Claude buries your conclusion.

Ask it for a recommendation and you get context, then background, then considerations, and somewhere near the bottom, the thing you actually wanted to say.

Barbara Minto solved this in the 1960s. McKinsey hired her in 1963, the first woman they took on as an MBA consultant, and made her their writing-skills director. The method she built there is still trained across the firm and it is three rules deep: lead with the answer, support it with a small number of reasons, put the evidence under the reasons.

Here are the five prompts, the Project setup that makes them permanent, and the style file.

Claude. Customize, then Skills, then upload minto-pyramid.zip. ChatGPT. Open a Project and paste the style file into the instructions. The same zip imports as a skill in Codex and the OpenAI API. It is one folder, minto, with SKILL.md inside.

Five prompts

Do not draft yet. Give me the Situation, the Complication, and the
one Question this document exists to answer. If the Complication is
weak, say so, because that means nobody needs this document.

In one sentence, what is the single answer this whole piece argues
for? Not a topic, not a summary. A claim someone could disagree
with. Then tell me whether my draft actually argues it.

Minto invented MECE, so hold me to it. Do my supporting points
overlap, and do they together cover the whole case? Name any
overlap and any gap. Tell me if I have four reasons because I
could not decide which two were the same.

For each level, does it answer the exact question raised by the
line above it? Where a reader would ask "why?" or "how?" and my
next line does not answer it, flag that line.

Reorder my supporting points by one consistent logic: time,
structure, or importance. Tell me which one you used and why the
others were wrong for this piece.

Make it permanent with a Project

Running five prompts by hand every time defeats the point. Put the method into a Claude Project and every conversation inside it inherits the structure.

Create a new Project. In the Project instructions field, paste the style file below. Then work inside that Project whenever you are writing something that has to land: a proposal, a board update, an internal ask, a client email.

The prompts above still work inside the Project. The Project sets the default, the prompts are for when a specific draft has gone wrong.

The style file

You write to the Minto Pyramid Principle. Non-negotiable.

STRUCTURE
- Open with the answer in one sentence. Never warm up, never
  set the scene first.
- Support it with two to four reasons. Never more.
- Evidence sits under the reason it supports, never floating.

BEFORE YOU DRAFT
- If the Situation, Complication and Question are not clear from
  what I gave you, ask for them. Do not guess and proceed.

RULES ON THE REASONS
- They must be mutually exclusive and collectively exhaustive.
- They must share one ordering logic: time, structure, or
  importance. Tell me which you used.
- A detail is not a reason. If it only supports one point, it is
  evidence, and it belongs underneath.

WHEN I AM WRONG
- If my governing thought is not actually supported by my reasons,
  open with that. Do not write around it.
- If the document has no real Complication, tell me it does not
  need to exist.

TONE
- British English. Short sentences. Active voice.
- No throat-clearing openers. No summary paragraph at the end
  restating what you just said.

Where this does not help

Do not use it for exploratory writing. When you are still working out what you think, the pyramid forces a conclusion you have not earned yet, and you end up defending the first idea you had.

It also fights creative drafting. Headlines, hooks and scripts want volume and mess before structure. Turn the Project off for those.

Try it on something you already sent

Take a document you sent last week and run prompt 2 on it. Most drafts do not survive the question, and the ones that do get shorter.

Reply and tell me what your buried conclusion turned out to be. I read every one.