Building an Automated Blog-Writing Pipeline — And What It Actually Produces

A while back, I got curious about a specific question: if you connect a news feed to an AI model and an auto-publish pipeline, what actually comes out the other end? Not in theory — what does the output look like, line by line, and is it something you’d actually want on a blog?

So I built the pipeline. It works, mechanically — a topic feed triggers an AI rewrite, which gets a generated image and publishes itself to WordPress, with no manual step in between. Here’s how it’s put together, and just as importantly, what I learned about where this kind of system is genuinely useful and where it isn’t.

Where the idea came from

Keeping a blog updated regularly is mostly a consistency problem, not a writing problem — the hardest part is often just noticing that something worth writing about happened, and doing something with it before the moment passes. I wanted to see whether a system could handle that noticing part automatically, while leaving room for a person to decide what happens next.

Step 1: Turning a topic into a feed

The first piece is a way to track new articles on a topic without checking manually. Google News makes this easy:

1. Go to news.google.com and search for your topic — in my case, “Artificial Intelligence.”
2. Copy the URL of the search results page.
3. Head to rss.app (free tier works fine), paste that URL, and generate an RSS feed.

What you get is a feed that updates automatically whenever a new article matching your search appears — essentially a structured “new content alert” you can plug into other tools.

Step 2: Where AI comes in — and what “rewriting” actually means

This is the part of the pipeline that needs the most care, so it’s worth slowing down here.

Mechanically, it’s simple: Make.com watches the RSS feed, and whenever a new article appears, it sends that article’s content to an AI model (I used GPT-4 via the OpenAI API) with a prompt asking it to produce a blog-style post based on the article. A second step sends a prompt to an image model (DALL·E 3) to generate a featured image.

Here’s the part that matters more than the setup: “rewriting a news article” and “writing an original post inspired by a news article” are not the same thing, even though they can look similar on the surface. An AI model asked to “rewrite this article” will often produce something that closely follows the structure, facts, and framing of the source — sometimes closely enough that it’s functionally a paraphrase of someone else’s reporting, just with different words. That’s worth being careful about, for two separate reasons.

The first is straightforward: republishing closely paraphrased versions of other outlets’ reporting, at scale, isn’t really “content” in any meaningful sense — it’s reprocessing someone else’s work without adding anything. The second is more practical: search engines have gotten noticeably better at identifying this pattern, and sites that publish large volumes of lightly-reworded existing content tend to get deprioritized, not boosted — which is the opposite of what a lot of “AI content for SEO” advice implies.

The more useful version of this step, in my experience, is prompting the model to use the news article as a starting point for analysis or explanation — “here’s what happened, here’s why it might matter, here’s a question it raises” — rather than asking it to retell the story. That output still needs a human pass before publishing, but it’s a genuinely different kind of writing, not just a reworded version of the source.

Step 3: Publishing automatically

The last piece connects everything to WordPress:

1. Install the Make (Integromat) plugin on your WordPress site.
2. Generate an API key from your WordPress dashboard.
3. Add that key to your Make.com scenario, connecting the AI-generated content (and image) to a “create post” action.

A setting worth using deliberately: posts can be set to publish as drafts rather than going live automatically. Given everything above about what AI output actually needs before it’s publishable, I’d treat the draft setting as the default, not an optional extra.

What the output actually looks like

In practice, the pipeline reliably produces a structurally complete draft — headline, body, image — within a minute of a source article appearing. What it doesn’t reliably produce is something ready to publish as-is. The drafts tend to need editing for: removing leftover phrasing that too closely tracks the source article, checking that any factual claims are actually accurate (AI models will confidently restate things incorrectly), and adjusting the framing so the post reads as commentary rather than a copy of someone else’s reporting.

So the honest description of what this system does is: it turns “noticing something happened” into “having a draft to work from” — automatically, and fast. It doesn’t turn “noticing something happened” into “a published, finished post,” and I don’t think that gap is something a better prompt fully closes.

If you want to try this yourself

I’ve published the full Make.com blueprint — the modules, connections, and prompt structure — here: Explore the full blueprint. It’s the actual workflow, so you can study it, adjust the prompts, or adapt it to a different topic feed entirely.

If you want to go deeper on the AI side specifically, OpenAI’s own documentation is a good reference for how the underlying model works and what prompt parameters are available: OpenAI’s GPT-4 API documentation.

For a related example of a small automated pipeline — this one focused on notifications rather than content — see: Automating Workflows: From Form to Notification in Seconds.

A few honest questions

Does this actually save time? Yes, for the part it’s built for — going from “an article exists” to “a structured draft exists.” It doesn’t save the editing time, and depending on the topic, editing can take longer than writing from scratch would have, if the AI draft tracks the source too closely and needs significant rework to become original.

Is AI-generated content bad for SEO? Not inherently — but content that’s primarily reprocessed from other sources, published at volume, is the kind of pattern search engines are specifically built to deprioritize. The output of a pipeline like this is a draft, and what happens between draft and published post is where that risk gets addressed.

Do I need to know how to code to set this up? No — every piece here (RSS generation, Make.com, the WordPress plugin) is configured through interfaces, not code. The part that takes the most thought isn’t technical at all; it’s the prompt, and what you’re actually asking the AI to do with the source material.

Closing thought

I went into this wanting to know if a blog could “write itself.” The honest answer is: a blog can draft itself, reliably and quickly — but the gap between a draft and something worth publishing turned out to be less about AI capability and more about a fairly specific judgment call: whether the output is about something, or just a reworded version of something. That distinction doesn’t show up in a demo, but it’s the part that actually matters.