Posting every day sounds like a writing task. In practice most of the time goes on small decisions: what is worth talking about today, did we already say it last week, when should it go out, and is this draft claiming something we cannot back up? This workflow takes those decisions off the blank page and leaves the final yes or no to a person.
It runs in n8n Community Edition (free, self-hosted) on an ordinary laptop with a small AI model running locally. No cloud AI account, no API key, no subscription. Everything described below was tested on test data on October 26, 2026 and every problem we hit is listed further down.
Who it is for
A social media team or a one-person business that has to post every day and loses time deciding what to post and when.
How it works, step by step
- Reads the log of topics already posted (first run: file missing, handled).
- Pulls a live news RSS feed for the day trends (falls back to a cached list if the feed is down).
- Drops topics already posted (word-overlap check) and ranks the rest by relevance.
- Works out the best posting hour per platform from past engagement (falls back to all-days history when a weekday has too few samples).
- Asks a small local model for a short post, then checks it in code: length, numbers not in the source, hype or guarantee words, product names, invented first-person results, ends with a question.
- If the draft fails it retries once with a stricter prompt, then falls back to a safe template marked NEEDS HUMAN REWRITE.
- Writes an approval queue (LinkedIn, Facebook, Instagram versions with brand hashtags added by code), updates the posted log and writes a run report. Nothing is posted.
The design rule we followed in every workflow of this series: the model only rephrases text, and plain code does everything that can be checked. So topic de-duplication, the best-hour calculation, the length and word checks, and the retry or fallback logic are all code. The model writes one short draft; the code decides whether that draft is allowed to reach a person.
How it runs: the execution strategy
It has two triggers: a schedule (Schedule: weekdays 07:30) and a manual trigger for testing. Each run checks that the local model server is up, pulls a live trend feed, works out the topic and time, drafts, checks and queues the post, then updates its posted log so the next run knows what was already used. If the feed is down it falls back to cached topics and says so in the run report. It has not been activated on a live account; the runs described here were run by hand.
The workflow on the n8n canvas
Screenshots taken in n8n right after a run. A green tick means the step finished, and the number on each line is how many items passed through it. Click an image to see it larger.



What went wrong while building it, in order
This list is copied from the build log kept while the workflow was made.
- Smoke test first: manual trigger > local model > file write ran end to end in about 30 s, so the plumbing (file access limits, local model call) was proven before building.
- First run stopped silently after "Read posted log". Cause: when the file does not exist yet (first ever run) the read node returns zero items, so every later node is skipped with no error. Fix: "Always Output Data" on the read and extract nodes, and the code treats a missing log as first run.
- Output files appeared flat in output/ instead of output/w1_social/. Cause: the write node keeps only the file name. Fix: put the folder in the write path.
- Test T2 (second run) did not skip yesterday's topic. Cause: "Extract From File > JSON" stores the parsed content under the property data, my code read the root. Fix: read x.data. Verified in T6: log loaded (1 topic), 1 duplicate skipped, a different topic chosen.
- Test runner crashed on workflows that end in an error (stop node) because n8n prints text after the JSON. Fixed the runner to parse tolerant.
- T3 simulated feed outage (feed URL pointed at a dead port): node reported ECONNREFUSED, workflow used the cached topics, status flagged in report.
- T4 simulated model server down: preflight failed, workflow stopped with a plain-English message and wrote nothing.
- T5 adversarial angle (asked the model to claim a 30-day test, add numbers and name a product): first draft failed 2 checks (names an AI product; claims an experience we did not have), retry draft passed, status NEEDS APPROVAL. This is the guard that stops invented results.
What we learned
- A first run is a different case from every other run. The very first run has no "already posted" file, and n8n quietly skips every later step when a read returns nothing. The workflow stayed silent until we turned on Always Output Data.
- The guard that matters most is the one against invented results. In the adversarial test we asked for a post claiming a 30-day test with numbers and a product name. The first draft failed two checks, the retry passed, and the status stayed "needs approval".
- A free public feed is enough for trend research, but a feed can fail. The workflow records the error and falls back to cached topics, and the run report says so.
What the finished run produced

This is the real output file from a clean first run, with every saved file removed beforehand. The whole run took 48 seconds on a CPU-only laptop, including about 15 seconds of n8n start-up.
# Run report Topic: The Top 10 AI Tools for Small Business: From Automation to a Coordinated Team of AI Agents Source: live feed Log: posted log missing -> first run, starting a new one Candidates: 100 | duplicates skipped: 0 Model attempts: 1 First-draft problems: none Status: NEEDS APPROVAL ## Scheduled posts - LinkedIn at 2026-10-29 09:00 (Friday history (16 posts)) - Facebook at 2026-10-29 19:00 (Friday history (19 posts)) - Instagram at 2026-10-29 13:00 (Friday history (25 posts)) ## LinkedIn text Small business owners don’t need to build or code anything to use AI helpful ways. They can start with tools that do routine tasks like sorting emails, scheduling meetings, or creating simple text drafts. These tools work like assistants that follow clear instructions given in everyday language. The key is understanding what the business actually needs to get done each day. Instead of searching for the newest tool, ask: what repeats? What takes time? Pick a tool that can handle that specific step without extra setup. Over time, using a few reliable helpers can reduce daily paperwork and free time for other work. When a few simple helpers work together for one routine job, it feels like a small coordinated team doing that one thing well. It keeps things straightforward and avoids overcomplicating. What specific daily task would you like to handle more easily with a simple helper?
Node-by-node trace of that run
| Node | Status | ms | Items out |
|---|---|---|---|
| Manual Trigger | success | 1 | [1] |
| Settings (edit me) | success | 7 | [1] |
| Preflight: local model server | success | 132 | [1] |
| Server up? | success | 9 | [1, 0] |
| Read posted log | success | 2 | [1] |
| Extract posted log | success | 1 | [1] |
| Fetch trend feed | success | 835 | [1] |
| Parse and rank trends | success | 56 | [1] |
| Any topic? | success | 5 | [1, 0] |
| Best posting time | success | 9 | [1] |
| Build prompt | success | 6 | [1] |
| Local model: write post | success | 30230 | [1] |
| Check post (attempt 1) | success | 14 | [1] |
| Passed checks? | success | 4 | [1, 0] |
| Assemble platform versions | success | 10 | [1] |
| Write post queue > to file | success | 14 | [1] |
| Write post queue | success | 23 | [1] |
| Build updated posted log | success | 9 | [1] |
| Write posted log > to file | success | 4 | [1] |
| Write posted log | success | 5 | [1] |
| Run report | success | 8 | [1] |
| Write run report > to file | success | 5 | [1] |
| Write run report | success | 14 | [1] |
Limits, stated plainly
The engagement history in this template is placeholder data, so the "best posting time" is only as good as the history you plug in. Headlines are used as prompts for commentary, not as facts to repeat; read the source before approving a draft. The workflow drafts and queues. It does not post anything. Everything above was tested on a laptop . We make no claim about results on your data.
What it costs
Software: nothing. n8n Community Edition is free to self-host and the AI model runs locally with a free model runner such as Ollama. You need a laptop (8 GB of memory minimum, 16 GB is comfortable, no graphics card), an internet connection for installing, and about 20 minutes the first time. No cloud AI account, no API key and no subscription.
Want the workflow file?
The workflow file and a plain-English guide exist for every workflow in this series. If you would like this one, or you have a repeated task in your own business you would like checked for fit, tell us about it. We will say honestly whether a workflow like this could take it off your plate, and we will not promise results we cannot show.
How this article was produced
This article was drafted with AI assistance from the build log, test traces and output files of a workflow built and run on the site owner's own laptop, then published by the site owner, who is responsible for it. The workflow was built and tested by an AI assistant under the owner's direction. If you find an error, tell us. Our standards are described in the editorial policy.
What is and is not verified
Verified: the build-log events, the node trace and the output shown come from real runs on October 26 and 27, 2026. Not verified: how the workflow behaves on real business data, with other models, or at larger volume. The node diagram is generated from the workflow file and is not an n8n screenshot.
Sources
Related: Search Console Says 43 Pages Are Not Indexed: What We Tested and What We Found and Moving to a custom Blogger theme: 9 errors we hit.
About the author
I'm Junaid. I'm a sales professional with more than seven years across several industries, and I have no coding background. I use AI to build and test tools like this and I write down exactly what happened, mistakes included. If a process in your business is eating your time, tell me about it and we can look at whether AI could take some of the load.