Writing better emails, reports and documents with AI

Learn in your browser · Chapter 01 of 07

Emails and short reports are the single most common thing people ask AI to help with at work. Used well, AI removes the blank-page problem and speeds up wording. Used badly, it produces something generic, over-long, or quietly wrong. This chapter sets the one rule the rest of the workbook builds on.

Ask people what they actually use AI for at work, and "help me write this email" or "turn these notes into a report" comes up more than almost anything else. That is not a coincidence: drafting text well takes real time, and AI is genuinely fast at producing a first pass once it has something to work from.

What AI is good at here is narrow and specific: turning facts and instructions you already have into clear, well-structured wording. It is not good at knowing your situation, your organisation's specific process, or what actually happened, unless you tell it. Ask it to write a report on a project you have not described, and it will not stop to ask; it will fill the gaps with something plausible-sounding, which is far more of a problem than an obviously blank space would be.

This matters more for reports than emails, because a report's numbers and claims often get repeated, forwarded and relied on well after you have stopped thinking about how it was drafted. A wrong tone in an email is awkward and easily fixed. A wrong figure in a report can travel.

• Wording and structure: yes, this is where AI genuinely saves time.
• Facts, figures and what actually happened: no, this stays entirely with you.
• Judging what the reader needs to know first: partly, AI can help once you tell it who is reading and why.

Learn in your browser · Chapter 02 of 07

Most disappointing AI drafts share the same cause: the person asked for a document before giving AI anything to build it from. This chapter is the method that fixes that, every time.

A vague request such as "write me an email about the delay" produces a vague, generic draft, because there is nothing specific to draft from. AI cannot read your inbox, your project history or your intentions. It can only use what you put in the request.

1. State the facts first, in your own rough words: what happened, what you actually want, and any real names, dates or figures. Keep this as plain notes, not a draft.
2. Say who is reading it: their role, how much they already know, and their likely mood or priorities.
3. Say what you want to happen after they read it: a decision, an approval, an acknowledgement, or simply being informed.
4. Say the tone and rough length you want, rather than accepting whatever AI defaults to.
5. Only then ask AI to draft from those notes, not from a one-line summary of the situation.

Scroll sideways to see every column.

Learn in your browser · Chapter 03 of 07

AI's default writing style leans long, hedgy and formal, whichever tone you actually need. This chapter is about noticing that default and correcting it deliberately, rather than accepting it.

Left to its own defaults, most AI tools write longer than necessary, add qualifying phrases that soften every statement, and reach for a formal register even for a quick internal update. None of that is wrong exactly. It is just a default, and a default is not the same as what your situation actually needs.

Tone and length are both things you can direct precisely, the same way you would direct a colleague: shorter, less formal, more direct, cut the second paragraph, say this in three sentences rather than three paragraphs. Vague feedback such as "make it better" produces another generic pass. Specific feedback produces a specific fix.

• "Cut this by a third" when a draft is padded rather than genuinely detailed.
• "Say the outcome in the first sentence" when the point is buried.
• "Drop the hedging" when every sentence has a qualifier such as "it seems" or "to some extent".
• "Match this tone" with a short example of how you actually write, when a draft sounds like nobody you know.

Learn in your browser · Chapter 04 of 07

AI-drafted writing has recognisable patterns. Learning to spot them in your own drafts, before you send them, is a skill worth having whether or not anyone else notices.

AI-generated text tends to share certain habits, whichever tool produced it: a throat-clearing opening that restates the obvious before getting to the point, filler transitional phrases used as padding, a fondness for rounding a document off with an unrequested "in conclusion" summary, and a rhythm where every sentence is roughly the same length. None of these make a draft wrong. They do make it read as generic, and a reader who has seen enough AI-drafted text will notice.

Scroll sideways to see every column.

Learn in your browser · Chapter 05 of 07

Seeing the method applied end to end is worth more than any single rule. Here is the same real situation drafted two ways.

The situation: you manage a small team, and one member's report is late for the second time this month. You need to raise it without escalating unnecessarily, since there is no sign of a wider problem yet, just a pattern worth naming early.

This draft is not rude, but it is nearly useless. It never names the actual report, the actual date, or the actual pattern of two late reports in a row. It reads as a template because, given a one-line request, that is exactly what it is.

• Names the actual report and the actual pattern (twice, not vaguely "delays").
• States there is no bigger worry, so the reader does not over-read the tone.
• Asks a real question rather than issuing an instruction, since the goal is understanding, not correction.
• Ends with a genuine offer, which changes how the whole message lands.

Learn in your browser · Chapter 06 of 07

Reports fail differently to emails: not usually in tone, but in structure, because the reader needs to find the actual point fast, not read start to finish.

The situation: a short internal update on a delayed software rollout, going to your manager, who needs to know the impact and the plan, not a full history.

Every sentence in that draft is technically true and says almost nothing. There is no actual date, no actual reason, no actual decision needed from the reader, and "in the coming weeks" could mean anything from one week to six.

• States the actual new date in the first line, not buried in a closing paragraph.
• Gives one concrete reason, not "various factors".
• States impact plainly, including that there is not one where that is true, rather than leaving the reader to guess.
• Ends with exactly what action is needed, so the reader knows what to do with the report, not just what happened.

Learn in your browser · Chapter 07 of 07

Every method in this workbook only works if the last step is a genuine read-through, not a glance. This closing chapter is that discipline, plus a recap.

The single most common way an AI-assisted email or report goes wrong is not a bad draft, it is an unchecked one. A specific, well-directed draft can still contain a wrong name, a fact that drifted during a tightening pass, or a tone that reads differently to the actual recipient than it did to you.

1. Read the whole thing once, in full, as if you were the person receiving it, not skimming for typos.
2. Check every name, date, figure and reference against your own original notes, not against how confident the sentence sounds.
3. Check the opening line states the actual point, not a throat-clearing lead-in.
4. Check for the generic-writing tells from Chapter 4: filler transitions, an unrequested conclusion, a metronome rhythm.
5. Ask yourself: would I be comfortable if the recipient asked me to explain any single line of this?

You can now give AI the context it needs before asking for a draft, direct its tone and length with specific instructions rather than vague feedback, catch its generic-writing habits in your own text, and apply an answer-first structure to a short report. None of that replaces the final read-through. It makes what you are reading through worth sending.

This workbook is free educational content published by Mickai LTD, not professional writing, legal or business advice, and Mickai accepts no liability for how you use it.

https://trust-agent.ai/exam/writing-better-emails-reports-with-ai

Being able to show your own real facts and judgement behind a piece of writing, rather than trusting a draft blindly, is the same principle behind Mickai's approach to trustworthy AI: a human stays in control, and that control means genuinely checking, not just skimming.