Writing and Documents

A volunteer named Rosa has to write a thank-you letter to the family who donated a van to the food pantry. She’s stared at the blank screen for twenty minutes. She knows what she feels. She just can’t find the first sentence. So the letter doesn’t get written today, and probably not tomorrow either, and the good deed goes unthanked because the writing got in the way.

Why this matters to you

Writing is the tollbooth on a hundred small tasks. The email to your kid’s teacher. The note in the neighborhood group. The cover letter, the apology, the reminder, the recommendation. Most people don’t dislike having written. They dislike starting.

An AI tool clears the tollbooth. It gives you a first draft in seconds, so the hard part stops being the blank page and becomes something much easier: reacting to words that already exist. You read a draft and you know immediately, “warmer than that,” or “too long,” or “that’s not how I’d say it.” That reaction is the whole skill. You already have it.

The theme from Chapter 2 carries this entire chapter: you are the editor, and the AI drafts. It is fast and tireless and a little generic. You are slow and particular and the only one who knows what you actually mean. Put those together correctly and you produce better writing in less time. Put them together lazily and you flood the world with text that sounds like everyone and no one. The difference is you doing your job.

From rough notes to a real draft

The single most useful move in this chapter is turning a pile of notes into prose. You don’t need a clever request. You need to hand over the raw material and say what you want made from it.

Notice the pattern: give the AI the actual stuff, not a description of the stuff. The cheat sheet calls this “the raw material,” and it’s worth more than any clever wording. If you have bullet points, paste the bullet points. If you have a messy paragraph, paste the messy paragraph. Then name the shape you want it in.

Try It Now Think of a real message you owe someone. A thank-you, a request, an apology, an update. Jot three or four rough facts, no full sentences, just the raw ingredients. For example: “neighbor watched our dog for the weekend. Dog usually anxious, was calm with them. Want to thank them and offer to return the favor.”

Open your AI tool and paste this:

“Turn these notes into a warm, short thank-you message to a neighbor. Keep it under 100 words and sound like a normal person, not a greeting card. Here are my notes: [paste your notes].”

Now the variation. Add one line: “Give me two versions: one casual, one a little more heartfelt.” What did you notice about having two to choose from instead of one to fix?

That variation is doing real work, so let’s slow down on it.

Tips & Tricks Ask for two versions. When you’re not sure what you want, don’t describe it into existence. Ask the AI for options. “Give me two versions, one brief and one warmer,” or “three subject lines, ranging from plain to playful.” Choosing is faster and more honest than specifying, because you recognize the right tone the second you see it, even when you couldn’t have named it in advance. This is one of the moves on your downloadable Prompt Cheat Sheet. Keep it nearby.

Tone, length, and reading level are dials you control

The same content can arrive as a curt three-line email or a gracious two-paragraph note. You decide which, and you can change your mind after you see it.

Tone is a dial. One word turns it: warm, formal, brief, firm, apologetic, upbeat. If a draft feels stiff, you don’t rewrite it yourself. You say “warmer” and read the next version.

Length is a dial. “Cut this in half.” “Give me a one-line version for a text and a longer one for email.” Beginners often accept whatever length the AI produces, as if it were fixed. It isn’t. You set it.

Reading level is the dial people forget, and it might be the most useful of the three. You can ask for writing a specific audience will actually absorb. A note going home to families in a school where many parents are still learning English should not read like a legal memo. Ask for it plainly: “Rewrite this so a busy parent can understand it in one quick read. Short sentences. No jargon.” The same trick works in reverse when you need to sound more polished for a formal request.

Think About It Picture two real audiences in your life who could not be more different: say, your closest friend and someone official you’re a little nervous to email. If you had to send them the same basic news, what would change between the two messages? Name three specific things. Those three things are exactly what you’ll hand the AI as instructions.

Summarizing without losing what matters

The other half of “documents” is the incoming tide: the long email thread, the twelve-page policy, the forty-message group chat that decided something while you were at work. AI is genuinely good at compressing these. It is also capable of smoothing over the one sentence that mattered most, so summarizing well takes a little care.

The care is this: tell it what you’re trying to keep. A summary made to answer a specific question is far more useful than a generic “summarize this.” If you need to know what you’re on the hook for, say so.

Try It Now Find something real and too long to reread: a lengthy email chain, an article you saved, a chapter of a rental agreement or handbook. Paste it into your AI tool with this:

“Summarize this in about five sentences for someone who has to act on it. Focus on any decisions made, deadlines, and anything I’m being asked to do. If something’s ambiguous or unresolved, say so instead of smoothing it over. Here it is: [paste the text].”

Now the step most people skip, and the one that matters most: open the original and check the summary against it. Did it catch the deadline? Did it miss a “not”? Did it turn a “maybe” into a “yes”? A “not available Tuesday” flipped to “available Tuesday” is the kind of small error that reschedules a real meeting. You are the editor. Reading the summary is not the same as trusting it.

This checking habit isn’t paranoia. It’s the same lesson from Chapter 3: AI writes confidently whether it’s right or wrong, and a summary is a factual claim about another document. If anything in that summary would change a decision, verify it against the source before you rely on it.

Editing like an editor, not a rewriter

Here’s where most people leave value on the table. They read the first draft, don’t quite like it, and either accept it anyway or throw the whole thing out and start over. Both are mistakes. The draft is raw clay, and your job is to shape it with specific feedback.

Vague feedback gets vague results. “Make it better” tells the AI nothing. “The second paragraph is too formal, and you never mentioned the date. Put the date up top and loosen the middle” tells it exactly what to fix. You are pointing at problems, one at a time, the way a good editor marks up a page. Two or three rounds of specific notes will take you further than any single perfect request, because you can’t predict what the draft will get wrong until you see it get it wrong.

This is the “iterate like an editor” move from your cheat sheet, and it’s the habit that separates people who fight their AI tool from people who direct it. The first output is a starting point. Your reactions are the steering wheel.

Keeping your own voice

There’s an honest catch, and skipping it is how people end up sounding like a corporate newsletter. Left alone, AI writing drifts toward smooth, agreeable, and forgettable. It reaches for the phrase everyone uses. It sands off the specific, slightly odd details that make writing sound like a person.

Your specifics are the cure. If you tell it the dog’s name is Biscuit and he hid under the deck during the thunderstorm, that detail survives into the letter and makes it yours. If you give it nothing but “thank neighbor for dog-sitting,” you get a sentence that could have been written for anyone about any dog. Feed it your real examples, your actual phrases, the true small facts. The more of you goes in, the more of you comes out.

A simple test: read the final version out loud. If it doesn’t sound like something you’d actually say, it isn’t done. Tell the AI “this sounds too polished, make it sound more like how I talk,” or just fix that one stiff line yourself. You’re allowed to touch the words. You’re supposed to.

What people get wrong here

The biggest mistake is treating the first draft as the finished product. People paste, copy the reply, and send. They get exactly the generic result the tool defaults to. The value was in the second and third rounds they never did.

The second mistake is the opposite fear: that using AI to write is cheating, or that it will erase their voice. It only erases your voice if you hand it nothing of yourself. Used well, it’s closer to a patient assistant who types fast and never gets tired of your revisions. The words that go out are still yours, because you chose and shaped every one.

The third, and the one with real consequences, is trusting a summary or a draft that contains facts you never checked. AI will confidently state a date, a name, a dollar amount, or a “the policy says you must” that isn’t quite right. For anything that matters, the rule from Chapter 3 stands: you verify. The AI drafts; you’re accountable for what you send.

Your move

Before the next chapter, finish one real piece of writing you’ve been avoiding. Rosa’s van letter, your version of it. Use the notes-to-draft move to start, ask for two versions, then give at least two rounds of specific editing feedback before you accept anything. Read it out loud. When it finally sounds like you, send it. Notice how long the whole thing took compared to the twenty minutes you used to lose to the blank screen.

This chapter was developed with AI assistance and reviewed by a human editor. It’s educational, fact-checked where applicable, and may contain minor errors. It’s not a substitute for professional advice.