Gains:
- Ability to control the structure itself by producing the document first as a skeleton, then as a draft, section by section, and then in the order of revision.
- Ability to prevent fabricated data by giving the main message, numbers and data itself and interweaving the sentence flow into artificial intelligence
- Ability to reuse the same content by converting it into different formats (report, summary, presentation, e-mail) and verify each fact without sending it
Writing is the knowledge worker's most visible output—report, proposal, presentation, proposal text, internal document. But writing is also the most time-consuming, most postponed task: fear of the blank page, difficulty in establishing structure, rounds of revisions. AI is a lever here; creates a skeleton, produces the first draft, revises it, and translates it into different formats. However, the real value is in you: the idea, the message and the responsibility are yours. AI produces the muscles of writing; You provide the framework (what to say) and approval (is it correct, can it be sent).
Terms. The skeleton/outline is the structure of the document consisting of section headings. A draft is text that is filled in but not final. Revision is a round of improving the manuscript. An executive summary is a short summary at the beginning of a long document that the decision maker will read. Repurposing is converting the same content into a different format (report→presentation→e-mail).
Correct order of authoring workflow
The most common mistake is to directly tell the AI “write me a report” — this produces generic text with no message. The correct order is: skeleton first, then draft, then revision.
- You choose the skeleton (let AI help you). What do you want to say, in what order? AI suggests skeleton, you choose.
- Draft it section by section. Request each section separately; Having AI focus on one section is better than requesting the entire document at once.
- Provide your own material. The numbers, the examples, the inside knowledge come from you; AI weaves language.
- Revise. Tour for tone, length, clarity.
- Verify. Is every fact, figure, claim verified?
- Own it. Give it the final polish with your own voice; clear the clichés.
Tip: Tell the AI "make the skeleton first, don't write the content before I approve it." So you check the structure and course correct before the AI writes 3 pages in the wrong direction.
Message from you, language from AI
Critical principle: You can't write something the AI doesn't know. The real value in your report—why it dropped this quarter, what the customer really wants, what proposition you're advocating—is in your head. AI doesn't know this. If you just tell it to "write a sales report" it will produce empty, one-size-fits-all text. If you give it your own data, findings, and opinions, it will turn it into a fluent and organized text. So: you provide the density of ideas, let AI provide the fluidity.
piece of document
Who produces
note
Main message / thesis
you
The one thing AI can't know
Figures/data
you
AI makes up; you give
Structure/skeleton
You + AI
AI recommends, you choose
Sentence flow / language
A.I.
Where is it strongest?
Tone/format
AI (with your description)
Describe clearly
Accuracy confirmation
you
Non-transferable
Four copyable writing templates
For the following document, ONLY the skeleton (section headings) appear first:Document type: [report / proposal / recommendation]. Reader: [who]. Purpose: [what].My main message: [one sentence].Do not write content without confirming the framework; Suggest 5-7 titles.
Write this section. My material is below; just use this, data fitting.Section: [title]. Tone: [professional]. Length: [~200 words].Material: """[own notes, figures, findings]"""
Write an executive summary for the following completed document:- No more than 5 items, in the language the decision maker will read.- Findings and recommendations only in the document; adding new information.- A clear "recommended action" line at the end.Document: """[paste]"""
Convert the following report into a [10-slide presentation / short email] format:- Title + 3 bullet points for each slide.- Keep numbers, don't add comments.- Keep the presentation flow logical (situation → finding → recommendation → next step). Report: """[paste]"""
Weak prompt / Strong prompt
Weak: "Write me a quarterly sales report." (The AI makes up a generic text with no message, no data — possibly with fake numbers.)
Strong: "I've approved the following framework. Now write the 'Territory Performance' section. Material: North +12%, South -4% (one customer churn), West steady. My main message: decline single-sourced, temporary. Tone: neutral-analytical, ~150 words. Don't go beyond the numbers I gave." Since the strong version gives the message, data and boundary, it returns a real and usable text.
three mini cases
Case 1 — Beating the blank page. A manager was postponing his annual evaluation report by 2 days each year. He had the AI first create a skeleton and then draft each section, feeding it with his own notes; My total writing was reduced to 3 hours. The payoff came not from the AI typing, but from making it easier to get started — once the skeleton was in hand, the rest flowed.
Case 2 — Fake number disaster. “Write a sales report,” one analyst said, slightly editing the text the AI returned — with fake growth numbers in it — and presenting it to management. The management thought the numbers were real and made a decision; When the mistake was later revealed, they had to reverse the decision. Lesson: Never make AI fit data; You give the numbers, he just weaves the language.
Case 3 — One content, three formats. A marketer translated his 8-page strategy into three formats with AI: 1-page summary for management, 10-slide presentation for the team, short email for the customer. He preserved the numbers and set the tone in every format. Three outputs from the same content, 40 minutes in total. The gain came from reuse.
Common mistakes
- Saying "write" directly: Outputs generic text without skeleton; Build the structure first.
- Making up data: You give the numbers; Restrict AI to “generate data”.
- Requesting the entire document at once: Section by section provides better quality.
- Ditch the clichés: Clear out AI phrases like “This report covers our dynamic marketplace…”
- Trying to get the AI to establish the message: The main thesis is yours; AI can't know that, you give it to him.
- Bypass verification: The document you send bears your signature; verify every claim.
Caution: The AI-generated text feels "correct" because it is fluid. Fluency is not accuracy. Verify sentences one by one, especially those containing numbers, dates, names and claims, no matter how beautifully written they are.
In summary
- The correct order: first the skeleton (you determine), then the chapter-by-chapter draft, then revision.
- Messages, data and confirmation are from you; structure common; sentence flow and tone are from AI.
- You can't write what AI doesn't know; You provide the density of ideas, let AI provide the fluency.
- Never make the AI fit the numbers; impose a "data generation" constraint and verify every fact.
- Reuse the same content by converting it to different formats (report→summary→presentation→e-mail).
Application task
Choose an actual document you need to write. With the first template, first just extract the skeleton and edit it yourself. Then have a second template draft a section with your own data — make sure you don't make the AI make up the data. When finished, convert the document to a second format (summary or presentation) with the fourth template and verify that the numbers are preserved.
checklist
- [ ] First, I removed the skeleton and approved the structure myself.
- [ ] I gave the numbers and data, I did not make them fit it to AI.
- [ ] I had the chapters drafted separately.
- [ ] I determined the main message/thesis myself.
- [ ] I cleaned up the cliché AI patterns and embraced them with my own voice.
- [ ] I verified every fact and figure before posting.