Why "Write Me a Blog Post" Gets You Garbage (and the Chain That Doesn't)

"Write me a blog post about roof repair." That's how most people use AI for content. One request, one answer, copy, paste, publish.

And it's why so much AI content is garbage. Not because the AI can't write. Because you asked it to do seven jobs at once, in one breath, with no chance to check its work. The fix is called prompt chaining, and it's simpler than it sounds.

What is prompt chaining?

Prompt chaining means breaking a big task into a series of smaller requests, where the output of each step becomes the input for the next. You check each step before moving on.

It's how you'd manage a person, too. You wouldn't tell a new writer "go write something about water heaters" and wait for the finished piece. You'd agree on the topic, look at an outline, read a draft, and give notes. Chaining just does that with AI.

Anthropic's prompting guide describes the most common version: generate a draft, have the AI review it against criteria, then have it refine. Breaking the work into separate steps also means you can inspect what came out in the middle, before a mistake gets baked into the final piece.

One-shot vs. chained: what's the difference?

"One-shotting" is asking for the finished product in a single request. Here's how the two approaches compare on a typical service-business blog post.

The chained version takes more clicks. It takes a lot less fixing, and it produces something you'd actually put your name on.

Before the chain: give the AI a brain

A chain works best inside an AI project that already knows your business: your voice, services, prices, customer questions, and rules. Without that, every step starts from zero.

I covered how to set that up in stop prompting, start briefing, and what goes into it in the 5 documents your AI needs. Build the folder first. Then run the chain inside it.

How to write blog posts with AI: the 7-step chain

Here's the chain I use. Each step is its own request. You read the output, fix anything off, then feed it into the next step.

1. Step 1

   Research the question

   Start with one question a customer actually asked. Ask the AI to list the related questions people search, what the top results already cover, and what's missing. If your tool can search the web, have it do that and cite sources. Look at the results yourself, too.

2. Step 2

   Build the brief

   Turn the research into a short brief: who the reader is, the one thing they need to walk away knowing, the search phrase you're targeting, and what from your business belongs in it (prices, towns, process). This is the step most people skip, and it's the one that matters most.

3. Step 3

   Outline

   Ask for headings only, written as the questions readers actually type, with one line under each saying what it covers. Cut anything that doesn't help the reader. Reorder. It's ten times easier to fix an outline than a finished draft.

4. Step 4

   Draft section by section

   Write one or two sections per request instead of the whole thing. Each section gets the AI's full attention, and you can catch drift early. Paste the approved outline each time so it stays on track.

5. Step 5

   Edit against the voice doc

   Hand the full draft back and ask the AI to review it against your voice and rules documents: flag banned phrases, filler, vague claims, and anything that doesn't sound like you. Then have it revise. Then you read it.

6. Step 6

   Fact check

   Ask the AI to list every factual claim in the post: prices, numbers, product details, anything about how Google or a manufacturer works. Then verify each one yourself or with sources. Cut what you can't confirm.

7. Step 7

   SEO pass

   Title, meta description, headings that use the search phrase naturally, internal links to your related pages, and a clear next step for the reader. This is where the post gets ready to be found.

This is the same layered approach I described in is AI content killing your SEO. The chain is just the hands-on version of it.

Where few-shot prompting fits in

Few-shot prompting sounds fancy. It just means giving the AI a few examples of what you want, instead of only describing it.

You can use examples at almost every step of the chain:

• Outline step. Show it an outline from a post you liked, so it learns your structure.
• Drafting step. Include two or three paragraphs of your real writing so it picks up your rhythm.
• Editing step. Give it a before and after: a sentence you rewrote and why.
• SEO step. Paste a title and meta description you were happy with as the model to follow.

The best place to keep these examples is in your project, so you're not pasting them every time. That's what the voice document is for.

One caution: pick examples that are different from each other. If all three samples are about furnaces, the AI may start steering every post toward furnaces. Mix topics, lengths, and formats so it learns your voice, not one subject.

Is chaining overkill for a small business?

Fair question. You're running a plumbing company, not a publishing house.

Here's how I'd think about it. A one-shot post takes two minutes to generate and then either goes up as filler that never ranks, or eats an hour while you rewrite it. A chained post takes maybe 30 to 45 minutes of your attention once you've done it a few times, and the result is something a customer would actually read.

And it gets faster. Once you save your chain steps as reusable instructions, each new post is mostly you reading and approving. The thinking you do in Steps 1 and 2 is what makes it yours, and that's the part you shouldn't hand off anyway.

Common mistakes when chaining

• Skipping the brief. Going straight from a topic to an outline is just a slower one-shot.
• Not reading between steps. The whole point is catching problems early. If you approve everything without looking, you've lost the benefit.
• Running it in an empty chat. No business context means generic output at every step.
• Trusting the fact check blindly. The AI can list claims well. Confirming them is your job, or someone's who knows the industry.
• Publishing one post at a time. Once the chain works, batch it. Research ten questions at once, brief ten posts, then draft them. Batching is why I produce social content a quarter at a time, too.

The short version

Don't ask for a blog post. Ask for research, then a brief, then an outline, then sections, then an edit, then a fact check, then an SEO pass. Check each one. Run it all inside a project that knows your business.

It's more steps. It's also the difference between content that exists and content that brings in calls.