RSS Auto-Posting: When It Works and When It Just Spams

Every RSS-to-social setup starts the same way. Someone wires a feed to an account, watches the first three items go out correctly, and feels like they have found free content. Two weeks later the account reads like a machine, and the thing everyone remembers is the Tuesday the CMS re-indexed and the feed dumped forty items into the timeline in about ninety seconds.

The mechanism is not the problem. Feeds are structured, reliable, free and older than every network you are posting to. The problem is that a feed item and a post are different objects, and most setups treat them as the same object under a different name. Everything below is about closing that gap.

Three Setups That Actually Work

Feed automation earns its keep in a narrow band: predictable, reasonably high-volume content from a feed you control, where being early matters more than being clever. Three patterns fit that description.

Your own publishing, announced once

You publish, the feed fires, an announcement goes out. This works because you own the feed, the timing and the source text, and because the message is genuinely useful to someone who follows you. The mistake is treating it as your only announcement. One automated post at publish time, plus a human writing the real hook two days later, beats two automated posts every single time.

Releases and changelogs

A release feed is the best RSS automation available. GitHub exposes one per repository at the releases.atom URL, a release is an unambiguous event with a version number attached, and the people who want that in their timeline are exactly the people who follow a project account. It is also the one case where robotic phrasing is completely fine. Nobody wants personality in a changelog announcement. They want the version, the highlights and a link.

Curated reading, with a person in the middle

Pull twenty industry feeds into a reader, let automation collect, dedupe and format the candidates, and have a human pick three a week. The machine does the boring half, which is watching. The person does the half that needs taste, which is deciding whether an item is worth your audience's attention. This is the highest value pattern of the three, and the first one people abandon, because it still asks for five minutes a week.

Filtering Is Most of the Job

The naive setup posts every item. The working setup rejects most of them. These are the filters worth having, roughly ordered by how much pain each one saves.

• Deduplicate on the GUID, never on the title or the link. Titles get edited after publishing and links collect tracking parameters, and either will happily re-post the same article as a brand new item.
• Ignore anything published before the moment you switched the automation on. This one rule prevents the classic first-run disaster where a five-year archive lands on your timeline at once.
• Cap items per run and per day. If a run finds twelve new items, something upstream is wrong. Post one, log the rest, and go look at it.
• Allow and deny lists on keywords, categories or author. Most feeds carry more than one kind of content: a company blog feed is usually announcements, engineering posts and job listings mixed together, and you probably want one of those three.
• Delay an hour or two after publication. It costs nothing, and it means typos, wrong titles and accidental publishes get fixed before you broadcast them.
• Require the fields you use to actually exist. No title or no link means no post. Feeds break in dull ways, and empty items are the dullest and most common.

The age filter and the per-run cap are the two that prevent public embarrassment. The rest only improve quality. If you implement nothing else from this article, implement those two, because they are the difference between a quiet automation and a story your colleagues retell.

A Feed Item Is Not a Post

The default template almost everywhere is title plus link. That is a headline, not a post, and it is the main reason automated accounts read as automated. What the item actually hands you is raw material: a title written to win a page of search results, a description field that might be a summary, might be the entire article and might be empty, all wrapped in HTML that arrives in a social box as literal tags and half-decoded entities.

So formatting is real work, and it is the step where a language model earns its place in the pipeline far more than it does in the writing. Strip the markup and decode the entities. Turn the description into a first line a human would read, or write a generic one that at least parses. Decide once, per network, what happens to links and images instead of sending one identical string everywhere. And skip the hashtag generator completely: nothing announces an unattended account faster than five auto-generated tags stapled to every post.

Automation should decide when something posts. It should almost never decide whether.

Rate and Rhythm

A feed has no sense of occasion. It does not know that your account posts twice a day, or that a person is reading a timeline rather than consuming a stream. The automation has to supply the rhythm the feed lacks, and the shape of that is always the same: collect continuously, publish on a drip.

In practice the feed writes to a queue, and something else empties that queue on a schedule you chose. One item every couple of hours inside working hours, nothing at 3am, nothing in a burst. That split gives you a natural cutoff too. If the queue holds thirty items, that is not a reason to post thirty times. It is a reason to post the best one and drop the rest.

The Review Gate

The highest leverage change available to any feed setup is small: create drafts instead of publishing. Two obvious ways to wire it. The n8n node (n8n-nodes-postqueen) reads the feed and creates the posts, or a short script does the same over the postqueen CLI or the REST API. Either way, point the final step at a draft, and reviewing the calendar once a day costs about two minutes.

# One item every two hours, weekdays only. Never a burst.
# crontab -e
0 9-17/2 * * 1-5 /srv/feeds/drip.sh
 
# Inside drip.sh: take the oldest unposted item,
# format it, and queue it as a draft for review
postqueen integrations:list
postqueen posts:create --json /srv/feeds/next-item.json

Those two minutes convert the entire failure mode. An unreviewed feed will eventually post a competitor's marketing, a sponsored item you did not spot, or an article whose headline aged badly in the four hours between publication and your queue. A reviewed feed never does any of that. If even two minutes is too much, hand the review to an agent over MCP with an explicit instruction to publish nothing on its own: it can watch, filter, format and stage all day, and the approval still belongs to you.

When Not to Automate a Feed at All

Some cases are not worth wiring, and recognizing them early saves you from building something you switch off in a month.

• Low-volume accounts. If you publish twice a month, the automation saves two minutes a month and costs an afternoon to build.
• Feeds you do not control. A third-party feed can change format, get acquired, start carrying sponsored items or begin re-publishing old posts with fresh dates, and your account carries the consequence.
• Aggregator feeds. Automating an aggregator turns you into a slower copy of it, and nobody follows a mirror.
• Anything where a headline can age badly. News is the obvious case: a story that reads fine at publication can read very differently four hours later, and your queue does not read the news.
• Accounts whose entire value is that a person is behind them. Automating those trades trust for time, at a bad rate.

Which leaves an honest summary: RSS automation is a distribution mechanism, not a content strategy. It is very good at making sure the thing you already made gets announced, on time, without anyone remembering to do it. It is very bad at deciding what deserves attention. Wire it for the first job, keep a person on the second, and the feed becomes a quiet and useful piece of your stack instead of the reason people muted you.