<- all tokdocs
2026.09.25@bengusberg

The Minto prompt in this clip asks Claude for 2-4 reasons while the voiceover says three

AI
Watch on TikTok

View on TikTok ->

This is a 45-second selfie clip posted by @bengusberg on 2026-09-25, at 1080x1920, carrying 162,400 views, 7,263 likes, 65 comments, 1,726 reposts, and 9,056 saves at the time of capture. The transcript runs 181 words across 18 subtitle cues, and 23 frames were extracted at 2-second intervals. Every frame shows the same speaker in a corner with grey acoustic felt on the left and vertical wood slats on the right, with burned-in captions in white and yellow at the bottom. Five distinct overlay cards appear. Frame 001 shows a dark navy card reading "McKinsey & Company" in white serif. Frame 002 shows a low-resolution card with the watermark "untools.co" in the top right and a three-tier grey pyramid. Frame 008 shows the same card at full resolution reading "Minto Pyramid" with the subhead "Make your communication more efficient and clear." Frame 009 shows a grey-background portrait of a blonde woman in a checked jacket while the caption reads "MCKINSEY CONSULTANT BARBARA". Frames 011 through 013 build a blue-box tree diagram labeled "Main Recommendation/Goal/Idea" at the top, "Supporting Arguments" across three linked boxes in the middle, and "Supporting Data / Insights / Conclusions" under nine boxes at the base. Frames 017 and 018 show a Claude composer with the full prompt typed and not yet sent, a Chat and Cowork toggle, a "+" attachment button, a model label reading "Opus 5.5" beside "Medium", a red send arrow, and a row beneath reading "Project or folder", "Manual", and "Output".

The prompt is the whole artifact, and it is fully legible on screen

Frames 017 and 018 hold the composer steady for roughly four seconds. The text cursor sits after the final period, which means the prompt was typed and never submitted on camera. Reproduced verbatim from the screen:

Rewrite this using the Minto Pyramid Principle. Start with the conclusion, add 2-4 supporting
reasons, and then end with our supporting evidence. Remove anything that is not necessary to
meet the principle's guidelines, and let me know if anything is missing.

Two details are worth preserving exactly. The prompt says "our supporting evidence" where "your" reads as the intended word. The prompt also asks Claude to flag gaps, which is the only instruction in it that produces something other than a rewrite.

The surrounding interface is checkable. The model label reads "Opus 5.5" with "Medium" next to it. Anthropic released Claude Opus 5.5 on 2026-09-22 under the API ID claude-opus-5-5, three days before this clip was posted, and Anthropic's own model table lists Opus 5.5 with a default effort of medium. The on-screen state is the shipping default. The Chat and Cowork toggle is consistent with Cowork, Anthropic's agent for general knowledge work, which launched on desktop in January 2026 and reached web and mobile in July 2026.

The audio and the prompt give two different numbers

The voiceover states the framework in three parts:

Number one, start with your conclusion. Two, add three reasons. Three, end with your supporting evidence or data.

Frame 012 reinforces this with a single yellow "3" caption and a raised three-finger gesture. The prompt on screen asks for "2-4 supporting reasons". A viewer who copies the prompt gets a range. A viewer who follows the spoken instruction gets a fixed count.

The range is the better instruction. Barbara Minto's actual rules govern the relationship between levels rather than the headcount at any one level. Her three rules are that ideas in each grouping must be the same kind of idea, that ideas at any level must summarize the ideas grouped below them, and that ideas in each grouping must be logically ordered. Untools, which published the pyramid card shown in frames 002 and 008, states no required number and notes that the detail layer can be skipped when the key arguments carry the point. The "three reasons" line is a convention that travels with the framework, not a rule inside it.

The McKinsey attribution is accurate, and the diagram is not McKinsey's

The clip says the principle was "created by legendary McKinsey consultant Barbara Minto." That checks out. Minto joined McKinsey in Cleveland in 1963 as the firm's first female MBA hire, moved to London in 1966, and stayed until 1973. She published The Pyramid Principle: Logic in Writing and Thinking in 1985 and The Minto Pyramid Principle: Logic in Writing, Thinking and Problem Solving in 1996. Her own site states the core idea directly: "The Minto Pyramid Principle says that your thinking will be easy for a reader to grasp if you present the ideas organized as a pyramid under a single point."

The opening claim, "Every McKinsey consultant learns this one writing framework," is the part I could not confirm. Secondary consulting-industry sources say it is taught in McKinsey's onboarding program, and McKinsey's alumni site hosts an interview with Minto about MECE, but that page timed out on repeated fetch attempts and I have no primary McKinsey source describing the curriculum. The visual evidence in the clip does not support the claim either. The "McKinsey & Company" card in frame 001 is a wordmark with no document behind it, and the pyramid graphic in frames 002 and 008 comes from untools.co, a third-party mental-model site.

The clip never shows Claude's output

The stated payoff is "now every doc and email I write sounds like an expert" and "Claude will then polish your writing to always sound like a McKinsey consultant every time you write anything." No frame in the clip shows a Claude response. The composer appears in exactly two frames, both with the prompt unsent. There is no before-and-after, no sample document, and no rewritten paragraph anywhere in the 23 frames.

This matters because the prompt is a transform that needs an input. It begins with "Rewrite this," which assumes text is already pasted or a file is already attached. The clip shows an empty composer with only the instruction in it.

"Custom instructions" is not the name of the setting it points you to

The voiceover says:

You can even add this to your custom instructions if you want.

Anthropic's personalization documentation names the account-wide setting "Instructions for Claude," alongside project instructions and skills. "Custom instructions" is widely used informally, so a viewer will find the right screen, but the label in the product differs.

There is also a practical problem with the advice that the clip does not raise. A prompt whose first word is "Rewrite this" is written for a single turn against a specific input. Promoting it to an account-wide instruction applies it to every conversation, including ones where nothing has been pasted to rewrite.

Key Takeaways

  • The reusable artifact is one prompt, legible in frames 017 and 018, asking Claude to rewrite text using the Minto Pyramid Principle with the conclusion first, 2-4 supporting reasons, supporting evidence last, non-essential content removed, and gaps flagged.
  • The spoken "three reasons" and the on-screen "2-4 supporting reasons" do not match. The written range is closer to Minto's published rules, which constrain how levels relate rather than how many items sit at each level.
  • Barbara Minto's biography as stated in the clip is accurate. She was McKinsey's first female MBA hire in 1963 and published the principle in book form in 1985.
  • The on-screen Claude state is current and correct. Opus 5.5 shipped on 2026-09-22 and its documented default effort is medium, which is exactly what the composer shows.
  • Unverified: the claim that "every McKinsey consultant learns this one writing framework." McKinsey's alumni page on Minto timed out on repeated fetches, and I found no primary McKinsey source describing its onboarding curriculum. Also unverified: that the technique makes writing "sound like an expert," since the clip shows no Claude output at all.
  • Unverified: whether Whisper mis-transcribed anything. At every point where a burned-in caption was readable, the caption matched the Whisper transcript word for word, including "$4 MILLION", "MINTO PYRAMID PRINCIPLE", and "I'M BEN GUSBERG".

Resources

Published September 25, 2026. Writeup generated from a favorited TikTok.