Prompt Chain: Blog Post from Brief

Use this whenever you need a blog post, article, or long-form content that is higher quality than what a single prompt can produce. The chain works because it mirrors how professional content is actually created: plan first, write second, edit third.

Particularly useful for content that will be published under your name or your brand's name, where quality matters more than speed. For throwaway internal drafts or quick social posts, a single prompt is fine. For anything public-facing, this chain pays for itself.

Single prompts overload the model. Asking an LLM to understand a brief, plan the structure, write the content, and self-edit — all in one step — produces mediocre results because the model is optimising for too many objectives simultaneously. Splitting into three steps lets the model focus on one task at a time.

Step 1 (Structure) is the highest-leverage step. A bad outline produces a bad article regardless of how well Step 2 and 3 are written. By separating the outline, you get to review and approve the structure before any writing begins. This is where most AI content goes wrong: the structure is never planned, so the article meanders.

The human review between Step 1 and Step 2 is critical. The chain is designed with a pause point. You review the outline, adjust it, then feed it to Step 2. This is not optional. Skipping the review and auto-chaining all three steps produces only marginally better results than a single prompt.

Step 3 (Edit) applies a different lens. The "senior editor" role activates different evaluation patterns than the writing role. The editing checklist gives the model specific things to look for rather than a vague "improve this." The editor's note forces the model to articulate what changed, which often catches issues the edit itself missed.

The rules in Step 2 prevent the most common drafting failures. "Follow the outline exactly" prevents the model from freelancing. "Keep paragraphs to 3-4 sentences" prevents walls of text. "Do not use filler phrases" catches the obvious AI tells.

Add your style guide to Step 2. If you have brand voice guidelines, banned words, preferred sentence structures, or formatting rules, add them to the Rules section in Step 2. The more specific your rules, the closer the output matches your expectations.

Adjust the editing checklist. The seven criteria provided are general-purpose. Add industry-specific checks: "Does it include at least one data point per section?" or "Is every claim supported by an example?" or "Does it avoid making promises we cannot back up?"

Add a Step 0 for research. If the topic requires current information, prepend a research step: "Research [topic] and produce a brief of key facts, recent developments, and notable perspectives. I will use this to write a content brief." This is especially useful when paired with web search tools.

Simplify for shorter content. For articles under 800 words, you can merge Steps 2 and 3: write and edit in the same prompt, using the editing checklist as a self-review instruction.

This chain requires three separate prompts, which means three separate model calls. It takes longer and costs more than a single prompt. For high-volume, lower-stakes content, the trade-off may not be worth it.

The quality ceiling is still bounded by the quality of the brief you provide. "Write a blog post about AI" will produce a generic outline regardless of how good the chain is. The brief should include: topic, audience, key messages, tone, target length, and ideally a specific angle or thesis.

The human review between Steps 1 and 2 is what makes this chain work. If you auto-chain all three steps without reviewing the outline, you lose most of the quality benefit. Treat Step 1's output as a draft you actively shape, not a finished plan.

Claude: Handles all three steps well. The outline in Step 1 tends to be detailed and well-structured. The editing step is thorough. Claude will sometimes flag issues in the editor's note that it then does not fix in the revised text — review both.

GPT: Produces good outlines but may need the Step 2 rules reinforced. GPT tends to add embellishments not present in the outline. "Follow the outline exactly. Do not add sections or topics not in the outline" may need repeating.

Gemini: Works well for the structure and drafting steps. The editing step may be less aggressive than Claude or GPT — Gemini tends to polish rather than cut. If you want substantive editing, add: "Be aggressive. Cut anything that does not earn its place."