How I Wrote 30 Days of Blog Posts in One Weekend

Not by writing faster. By removing every step that was not writing. Here is the exact pipeline, the prompt pack that runs it, and the honest part about what it does not do.

The claim, in plain wordsBetween Friday night and Sunday afternoon, I went from "this site has 4 posts" to "this site has 30 posts queued for the next month." I did not write 30 posts in a weekend. I wrote 30 drafts in a weekend. Twelve of them are on the site now. The other eighteen are in a folder, scheduled, waiting their turn.That is the honest version. Here is how.The pipelineSix steps. Each one is a prompt. Each prompt takes the output of the previous one and turns it into the input of the next. None of the prompts is doing the writing. The prompts are doing the non-writing work, which is what was eating my evenings.Seed. A one-line idea. (e.g. "why my cron jobs kept double-running") That is the only human input to the whole pipeline. Once you have a seed, the rest is automated.Research prompt. Takes the seed, returns 5-10 sub-questions worth answering. The model does not browse the web; it returns the questions it would ask if it were a person researching the topic. I read the questions, delete the ones I do not care about, add the ones I do.Outline prompt. Takes the seed + the research questions, returns 3-7 sections, each with a one-sentence purpose. I edit this. The model's first outline is usually 60% right. I add, cut, reorder.Voice prompt. Takes 2-3 of my existing posts as samples, returns a "voice anchor". a paragraph of stylistic notes (e.g. "short sentences, parenthetical asides, no em-dashes, no 'in conclusion,' teacherly register"). This gets fed to the next prompt.Draft prompt. Takes the outline + the voice anchor, returns a full draft. One section at a time, run in sequence. (Parallelizing sections breaks the through-line; do not do it.)Edit pass. Takes the draft, returns a revised version. Strips AI tells, kills em-dashes, tightens sentences, checks the lede, verifies the constraints in the original brief.That is the pipeline. Six prompts. About 4 minutes per post on Claude Sonnet, 7 on GPT-4. For 30 posts, that is 2 to 3.5 hours of model time. I spent the rest of the weekend reviewing outputs and tweaking seeds.What the pipeline does for meIt removes the 30 minutes of decision-making I was spending per post. That is the actual benefit. The 30 minutes was not a warmup; it was the friction that kept me from starting. Once the friction was gone, the writing was the easy part.It also removes the consistency problem. Before the pipeline, my posts varied wildly in tone, length, and quality, because I was writing each one from scratch in a different mood. With the pipeline, every post goes through the same voice anchor and the same edit pass. The output is consistent. That is the part that makes a 30-post run possible. without the consistency, the run would be 8 posts of "this is the best thing I have ever written" and 22 posts of "I cannot believe I am publishing this."What it does not doIt does not write in your voice from scratch. You have to feed it 2-3 of your existing posts first. If you do not have any posts, you do not have a voice to anchor to, and the output will be generic-AI prose. The pipeline cannot fix a missing corpus.It does not publish for you. The output is a markdown file. You can publish that markdown file anywhere. I covered this in [the flagship post. WordPress, Ghost, Substack, Medium, a static site generator, a plain HTML file, a PDF. The pipeline does not pick a platform. You do.It does not replace editing. The edit pass catches the obvious AI tells. It does not catch the weird sentence only your spouse would notice. It does not know which of your jokes land and which fall flat. You still have to read the draft.It does not make a bad seed into a good post. If the seed is "AI is changing the world" (vague, unoriginal, no angle), the post will be vague and unoriginal. The pipeline amplifies a seed; it does not invent one. A strong seed is "why my prompts kept failing at work," which is a real problem I had, in a domain I know, with a specific audience in mind. Strong seed, strong post. Weak seed, weak post. The pipeline does not change that.The 30 posts, by topicFor the curious: the 30 posts broke down roughly as 12 on automation (cron, scripts, glue), 10 on prompts and GenAI (the formats, the failure modes, the prompt pack itself), 5 on building this site (the static-vs-WordPress decision, the cost math, the deploy story), and 3 on home automation (the Raspberry Pi piece, the smart-home fallback plan, the security model). The first five are on the site now. The rest are queued.The honest accountingOf the 30 drafts, 5 needed substantial rewriting (e.g. the seed was too thin, the outline was wrong, the voice anchor did not take). 12 needed light editing (a sentence tightened, a paragraph cut, an example added). 13 were publish-as-is, or close enough.That is the part I would not have predicted. I expected to throw away half of them. I threw away maybe 5, and the other 25 were usable in some form. The pipeline's edit pass and voice anchor are doing real work, not just decoration.Try this firstIf you want to see the pipeline in use, this post is one of the 30. So are [the flagship post, [the 3-field fix post, [the cron stack post, and [the JSON vs YAML vs TOML post. If those read like me, the pack is for you. If they read like every other AI-written post on the internet, the pack is not for you, and I would rather you save your $14.If you do want the pack, it is at [brianackerman.gumroad.com/l/blog-writing-factory. Ten prompts, a 600-word README, a short list of what it does not do. Yours for $14, no subscription, updates free when I ship new ones.Get the Blog Writing FactoryOne ZIP. Ten prompts. 600-word README. Yours forever. Updates free when I ship new ones.Buy on Gumroad. $14

Send me the rough edges if you try it. I read every message.