How to Write a Blog Post With AI: Outline First

You know what you want to say. You've got an idea, maybe some research behind it. So you open a chat assistant and type the most natural sentence in the world.

"Write me a blog post about native plantings for office parks."

Ten seconds later you have 800 words. Clean. Headings. An intro that says the topic is "more important than ever" and a conclusion that starts with "In conclusion." And it's close to useless. Not because the grammar is off. Because nothing in it came from you.

Here's the part people don't expect. The better AI gets at writing, the worse that one-line request gets for you. A weak tool gives you something obviously bad, and you throw it away. A strong tool gives you something that looks finished. Smooth enough that you're tempted to publish it. And then your name is on a piece anyone could have produced with the same ten words.

If you want to write a blog post with AI that actually sounds like you and says something, the fix isn't a cleverer one-liner. It's getting an outline first, arguing with it, and only then asking for prose.

Every gap you leave gets filled with average

A chat assistant writes by predicting what usually comes next, given everything in front of it. When "everything in front of it" is one sentence, the likeliest next words are the most common words ever written on that topic. The model isn't lazy. It's filling the gaps you left with the safest, most typical choice.

Count the gaps in "write me a blog post about native plantings for office parks":

• Who's reading? A property manager? A homeowner? The AI guesses "general audience."
• What should they do after? Call you? Rethink a budget? It guesses "be informed."
• What's the point of view? Smart, overrated, smart only sometimes? It guesses "balanced overview."
• What facts are in play? Whatever it absorbed in training, which may be stale or wrong for your region.
• What do you sound like? Its own house style: polished, upbeat, a little formal.
• How long, what format, where does it run? About 800 words with H2s.

Six gaps, six guesses, one generic post.

Here's a worked example from the course. Marcus runs a landscaping company that maintains small office parks and medical buildings. He'd done real research on whether converting mowed lawn to native plantings saves money: his own job records, a state extension guide, cost comparisons from property managers he works with. His insight was sharp. The savings don't show up in year one. They show up in years three through five, after establishment costs are paid off, and most property managers quit before then.

He typed "Write me a LinkedIn article about native plantings for office parks." The draft opened with "In an era of growing environmental awareness." It listed five benefits. It had no year-three idea and none of his numbers. It ended with "By embracing native plantings, property managers can create a greener future."

Marcus didn't have a writing problem. He had a sequence problem.

Outline-first drafting: approve the structure before any prose

The order I teach is Brief, Outline, Draft, Human Pass. This post is about the middle hinge: never ask AI for prose until you've approved a structure.

Think about a contractor. Nobody pours a foundation and then decides where the rooms go. You agree on the plan, move a wall or two on paper where it costs nothing, then build. Moving a section in an outline takes ten seconds. Moving it in a finished draft means rewriting the transitions on both sides, fixing the intro that promised a different order, and rereading the whole thing.

There's a psychological trap too. Once you see 1,200 smooth words, you start editing sentences instead of questioning the structure. You fix the word "crucial" in paragraph four and miss that your main point doesn't show up until paragraph nine. The outline makes you look at the skeleton while it's still easy to see.

Almost nobody does this, because AI can produce a full draft as fast as an outline, and the draft feels like more progress. It isn't. A full draft on a weak structure is a weak draft with more words.

The outline prompt, and the one rule that matters most

A good outline prompt has five parts:

1. Role and task. "You're helping me outline one piece of content. Don't write it yet."
2. Your brief. Reader, goal, main point, the facts you've verified, angle, format, voice. Paste it in.
3. Source rules. "Use only the facts in my brief. If a section needs a fact I haven't given you, write [NEED FACT: what kind] instead of inventing one."
4. Output shape. "A numbered outline. For each section: working heading, one-line purpose, the fact or insight it uses, rough word count."
5. Check-back. "Then list anything you were unsure about or had to guess."

That [NEED FACT] rule is the one that protects your credibility. Left alone, an AI writer fills any thin spot with something that sounds like a fact. The placeholder gives it a legal way to say "I need more here." Every [NEED FACT] you see is a spot where your draft would otherwise have contained a made-up claim.

When the outline comes back, don't approve it because it looks organized. AI outlines always look organized. Read it like an editor:

• Where's the main point? If it isn't in the first two sections, move it up.
• Does every section earn its place? Would the piece be weaker without it? AI loves a "Background" section that tells readers what they already know.
• Is the angle visible in the structure? If your angle is a timing problem, the outline should be organized around time.
• What will you do with each [NEED FACT]? Find the fact, cut the point, or rewrite it as your own experience and say so.
• Is there a generic section? The classic is "Benefits of X." Cut it or make it specific.

What it looks like in practice

Marcus ran the outline prompt with his brief. He got eight sections back: an intro on "growing interest," what native plantings are, the year-one cost reality, year two, year three and beyond, why managers cancel, how to plan for five years, and a conclusion. Section six had a flag: [NEED FACT: data on how often conversions are cancelled]. The check-back said, "I did not have data on cancellation rates, so section 6 needs a fact or a different approach."

His edits took about six minutes.

He cut "what native plantings are." His readers already know. He cut the "growing interest" intro and put his main point in the opening, using the AI's suggested first line almost as written. He merged year two with the cancellation section, because the year-two complaints and the cancellations are the same story.

And at the [NEED FACT], he made a call only he could make. He didn't go hunting for a national statistic he didn't have. He rewrote the section from his own records: of the conversions he'd been asked to quote over the last few years, he'd seen several cancelled in year two, and here's what they had in common. True, his, and more believable than a number he couldn't source.

Five sections instead of eight. Main point in the first 50 words.

If he'd asked for the full draft straight away, the AI would very likely have filled that gap with something like "studies suggest up to half of conversions are abandoned." And he might not have caught it.

When outline-first is overkill

Not everything needs this. A two-line social reply doesn't need an outline. For a short LinkedIn post, your brief might be four lines and your outline three bullets. That's fine. The point isn't paperwork. It's getting just enough structure that the AI writes your post instead of the average post.

The time math surprises people. The one-liner feels fast because the draft shows up in ten seconds. Then you spend forty minutes rewriting it because nothing in it is right. A short brief and a five-minute outline usually get you a usable draft on the first try. It isn't slower. It moves the thinking to the front, where it's cheap.

One more thing. When a draft isn't working, don't type "make it better." That has the same problem as "write me a blog post." Name the problem. "Paragraph two buries the main point. Move the year-three sentence to the first line." Specific feedback gets specific fixes.

Key takeaways

• A one-line request leaves gaps (reader, goal, point of view, facts, voice, format), and AI fills every gap with average.
• Approve an outline before asking for prose. Structure is cheap to change early and expensive late.
• Tell the AI to write [NEED FACT] instead of inventing a fact. Each flag is a made-up claim you avoided.
• Read the outline like an editor: main point early, every section earns its place, angle visible, no generic sections.
• Don't say "make it better." Name the specific problem.

Frequently asked questions

Why does "write me a blog post" produce generic AI content?

A one-line request leaves the AI to guess the reader, the goal, the point of view, the facts, the voice, and the format. AI fills each gap with the most typical choice, so you get the average post on that topic. Closing those gaps with a short brief and an approved outline before drafting is what makes the output specific to you.

What is outline-first drafting with AI?

Outline-first drafting means you never ask an AI tool for prose until you have reviewed and approved a structure. You ask for an outline only, edit it yourself by cutting, moving, and merging sections, and then ask the AI to draft from the approved outline. Fixing structure in an outline takes seconds; fixing it in a finished draft takes much longer.

How do I stop AI from making up facts in a draft?

Give the AI a list of verified facts and a source rule: use only these facts, and if a section needs one you didn't provide, write [NEED FACT] instead of inventing it. Each placeholder marks a spot where the draft would otherwise contain a made-up claim. You then find the fact, cut the point, or rewrite it as clearly labeled first-hand experience.

Should I have AI write the whole draft at once or section by section?

For short pieces like social posts or brief emails, drafting all at once is fine. For articles over roughly 800 words, drafting one section at a time usually works better: the AI stays focused, later sections don't get thin or repetitive, and you catch voice problems in the first section before they spread through the rest.

What should I check in an AI-generated outline?

Check five things: whether the main point appears in the first two sections, whether every section would be missed if cut, whether your angle shapes the structure, what you will do with each [NEED FACT] flag, and whether any section is generic, such as a list of benefits anyone could write. Approve the outline only after fixing those.

Where this goes next

Outline-first drafting is the second step of the Kickstart Sequence (Brief, Outline, Draft, Human Pass) in Module 4 of my class How to Use AI for Smarter Research & Creative Ideas, which also covers the nine-field content brief and voice notes.

See the class