Build a No-Code AI Pipeline That Turns RSS News Into WordPress Posts
How to build an automated AI content pipeline with Make: RSS ingestion, AI drafting, human approval via Telegram, and WordPress publishing.
What is an AI news-to-WordPress pipeline?
An AI news-to-WordPress pipeline is a no-code automation that watches a news source, feeds new stories to an AI model for drafting, and publishes the result to a WordPress site with a human approval step in between. It replaces the manual loop of finding a story, summarizing it, writing it up, and posting it, with a chain of connected modules that run on their own. The version described here uses five modules built in Make, an automation platform, and needs no custom code.
TL;DR
- RSS feeds from almost any news site can trigger an automation the moment a new article publishes, no scraping or API keys required.
- AI drafting modules inside Make connect directly to models like GPT or Claude, or Make’s own built-in AI provider, without a separate API key.
- Strict prompting that tells the model to write only from the supplied source material prevents fabricated quotes or numbers when a story is thin or missing.
- A JSON parsing step breaks the AI’s draft into title and body fields so WordPress can receive structured content instead of a wall of text.
- Posts land as drafts, not live pages, which keeps a human in the loop before anything goes public.
- A Telegram notification with an approve button lets a person greenlight a post from a phone in about ten seconds, then a second small scenario flips the post from draft to published.
- The same trigger can fan out to a newsletter, a social post, or a Slack message through a router, turning one news event into multiple outputs.
Remy doesn't write the code. It manages the agents who do.
Remy runs the project. The specialists do the work. You work with the PM, not the implementers.
How does the pipeline actually work?
The workflow runs as two small scenarios in Make rather than one long chain.
The first scenario handles ingestion and drafting. It watches a specific RSS feed (in the demonstrated build, TechCrunch’s AI feed) at a set interval. When a new article appears, it flows into an AI module with a defined prompt: adopt a specific tone and format, write only from the information given, and shorten or skip a story if the source material is thin. That draft passes to a JSON parser that splits it into structured fields like title and body. Those fields get pushed into WordPress as a new post, but with its status set to draft, meaning nothing is publicly visible yet. The final step in this scenario pings a Telegram chat with the draft’s title, a short excerpt, and an approval link.
The second scenario is much smaller. It’s a webhook that listens continuously for that approval click. When triggered, it filters for approved posts only, then calls the WordPress module again, this time just to flip the post’s status from draft to published.
The two scenarios together took minutes to build, according to the demonstration, not days, and the entire thing runs without any custom scripting.
Why keep a human in the loop?
Letting an AI model publish directly to a live website without review is a risk most site owners shouldn’t take. Models can misread thin source material, restate a rumor as fact, or produce a serviceable but off-tone draft. Building the WordPress step to create a draft, not a published post, means the AI never has unsupervised write access to a public page.
The approval mechanism in this build is a Telegram message with a one-tap link. That single click is the only manual work left in the entire pipeline. Everything else, the fetching, the drafting, the formatting, the categorizing, happens automatically. This is the practical version of “human in the loop”: not constant supervision, but a fast checkpoint at the one moment that matters, publication.
There’s also a real-world argument for this pattern beyond convenience. A recent high-profile incident (an autonomous AI agent running an unsupervised, multi-phase intrusion against Hugging Face’s infrastructure using frontier models) is a reminder that AI systems given too much autonomy and too little oversight can take unexpected, damaging actions at machine speed. A content pipeline is lower stakes than a security breach, but the underlying design principle is the same: give automation the repetitive work, keep a person on the decision that carries risk.
What stops the AI from making things up?
The prompt used in this build includes an explicit instruction: write only from the information provided in the source, and never invent quotes, numbers, or details. If the summary from the RSS feed is thin, the model is told to write a shorter post rather than pad it with invented specifics.
This rule was tested in practice when the model was accidentally handed a blank story. Instead of generating plausible-sounding filler, it declined to write anything, effectively responding that there was no source material to report on. That’s the behavior you want from a model with any kind of access to a live website: a refusal to fabricate is far safer than a confident, wrong answer.
This matters more as these pipelines scale. An automation that runs unattended for weeks needs guardrails baked into the prompt itself, not just good intentions, because nobody is reading every draft as it’s generated. The approval step catches most problems, but a model that already refuses to guess reduces how often a human has to catch a bad draft in the first place.
Is this worth building instead of just using a chatbot?
Using a chatbot means asking a question, reading the answer, and doing something with it yourself, one interaction at a time. A workflow like this does the work continuously, in the background, without needing you to open a chat window. That’s the core distinction between using AI and building with AI: a chatbot answers, a workflow acts.
The specific value here is in the chaining. RSS feeds already exist on nearly every publisher’s site as a machine-readable list of new articles. AI models can already summarize and draft. WordPress already accepts posts through its API. None of these pieces are new. What Make (and tools like it) contribute is the connective tissue: modules that watch, trigger, transform, and hand off data between services without requiring anyone to write integration code.
Whether it’s worth building depends on how often the manual version of this task happens. Someone summarizing one article a week may not need automation. Someone tracking a fast-moving beat, publishing multiple posts a day, or maintaining several content pipelines (news roundups, newsletters, social captions) at once gets a much bigger return, because the fixed cost of building the automation is paid once and the marginal cost of each new post drops close to zero.
How far can this pipeline be extended?
The five-module structure described here (RSS trigger, AI draft, JSON parse, WordPress draft, Telegram approval) is a minimal version. Because Make supports routing, the same triggering event, a new AI news article, can branch into multiple parallel outputs. A router module can send the same story down different paths: one branch turns it into a WordPress post, another formats it as a newsletter blurb, another drafts a social media caption, another posts a summary to a Slack channel for a team. Each branch can have its own AI prompt tuned for that format and its own approval step if needed.
The platform lists thousands of app integrations, which means the same pattern generalizes well beyond news content. Sales lead intake, customer email triage, research aggregation, and similar repetitive, multi-step tasks can be built the same way: a trigger, one or more AI processing steps, a structuring step, and a delivery step, with a human checkpoint wherever the risk of an automated mistake is highest.
Frequently Asked Questions
What is an RSS feed and why does automation use it?
Other agents start typing. Remy starts asking.
Scoping, trade-offs, edge cases — the real work. Before a line of code.
An RSS feed is a machine-readable version of a website’s latest content, usually a list of article titles, links, and summaries that update automatically. Automations can check it on a schedule and detect new articles without scraping the site or needing an API key.
Do I need to know how to code to build this?
No. The pipeline described here uses Make’s visual, drag-and-drop module system. Modules are connected on a canvas and configured through settings panels rather than written in code.
Why does the AI post go to “draft” instead of publishing immediately?
Setting the WordPress post status to draft keeps a human approval step in the loop. It prevents an AI-generated post from going live on a public site without someone reviewing the title, content, and framing first.
What happens if the AI is given a story with no real content?
In testing, when the model received a blank or empty story, it declined to generate an article rather than inventing details. The prompt explicitly instructs the model to write only from the source material provided and to avoid fabricating quotes, numbers, or facts.
Can this same structure be used for things other than news content?
Yes. The pattern of trigger, AI processing, data structuring, and delivery with a human checkpoint applies to many repetitive workflows, including lead handling, customer support triage, research summarization, and multi-channel content distribution.
