Drafting professional documents

~ min read

30-second summary

The blank page of a professional document is worse than the blank page of an email: twenty pages to fill, not twenty lines. The AI solves the eighty percent that is “where do I start,” not the rest.
Three canonical documents: proposal (problem, solution, approach, costs), brief (for whoever continues the work or decides), report (data, interpretation, next steps).
The pattern is outline + filling, not “write everything, I’ll fix it.” First the outline, you critique it, then section by section. Re-reading eight hundred new words and cutting three hundred is slower than writing four hundred on top of the right outline.
Calibrate explicitly on audience, length constraints, level of detail. “For the CEO” and “for the technical team” produce two different documents.

The blank page of an email is annoying. The blank page of a professional document is worse: twenty pages to fill, not twenty lines, and on the other side you have a client, a boss, or a committee who will read. Eighty percent of the block isn’t “I don’t know how to write”: it’s “I don’t know where to start.” That’s where AI works best, not on the fine prose.

On a long document, AI as ghostwriter fails in a specific way: it produces two thousand plausible-sounding words, but re-reading them and cutting eight hundred is slower than writing a thousand on top of the right outline. The work doesn’t move: it multiplies. What works is splitting the job in two moments: first the structure (what goes where, how long, for whom), then the text (section by section, with you as the filter).

This doesn’t mean “AI as ghostwriter” is always to avoid: on a quick email, a note to yourself, a group greeting to colleagues, pasting the first version is fine, the cost of a slip is low. On documents with a signature and someone deciding as they read, the cost goes up, and the pattern changes.

Three canonical documents

Commercial or project proposal

A proposal has standard sections: problem, solution, approach, timeline, costs, team or references. The sections aren’t yours, they belong to the genre: a client who reads proposals every month expects to find them in roughly the same order. Your value isn’t inventing the structure, it’s calibrating the sections on the specific case: how much weight the problem carries (a lot if the client doesn’t yet know why they need it), how much the team carries (a lot if you’re a small outfit selling trust).

AI is useful for proposing you a clean outline and getting your thinking moving. You critique it: cut a section that isn’t needed, add one the client explicitly asked for, reorder. Often the final outline is seventy percent what the AI gave you plus thirty percent of your corrections.

Internal brief

A brief is short by construction: it’s read by people with no time. Three recurring types: a colleague who has to continue your work, a team receiving a handoff (the transfer of a project in progress), a boss who has to decide something. The format shifts: for continuing the work you need the operational steps; for deciding you need the problem plus two or three options with trade-offs.

The brief’s trap is the opposite of the proposal: writing it too long. If you tell the AI “handoff brief, one page max, who does what and with which files,” it gives you a short skeleton that’s easier to shorten than to fill. Typically four slots: current state (where we are, in three lines), what’s already been decided (so closed discussions don’t reopen), what’s left to do and who owns each piece, essential files and links. If it starts long, you’re starting behind.

Report or review

A report has a life cycle: data → interpretation → next steps. Weekly status update, post-mortem of a closed project, quarterly review. The format is stable, what shifts is the weight of the three parts: a weekly update is twenty percent data and eighty percent next steps; a post-mortem is sixty percent interpretation.

AI writes the interpretation part well if you give it structured data (numbers, events, dates) and ask what it sees. Example: you hand it a table with monthly revenue for the past six months, number of projects delivered, hours over budget, and ask “write two paragraphs reading this data, without inventing things you don’t see.” It finds the patterns (anomalous month, correlation between hours over and margin), and you cut the readings you know are wrong for contextual reasons it doesn’t have. It doesn’t write the next steps part well, because those depend on the company context it doesn’t know. You often use it for the first part, and write the second yourself.

This pattern (structured data in → reading out) extends well beyond the three canonical documents: internal policies, technical specifications, short pitches all work similarly, with recurring structures that shift only in weighting. If your document isn’t one of the three, start by asking the AI “what standard structure does a document of type X have in my industry?” and then apply the pattern below.

The pattern: outline first, filling second

The pattern that works well is opposite to instinct. Instinct is “ask the AI to write me the proposal,” read, fix. The pattern is: ask for an outline, critique it, then section by section.

It works better for two reasons. Critiquing the outline costs a lot less than critiquing the text: changing “drop the timeline section, add a why-us section” is two lines, while changing the text of a whole section is ten minutes of reading plus five of rewriting. And when the AI writes on top of an outline you’ve already validated, the text comes out more targeted: it has fewer degrees of freedom to invent sections or wander, because the perimeter is fixed.

“But I don’t know what should be there, that’s why I’m asking the AI”: fair objection, and the answer is that you don’t start from zero. You start from the outline it proposes and you read it entry by entry with a single question in mind: “does this section speak to my client, or to a generic client?” That question alone surfaces seventy percent of the critiques. The rest comes by asking for a second outline with a different constraint (“shorter,” “for a non-technical audience,” “dropping the sections a non-paying client would see”) and comparing the two: the differences tell you what the AI is tuned on and what you are tuned on.

The second step (section by section) isn’t “give me the full document based on the outline.” It’s one section at a time: “write the Problem section, 150-200 words, neutral tone.” Small pieces you re-read and correct as you go. Keep a single chat for the whole document: the AI sees the sections already written and keeps tone consistent; opening separate chats for each section fragments the voice. If a section leaves you unhappy, you rewrite it while it’s fresh in your head, not at the end over a block of two thousand words.

A concrete example

You’re writing a commercial proposal for a potential client, a fifty-person family business that has never hired an external consultant. They know you through an introduction from a previous client of yours. The first pass is asking for a generic outline.

You

Propose an outline for a commercial proposal for organizational consulting, six months, for a medium-sized company. Just the section index with a one-line description each, not the text.

Organizational consulting proposal: outline 1. Executive summary (half page): summary of what we're proposing and why. 2. The problem: analysis of the current situation and its costs. 3. The proposed solution: what we want to achieve in six months. 4. Approach and methodology: phases, techniques, tools. 5. Timeline: six-month gantt chart with main milestones. 6. Investment: costs per phase, total, invoicing terms. 7. The team: who will work on the project and why they're the right people. 8. References: similar projects already completed. 9. Next steps: what's needed to begin. Organizational consulting proposal: outline 1. Executive summary (half page): summary of what we're proposing and why. 2. The problem: analysis of the current situation and its costs. 3. The proposed solution: what we want to achieve in six months. 4. Approach and methodology: phases, techniques, tools. 5. Timeline: six-month gantt chart with main milestones. 6. Investment: costs per phase, total, invoicing terms. 7. The team: who will work on the project and why they're the right people. 8. References: similar projects already completed. 9. Next steps: what's needed to begin.

This is the textbook outline. It works for a generic client. For this specific client (family, first time, indirect introduction) it’s too neutral: trust is missing, and the detailed timeline so early intimidates a client who hasn’t decided yet.

You critique it in plain words:

You

Good start. But this client is a 50-person family business, first time hiring an external consultant, introduced by a mutual contact. Revise the outline taking these things into account: - before the problem, add a short 'Why us, why now' section that hooks into the indirect introduction; - drop the detailed timeline, for now we just need a sense of timing, the gantt we do at the kick-off; - the costs section isn't a price list, it has to justify the value against the cost of doing nothing; - shorten to 6 sections max, not 9. Good start. But this client is a 50-person family business, first time hiring an external consultant, introduced by a mutual contact. Revise the outline taking these things into account: - before the problem, add a short 'Why us, why now' section that hooks into the indirect introduction; - drop the detailed timeline, for now we just need a sense of timing, the gantt we do at the kick-off; - the costs section isn't a price list, it has to justify the value against the cost of doing nothing; - shorten to 6 sections max, not 9.

Revised proposal, six sections 1. Why us, why now (half page): who we are, the introduction that brought us here, why we think we're right for you at this moment. 2. What we observe (one page): the current situation as we've understood it from early conversations, without premature diagnoses. 3. What we propose (one and a half pages): the expected outcome in six months, with two or three concrete priorities, not an exhaustive list. 4. How we'll work (one page): broad phases and working style (meeting cadence, people involved, decisions in the client's hands), without a gantt. 5. Investment and return (half page): project cost against the estimated cost of doing nothing. Invoicing terms at close. 6. Next steps (a few lines): what's needed to begin (meeting with the owner, access to essential data, engagement letter). Revised proposal, six sections 1. Why us, why now (half page): who we are, the introduction that brought us here, why we think we're right for you at this moment. 2. What we observe (one page): the current situation as we've understood it from early conversations, without premature diagnoses. 3. What we propose (one and a half pages): the expected outcome in six months, with two or three concrete priorities, not an exhaustive list. 4. How we'll work (one page): broad phases and working style (meeting cadence, people involved, decisions in the client's hands), without a gantt. 5. Investment and return (half page): project cost against the estimated cost of doing nothing. Invoicing terms at close. 6. Next steps (a few lines): what's needed to begin (meeting with the owner, access to essential data, engagement letter).

The final outline is shorter, speaks to that client instead of a generic one, and still has everything it needs. Ten minutes to get there. Now you ask the AI section by section to write the draft, with the length constraints from your index. What you’ll read is faster to fix, because you already know where each piece should land.

What NOT to delegate

The strategy. If you don’t know what to propose, the AI won’t figure it out for you. AI knows how a proposal is structured in general; it doesn’t know what the right solution for your client is. If you ask for a proposal without having decided your approach yourself, you’ll get a textbook proposal, sounding good but not yours. The initial thinking stays yours, then the AI helps you write it down.

Specific data. Revenue numbers, names of previous clients, concrete results from past projects: the AI doesn’t have them and, if you ask it to invent them to fill space, it will. It sounds obvious and yet it’s a mistake that happens (the document reads smoothly, an invented figure slips by). Give it real data as input, or leave explicit blanks you fill in later (“[INSERT THREE 2025 REFERENCES HERE]”).

Contextual decisions. If two options are on the table (approach A or B, package X or Y, include a section or not), the choice is yours. The AI can list the trade-offs, it can’t choose for you: it doesn’t know the internal priorities, the client’s constraints, the political reasons. The corollary is in Things NOT to do: where you sign, you decide.

What comes next

A document rarely lives alone: almost always it gets presented, discussed, or followed by a meeting. The next lesson, Preparing a meeting, closes the loop: how to use AI to arrive at the meeting prepared without losing two hours of prep.

Check what you learned

About Privacy Open manual, written by a real person, in collaboration with AI.