
Part 3 · A worked example
Here is the block the generator hands back, for our running example company, the fictional consultancy Fenwick & Cole.
Example: Fenwick & Cole, a boutique management consultancy
CONTEXT: Fenwick & Cole is a boutique management consultancy. Documents we produce: client proposals, engagement reports, statements of work, board memos.
SKELETONS (follow the exact order; never reorder, never invent a heading):
- Proposal: In short, Context, Approach, Team, Timeline, Commercials, Next steps.
- Engagement report: In short, Context, Findings, Recommendations, Next steps, Appendices.
- Statement of work: Summary, Scope, Deliverables, Timeline, Fees, Assumptions, Sign-off.
- Board memo: Recommendation, Background, Options, Risks, Ask.
OPENING: every client-facing document opens with a one-paragraph summary under its first heading ("In short" or "Summary"). Board memos open with the recommendation, not the background.
MANDATORY: proposals and SOWs always carry a Commercials/Fees section and a Next steps/Sign-off section. Reports always carry numbered Recommendations. Never drop a mandatory section; if you lack the input for one, ask me rather than inventing it.
SELF-AUDIT: before output, check the draft against the correct skeleton for its type. Append one line: "Structure check: pass", or list any section that is out of order, missing, or invented.
Common questions
Why does AI ignore my document template?
The AI has no memory of your house structure between chats, so it falls back to a generic essay shape: an intro, a few invented headings, a conclusion. It isn't refusing your template, it never had it. Giving it your section order as an explicit instruction, every time, is what makes it comply.
How do I make AI follow a set document structure?
Give it a structure block that lists, per document type, the exact section names in order and which sections are mandatory, and tell it never to reorder or invent headings and to ask when an input is missing. The generator prompt on this page builds that block from your real documents. Store it in your AI tool's project instructions so it applies to every draft.
How do I stop re-explaining our format in every chat?
Move the structure block out of copy-paste and into always-on context: a Claude Project or a ChatGPT project's instructions, so the AI reads it on every message without you pasting it. That is the difference between a workaround you repeat and a fix that sticks, and it is exactly what an AI Brain install sets up across every workflow.
More fixes for AI output
AI keeps using em dashes and other AI tells
AI leans on the em dash, 'not X but Y' sentences, lists of three and stock openers. Readers spot them instantly, so someone edits every draft to take the tells out.
Open the fixAI answers are too long and padded
Ask a simple question and get a wall of text: a restated question, a bulleted overview, caveats and a summary of what it just said. You skim for the answer every time.
Open the fixThe prompt is the aspirin. Stored context is the cure.
The structure block you just generated is exactly what an AI Brain stores as always-on context. Paste it and it works for this chat; installed, it lives in the system and every proposal, report and SOW comes out in house order without anyone pasting anything. This page proved it on your own documents. The install is how it becomes the default.
What else does AI get wrong that you fix by hand? Browse the other fixes