Skip to content
All posts

When Should You Ask AI for an Outline, and When Should You Ask for a First Draft?

By Emily Kirschenbaum7 min read

An outline, a set of questions, or a full first draft? What to ask AI for depends on how much you already know about what you want to say.

An outline with two headings bracketed together in red pen and a question mark, chat bubbles of a question and an answer, and a dotted arrow leading to a full draft page with one sentence underlined in red

One of my favorite questions to ask AI when I’m working on an outline is, “what’s the difference between these two sections?” Usually I’m asking because I’ve read both headings a couple of times and I’m fairly sure I’d end up writing the same paragraph under each.

That’s a much easier fix when all I have in front of me is an outline (and one I spent 30 seconds prompting AI to write so it’s not like I’ve spent a lot of time on it yet). It’s considerably more annoying when I’ve already written 1500 words and find myself spending 20 minutes trying to save a sentence I like in a section I definitely don’t even need.

This is a decent illustration of how I decide what to ask AI for and how I work with it. If I’m still figuring out which ideas belong in a piece and how one leads to the next, I want an outline I can talk through. If I can already explain the point I want to make, the example I’d use, and why it matters to the person reading, I might be comfortable asking for a first draft or at least something more than just an outline.

If I have a topic and a vague feeling that I have something to say about it, that’s when the fun begins and I ask AI to ask me a few questions.

What to Look for in an Outline

Imagine you’re writing an article about a new reporting feature (fascinating, I know). You ask ChatGPT to give you an outline and it comes back with sections called “The Challenges of Manual Reporting,” “Why Reporting Takes So Long,” and “How Automated Reporting Saves Time.” Sounds like a reasonable progression, sure. It goes from problem to explanation to solution - that’s a solid standard B2B blog, right?

Maybe. But now try putting one example under each heading. For the first - the challenge - you might describe someone reporting data from three systems and combining it into a spreadsheet. Under the second, you’ll explain how long it takes to export data from the three systems and combine it in a spreadsheet. You can probably guess what this imaginary person is delighted to no longer be doing by the time you get to section three.

Here’s where it’s helpful to talk through the outline with ChatGPT much as you would with another person. At this point, I would say (or type) something like “I think we’ve covered the problem by the end of the first section. What would someone who actually prepares this report want to know next?” It could be whether the new feature can handle the adjustments made by hand, or what happens when figures from two systems disagree. Either of those gives the article somewhere more interesting to go. The choice depends on what the product does and what its users care about, which is a decision I need to make, not ChatGPT.

I also use an outline to check the order of my thoughts. I’m not the most organized of people and sometimes neither are my thoughts. It’s always possible that a detail I’ve put halfway down is the very thing a reader needs to understand at the start. Going back to our example, let’s say the whole reason this person was combining the three reports to begin with was to present to the head of finance for some specific purpose that resonates with whoever is going to be reading this imaginary blog post. That reader would want to know that before reading several paragraphs about spreadsheets because this information would make it instantly clear how relevant this article is to them or not.

One question I find helpful is to ask “by the time the reader reaches this section, what do they already know, and what are we assuming they know?” Sometimes we’ve explained a point twice and sometimes we’ve explained it exactly zero times because everyone inside the company knows it so well. Either way, it’s good to catch before writing the whole draft.

The Value of Getting AI to Ask Questions

One of the easiest things to forget when giving AI a brief is how much you’ve left in your own head. You know which customer conversation prompted the article and you remember the part of the product demo that people had questions about. You probably even have a strong opinion on what other companies get wrong when they talk about this subject. Despite all of this, the prompt you’ve given AI is “Write a blog post about our new reporting feature.”

AI can’t read our minds (yet). What it can do is ask the questions that get us to formulate the ideas that are floating around in our heads that we can’t quite grasp. If I was writing the blog in our example, I would try a prompt along these lines: “I’m writing for the person who prepares this report every week. Ask me a few questions, one at a time, about how they do it today and what changes for them when they use the new feature.”

The “one at a time” part is useful because this way the AI can base subsequent questions on your previous answers. For example, the first question might be what takes the longest about preparing the report. You answer that it’s pulling together all the numbers. But then its next question is what does the person do once they have the numbers and that triggers you to remember something you heard in a customer interview. The exports themselves only take a few minutes, but what takes so long is figuring out why two totals that should match are so often slightly different.

That changes the whole article. What started as a broad claim about saving time has turned into a specific problem to be investigated. Does the feature help uncover the discrepancy or does it prevent it from happening in the first place? Or does it just put the numbers in one place, leaving someone to do the same detective work as before. Whatever the answer is has given you a clear direction for the article.

When to Ask AI for a First Draft

This may be unpopular, but I’m going to admit it. There are situations in which I do ask AI to write a complete first draft. Sometimes I have all the ingredients. I’ve spoken to the subject matter expert, I have great notes, and I know the specific details that make the story worth telling that I want to build it around. At that point, I’m happy to ask AI for a first draft after sharing all that information with it. There’s value in getting the material onto the page in a form I can then react to.

Sticking with the same reporting feature article example, this is the kind of prompt I would use:

“Write a first draft for a finance operations manager who reconciles this report every week. Open with the customer’s description in my interview notes. Use the product walkthrough to explain which steps change and which still need someone's judgment. Flag anything you're not sure about and need me to confirm.”

That prompt gives AI something substantial to start from, and it gives me a benchmark for checking what comes back. I can check whether the opening captures the customer’s frustration, if I can follow the explanation, and if someone who does this work would recognize themselves in the article.

I do have a firm line when it comes to asking AI for a draft though. I would never ever give it a topic, a target word count, and a few tone guidelines, and then publish whatever it gives me. No matter how “warm, conversational, and authoritative” I tell it to sound, a vague prompt doesn’t give AI any information on what the company can offer or why someone should listen to you on this particular subject.

How to Tell if the Draft is Good

The sentences to look at most closely in an AI draft are the ones that read perfectly well at first glance. A line like “Finance teams can finally trust their weekly reports” sounds like a plausible payoff. But, did the customer ever say they distrusted the report? Or did they actually trust it precisely because someone was checking the figures and that manual process was the problem in the first place!

I use this as a test of whether I’m ready to ask for a draft - do I know enough about this subject and what I want it to say to spot the sentence that sounds good but gets it slightly wrong? If yes, then I can edit that draft with confidence.

Final Words of Wisdom

The best thing about AI tools is that we can all choose how and when to use them. For me, sometimes that means talking through an outline with AI and then writing the whole piece myself. Other times the material is clear enough for me to ask it for a draft.

What I try to pay attention to is where I’m stuck and where it can be of most use. If two sections keep collapsing into each other, I can work that out in an outline. If I’ve got the story and need help getting it onto the page, a draft gives me somewhere to start.

And if I’m staring at a topic thinking that I know I have an opinion on this but I can’t formulate it, I’ll start with the simplest request of all. I ask AI to ask me a question and I’m often a little surprised by my own answer.

(Want to set up a way for AI to help you with blog posts or create content? I can help!)

Clarity first. Strategy follows.

Tell me what you’re working on, what’s challenging you, or where your content could be doing more. We’ll start from there.