From 'too long' to a 6-part order of argument── When a draft feels long, the problem is often its order; name each part and its position, and the AI reorders instead of just trimming

01In the same message that said "too long", the operator spelled out a six-part order

In September 2026, the operator read several columns that an AI had drafted. The verdict sent back to the AI was short: the pieces were wordy, and they took far too long to reach the subject of AI. The operator added a request to follow a more Western line of reasoning.

One of those drafts opened with early humans learning to use fire. It moved through farming and printing and reached AI only near the end. Its title was a short riddle. No single sentence was wrong. Yet a reader who finished it could struggle to say what it had been about.

The same message then described the order the operator wanted. Open with one essential question. Say what the answer means and why it holds. Build the body around four questions: what is going on, where it happens, why it happens, and how it works. Close with three key points and then a conclusion. The operator also asked that this skeleton stay invisible to readers, that each piece carry three flow diagrams, and that titles become more specific.

Words about a feeling, "too long" and "too far", had been replaced by words about structure: six parts in a fixed order. That replacement is the subject of this episode. When a draft feels long because of its order rather than its length, naming the order turns the complaint into a request an AI can act on.

02"Too long" sounds like a complaint about length, but it is often a complaint about order

Judging a "wordy" draft by its word count can mislead. Two drafts of equal length feel very different if one states its conclusion in the first paragraph and the other saves it for the last. What the reader feels is the wait before the point arrives.

The operator's second phrase, "too far", names that wait directly. Distance can be counted. In which sentence does the subject first appear? In which paragraph is the conclusion?

The previous episode turned the adjective "cooler" into a measurable property, letter spacing. This episode moves in the same direction, but the target is different. It is not a visual property. It is what goes where inside a piece of writing.

03Without a named order, the AI trims sentences and the point stays far away

Reply "too wordy" and an AI will usually cut. It shortens the opening, drops restatements and tightens paragraphs. If the order does not change, the point is still near the end. The draft is shorter, but the distance remains.

Cutting can also remove the wrong material. Supporting numbers and explanations of sources look, to a length-minded editor, like everything else that is "not the main point". A revision aimed at length can strip out evidence first.

The second cost is repetition. A word about a feeling applies only to the draft in front of you. On the next draft the AI picks its own order again, and the user writes "too long" again.

The operator dealt with that in the same message. The AI was asked to put the essence of the feedback into words and to keep it, so that the same correction would not be needed twice. That day, the six-part order was written into the writing specification the AI works from, as its top-priority section. It could be written down because it had already been put into words.

04Words about order fix three things: the role of each part, their sequence, and their visibility

Here, "words about order" does not mean another evaluative phrase such as "more logical". It means stating which part does which job, in what sequence, and how much of that skeleton the reader sees. The order the operator specified has six parts.

Visibility is part of the order too. The operator asked that this skeleton never appear in headings. A heading should state the claim of its section in a full sentence, not carry a label such as "Why it matters". Keep the sequence; hide the labels.

Some problems are not solved by order. A factual error stays an error in any position. Symbols and coined terms that a first-time reader cannot decode, or vocabulary pitched above the audience, need their own kind of feedback. And writing whose value lies in opening with a scene, such as an essay or a story, is a poor fit. This order suits reports and explanations that people read in order to decide or act.

05Pharmaceutical documents already require the conclusion near the front

For people in pharma, putting the summary first is not new. ICH E3, the guideline on the structure and content of clinical study reports, asks for a brief synopsis near the start of the report, usually limited to three pages. It adds that the synopsis should include numerical data to illustrate results, not just text or p-values.

Yet when a reviewer, a medical affairs specialist or a marketer asks an AI for a draft, that order is easy to lose. The draft starts with product background, the history of the disease or the market, and the actual claim slides to the back. A reviewer who replies "too long" or "I can't see the point" may get back a shorter draft in the same order.

Feedback written as an order lets the writer and the reviewer check the result against the same standard. Is the conclusion on slide 1? Does each response have all three parts? The answers are yes or no.

06An AI given no order picks the opening that is typical of the genre it was asked for

A language model builds text by choosing, step by step, the continuations that have most often followed similar words. Ask for a "column" or an "essay" and it tends toward the openings common in that genre: a scene, a piece of history, a familiar anecdote. Information about why the reader is reading does not enter unless someone supplies it.

Anthropic's published prompting guidance recommends giving instructions as sequential steps, in numbered lists, when the order or completeness of the steps matters. The same guidance says that explaining the reason behind an instruction helps the model understand the goal. The operator's request carried such a reason: readers had to wait too long to reach the subject.

Readers have their own reasons. Nielsen Norman Group, which studies web usability, recommends the inverted pyramid for web writing: the most important information, often the conclusion, comes first. People skim on screens, and some will read only the first paragraph.

Placement matters when the AI is the reader, too. A 2023 study found that language models used information in long inputs best when it sat at the beginning or the end, and noticeably worse when it sat in the middle. A conclusion buried mid-text is easy for both people and models to miss.

07Starting tomorrow, rewrite the unease as one line saying which sentence carries the point

There are four steps.

1. Write the feeling down as it is. "Too long" and "too far" are clues to where the problem sits. Keep them.
2. Measure the distance. Note the sentence or slide where the main claim first appears, and add that line to your feedback.
3. Write the order you want as a numbered list. For example: 1. question, 2. answer, 3. reason. Decide whether the part names may appear in headings.
4. Save the order and attach it every time. Once it is in words, keep it as a reusable instruction template and paste it into each new request.

Check the revised draft with two counts. Do the subject and the question appear in the first two sentences? Is the conclusion in the first paragraph? In the operator's own setup, whether the subject appears in the first two sentences became something a program checks before anything is published.