Every online community hits the same wall. Members ask questions that were already answered, because the answers are buried in a date-sorted feed. The fix is to organize by problem instead of by date, and you can automate most of that cleanup with a small script.
Here's how I'd build it.
The idea
- Export or copy your past questions (title + link) into a JSON file.
- Map each question to a "problem" bucket using simple keyword rules.
- Generate a Markdown "start here" post grouped by problem.
- Pin it, and re-run the script whenever new answers pile up.
No AI required for version one. Keyword rules are boring, predictable, and easy to debug, which is exactly what you want from a first automation.
The input
A file called questions.json:
[
{"title": "How do I follow up with new leads automatically?", "url": "https://example.com/post/1"},
{"title": "Best way to start a newsletter for my SaaS?", "url": "https://example.com/post/2"},
{"title": "Can I turn my spreadsheet into a weekly report?", "url": "https://example.com/post/3"}
]
The script
import json
PROBLEMS = {
"lead follow-up": ["lead", "follow up", "crm", "booking"],
"email and newsletters": ["email", "newsletter", "inbox"],
"data and reporting": ["spreadsheet", "report", "dashboard"],
"getting started": ["start", "first", "beginner"],
}
def categorize(title):
text = title.lower()
for problem, keywords in PROBLEMS.items():
if any(k in text for k in keywords):
return problem
return "uncategorized"
def build_index(questions):
index = {}
for q in questions:
index.setdefault(categorize(q["title"]), []).append(q)
return index
def to_markdown(index):
lines = ["# Start here: find your problem", ""]
for problem in sorted(index):
lines.append(f"## {problem.title()}")
for q in index[problem]:
lines.append(f"- [{q['title']}]({q['url']})")
lines.append("")
return "\n".join(lines)
if __name__ == "__main__":
with open("questions.json") as f:
questions = json.load(f)
print(to_markdown(build_index(questions)))
Run it with python build_index.py > start_here.md and paste the output into a pinned post.
How it works
PROBLEMS is your whole configuration. Each key is a problem in your members' words; each value is a list of keywords. categorize() returns the first bucket whose keyword appears in the title, so order matters: put the most specific problems first.
build_index() groups questions into buckets, and to_markdown() turns those buckets into headings and links.
Anything that lands in "uncategorized" is a signal. Either you need a new keyword, or you've found a new problem your members care about.
Making it better
- Schedule it. Run it weekly with cron or a GitHub Action and diff the output so you only update the pinned post when something changes.
- Add an LLM fallback. For uncategorized titles only, ask a model to pick one of your existing buckets. Keep the keyword rules as the first pass so results stay predictable and cheap.
- Track gaps. Count questions per bucket. The bucket with the most questions and the fewest good answers is your next lesson.
Why bother
A problem-sorted index gives new members a quick win, cuts repeat questions, and shows you what to create next. It's a tiny automation with an outsized payoff.
I'm building my own community, Automation Entrepreneur, around exactly this kind of practical automation. Join us here: https://www.skool.com/automation-enterenpeur-7666
Top comments (0)