From Links to Launch-Ready Ideas: How I Supercharged Buffer's “Ideas” with a Mobile-First Automation

Why I Built This

I wanted to get consistent on LinkedIn again without adding friction. Buffer's Send to Ideas works great, but the raw inputs felt flat: just a link and a line or two. I needed context-rich ideas—something I could react to in minutes, not hours.

So I built a micro-pipeline that:

• preserves my exact mobile workflow (Share → Shortcut)
• cleans noisy scraped content (no CTAs, auth walls, or counters)
• extracts the original post, separates comments, and enriches with:
• a tight summary
• highlights
• insights from comments (themes/sentiment)
• 5–15 questions to spark a POV post
• pushes everything into Buffer Ideas automatically

What You'll Need

• Apple Shortcuts (iOS/iPadOS) — to grab date/content/URL via Share Sheet
• Google Sheets — 2 tabs
• Dump: date | title | url (“title” currently holds the scraped content blob)
• FFT: date | url | title | summary | clean content | content highlights | clean comments | comment insights | thought-provoking questions
• Zapier — to transform & deliver to Buffer Ideas
• An LLM provider in Zapier (e.g., ChatGPT/OpenAI with gpt-4o-mini)

Architecture at a Glance

Option A (3 Steps): The Fastest Setup

If you have Zapier's ChatGPT: Extract Structured Data action, you can keep it minimal:

Option B (Pro): Bulletproof & Cheap at Scale

If your account doesn't have “Extract Structured Data,” or you want tougher pre-cleaning, use this. It's still mostly no-code.

Zap Steps

(?im)^(?:skip to main content|linkedin|join now|sign in|sign up|log in|open app|read in the substack app|read in app|get the app|view profile|more from this author|more from|explore topics|welcome back|email or phone|password|forgot password?|user agreement|privacy policy|cookie policy|guest controls|community guidelines|your california privacy choices|brand policy|copyright policy|language|start writing|previousnext|discussion about this post|comments? restacks?|restacks?|likes?|^like$|^comment$|^reply$|^share$|see all|see more comments|see more|show more|read more|to view or add a comment, sign in|create your free account|new to linkedin?|continue to join|continue to sign in|subscribe(?: now)?|ready for more?) $

(?im)^(?: *\d[\d,. ] *(?:likes?|restacks?|comments?|reactions?).$| *like *\d+\d+ *$| *\d{1,3}(?:,\d{3}) *)$

function toStr(v){return typeof v==='string'?v:(v==null?'':String(v));}
function unescape(s){return toStr(s).replace(/\\n/g,'\n').replace(/\\r\\n?/g,'\n');}
function clip(s,m=48000){s=toStr(s);return s.length>m?s.slice(0,m):s;}
let obj;try{obj=JSON.parse(inputData.json||'');}catch(e){throw new Error('Invalid JSON: '+e.message);}
return {
  date: toStr(obj.date||inputData.fallback_date||''),
  url: toStr(obj.url||inputData.fallback_url||''),
  title: clip(obj.title||''),
  summary: clip(unescape(obj.summary||'')),
  clean_content: clip(unescape(obj.clean_content||'')),
  content_highlights: clip(unescape(obj.content_highlights||'')),
  clean_comments: clip(unescape(obj.clean_comments||'')),
  comment_insights: clip(unescape(obj.comment_insights||'')),
  thought_provoking_questions: clip(unescape(obj.thought_provoking_questions||''))
};

Unified Prompt

Works whether you use Option A or Option B.

You are an expert content extraction and summarization assistant.

RETURN RULES
- Return ONLY valid JSON. No markdown, no code fences, no commentary.
- All fields must be strings. If information is missing, return "".
- Preserve meaningful emojis and original language.

SCHEMA (strings only)
{
  "date": "",
  "url": "",
  "title": "",
  "summary": "",
  "clean_content": "",
  "content_highlights": "• bullet 1\n• bullet 2\n• ...",
  "clean_comments": "[author]\n[headline]\n[comment]\n\n[author]\n[headline]\n[comment]",
  "comment_insights": "• insight 1\n• insight 2\n• ...",
  "thought_provoking_questions": "Question 1\nQuestion 2\n..."
}

TASK
You will receive pre-cleaned scraped text from a LinkedIn/Substack post that may still mix the original post with comments. Produce the JSON above with:
- "date": echo the provided date.
- "url": echo the provided URL.
- "title": extract if present; otherwise generate a concise factual title.
- "summary": 120–180 words on the ORIGINAL POST ONLY (no comments).
- "clean_content": ONLY the original post text, with remaining platform furniture removed (auth walls, CTAs, footers, reaction widgets, counters, "see more", etc.). No comments here.
- "content_highlights": 5–10 bullets, each starting with "• ".
- "clean_comments": ONLY the comments, formatted EXACTLY as blocks of three lines per comment:
  [author]\n[headline]\n[comment]
  Separate comments with a single blank line (\n\n). If author or headline is unknown, keep the line but leave it empty.
- "comment_insights": 5–8 bullets (each starting with "• ") summarizing themes, sentiment, notable disagreements from comments.
- "thought_provoking_questions": 5–15 questions I (a senior growth marketer in AI/automation for consumer tech & SaaS) could answer as a POV reaction; one question per line; no numbering.

INPUTS
- date: {{date}}
- url: {{url}}
- text: {{cleaned_text}}

Output ONLY the JSON object.

Turning FFT into an RSS Feed (for smarter Buffer suggestions)

In Google Sheets, create a public publish-to-web CSV for the FFT tab or use a simple Apps Script to render RSS XML.

Add columns if you want a neat <title> and <description> (e.g., map title and summary).

Plug the feed into Buffer so its AI nudges reflect what you've been saving.

Simple Apps Script RSS (minimal example):

function doGet(){
  const ss = SpreadsheetApp.openById('YOUR_SHEET_ID');
  const sh = ss.getSheetByName('FFT');
  const values = sh.getDataRange().getValues();
  const [headers, ...rows] = values;
  const idx = Object.fromEntries(headers.map((h,i)=>[h,i]));
  const items = rows.map(r=>({
    title: r[idx['title']]||'',
    link: r[idx['url']]||'',
    description: r[idx['summary']]||'',
    pubDate: new Date(r[idx['date']]||new Date()).toUTCString()
  }));
  const xml = '<?xml version="1.0" encoding="UTF-8"?>\n<rss version="2.0"><channel>\n<title>FFT Ideas</title>\n<link>https://yourdomain.com</link>\n<description>Food for Thought feed</description>\n' + items.map(i=>'<item><title>'+escapeXml(i.title)+'</title><link>'+escapeXml(i.link)+'</link><description>'+escapeXml(i.description)+'</description><pubDate>'+i.pubDate+'</pubDate></item>').join('') + '\n</channel></rss>';
  return ContentService.createTextOutput(xml).setMimeType(ContentService.MimeType.XML);
}
function escapeXml(s){return String(s).replace(/[<>&]/g,c=>({'<':'&lt;','>':'&gt;','&':'&amp;'}[c]));}

Deploy as a web app (Anyone with the link). Use the URL as your RSS endpoint.

Costs, Quality, and Reliability

One LLM call per idea keeps costs low; gpt-4o-mini is inexpensive and solid.

Pre-cleaning reduces tokens and hallucinations.

Add a de-dup step (lookup by url) if you clip the same post twice.

Fail-safe: If JSON parsing fails, retry once with the same prompt but a final line: “If you output anything except valid JSON, fix it and return only valid JSON.”

Troubleshooting

• Model invents names/headlines → Make sure you keep: “If information is missing, return "".”
• Newline rendering in Sheets → Wrap text (Format → Text wrapping → Wrap). If needed, convert literal \n to real newlines in the Code step.
• Weird platform text sneaks in → Drop more terms into the UI regex.
• Zapier “Parse JSON” missing → Use the Code-by-Zapier snippet above (recommended anyway).

What's Next

• Add a secondary Zap that detects certain keywords in content_highlights and auto-creates a Notion/Linear task.
• Build campaign tags (columns) to group ideas by topic.
• Gate higher-cost models (e.g., “Deep Dive” button) only when you choose to expand an idea.

Final Thought

I didn't overhaul my process. I kept my muscle memory intact—Share → Shortcut—and made what lands in Buffer Ideas richer. That tiny upgrade turned random links into launch-ready prompts.

If you want my exact Zap blueprint and prompt, copy from this post and adapt—or ping me and I'll publish a starter kit.