OpenClaw YouTube Content Pipeline: Hourly Idea Pitches, Deduped Research, and Instant Outlines

If you publish on YouTube consistently, you know the hidden cost is not editing. It is keeping your idea pipeline full.

You are constantly answering questions like:

• What is breaking right now in my niche?
• Did I already cover this topic last month?
• Is this idea truly new, or just a remix of the same story?
• Can I turn a link I found into a publishable outline in 20 minutes?

This OpenClaw use case is a creator-grade pipeline that automates the scouting and research phase.

You get:

• Hourly idea pitches sent to Telegram
• A 90 day “covered topics” catalog to avoid duplicates
• A pitch database with semantic deduplication via embeddings
• A link-to-outline workflow that turns a Slack message into an Asana card with sources and structure

It is not “generate random ideas”. It is a repeatable system that makes your channel feel like it has a researcher.

Internal references (useful later):

• Use cases overview: /blog/openclaw-use-cases
• Skills guide: /blog/openclaw-skills-guide

---

What you are building

The pipeline in one sentence

Every hour, OpenClaw scans the web and X for relevant news, checks what you already covered, dedupes ideas against past pitches, then posts only novel ideas to your Telegram “video ideas” topic. When you share a link in Slack, it generates research and creates an Asana task with a ready-to-record outline.

Components

1. Idea discovery

   • Web search for breaking stories
   • X/Twitter search for early signals and angles

2. Anti-duplicate layer

   • 90 day YouTube catalog: titles, topics, view counts, notes
   • Pitch database with embeddings for semantic similarity checks

3. Distribution and execution

   • Telegram for fast review
   • Slack trigger for “turn this link into an outline”
   • Asana (or Todoist) for your production board

---

Skills and tools you need

This flow is flexible. You can start small and add depth.

Required

• OpenClaw built-ins

  • web_search for web scanning

• X research

  • x-research-v2 (or any X search skill you trust)

• Task system

  • Asana integration (recommended) or Todoist

• Telegram

  • A chat or topic dedicated to “video ideas”

Strongly recommended

• Knowledge base / RAG

  • A knowledge-base skill so OpenClaw can pull context from your notes, past scripts, and research docs

• YouTube Analytics access

  • gog CLI for extracting last 90 day performance and video titles

Nice to have

• SQLite tooling (local file is enough)
• Slack integration or any webhook that captures links you share

---

Setup: step by step

Step 1: Create a Telegram destination

Create a Telegram group (or a forum topic) named something like:

• “Video ideas (OpenClaw)”

Then tell OpenClaw where to post.

Prompt example:

Create a Telegram destination called "Video ideas".
When you pitch ideas, always post:
- Title
- One-line hook
- Why now (recency)
- 3 bullet angles
- Sources (links)
- Confidence score (0-100)

Remember this Telegram destination for future runs.

Step 2: Decide your channels and niche filters

Write down:

• Your niche boundaries (what you will not cover)
• Your preferred formats (news reaction, deep dive, tutorial, contrarian take)
• Your posting cadence

This becomes a scoring rubric.

Prompt example:

My channel is about: practical AI for solopreneurs.
Exclude: crypto speculation, meme coins, celebrity drama.
Prefer: hands-on tools, launches with real user impact, and workflows.
My video formats:
1) 8-12 min tutorial
2) 5 min news reaction
3) 15-20 min deep dive (max 1 per week)
Create a scoring rubric and store it in memory as "youtube-pitch-rules".

Step 3: Build the pitch database (SQLite)

Create a SQLite database file. One table is enough to start.

CREATE TABLE IF NOT EXISTS pitches (
  id INTEGER PRIMARY KEY,
  timestamp TEXT NOT NULL,
  topic TEXT NOT NULL,
  hook TEXT,
  embedding BLOB,
  sources TEXT,
  score INTEGER,
  status TEXT DEFAULT 'new'
);

CREATE INDEX IF NOT EXISTS idx_pitches_timestamp ON pitches(timestamp);

Notes:

• embedding can be stored as bytes or JSON, depending on your embedding provider.
• sources can be a JSON string.

Step 4: Create a 90 day YouTube catalog

The goal is simple: when OpenClaw finds a candidate idea, it should check your recent uploads so you do not repeat yourself.

If you have YouTube Analytics access via gog, ask OpenClaw to export your last 90 days of uploads:

• Video title
• Upload date
• Views
• Topic tags (if you maintain them)

Prompt example:

Using YouTube Analytics via gog, export my uploads from the last 90 days.
Create a local file called youtube-catalog-90d.json with fields:
- videoId
- title
- publishedAt
- views
- inferredTopics (array)

Update this file daily at 7:30am.

If you do not have analytics access, start with a manual list in a Markdown file and improve later.

Step 5: Define the hourly scout job

This is your always-on radar.

Prompt example (hourly):

Run an hourly scheduled job.

Goal: pitch 1-3 novel YouTube video ideas.

Steps:
1) Search the web for breaking news in my niche.
2) Search X/Twitter for discussions and early signals.
3) Load my youtube-catalog-90d.json and reject ideas that overlap heavily with recent uploads.
4) Check semantic similarity against past pitches in my SQLite pitches table.
   - If similarity > 0.85, reject as duplicate.
5) For each remaining candidate, compute a score using "youtube-pitch-rules".
6) Post the top ideas to Telegram "Video ideas" with sources.
7) Save every pitched idea to the SQLite database, including embedding, sources, and score.

Rules:
- Do not pitch the same story twice.
- Prefer primary sources.
- If sources are paywalled, find an alternative reference.

Step 6: Add the Slack link-to-outline trigger

This is how you go from “interesting link” to “production-ready task”.

Define a channel (example: #ai_trends) where you drop links.

Prompt example (Slack to Asana):

When I share a link in Slack channel #ai_trends, do this automatically:

1) Research the link:
   - summarize the core claim
   - extract key facts and numbers
   - list counterpoints and risks
2) Search X/Twitter for:
   - top posts about the link
   - experts disagreeing
   - alternative angles
3) Query my knowledge base for related notes and past scripts.
4) Create an Asana task in project "Video Pipeline" with:
   - title
   - description
   - bullet outline (hook, setup, main points, call to action)
   - list of sources
   - suggested thumbnail text (3 variants)

Also add a checklist to the task:
- Confirm idea novelty
- Capture 3 supporting screenshots
- Draft intro in first-person

---

Practical prompt patterns that work

1) The “one idea is enough” prompt

When you feel overwhelmed, constrain output.

Give me exactly 1 idea.
It must be:
- novel vs last 90 days
- backed by at least 2 sources
- explained in a 15 second hook
Provide a recommended title and 3 thumbnail text options.

2) The “angle bank” prompt

Sometimes the story is obvious, but the angle is not.

Here is a link.
Generate 10 unique angles for a YouTube video.
At least 3 angles must be contrarian.
At least 3 must be tutorial-style (do X with Y).
Do not repeat wording across angles.

3) The “thumbnail first” prompt

Based on this topic, propose:
- 5 thumbnail concepts (visual idea + 2-4 words text)
- 5 title variations
Make sure each pair (title + thumbnail) tells a clear story in under 2 seconds.

---

Suggested data model and dedup strategy

Semantic deduplication is what turns this from “spammy idea bot” into a reliable pipeline.

How dedup works (simple version)

• Compute an embedding for each pitch using a stable model.
• Compare with existing embeddings in your pitch database.
• Reject anything above a similarity threshold.

Recommended fields to store per pitch

• topic (short)
• hook (one sentence)
• angleTags (array)
• sources (array of URLs)
• score (0-100)
• similarToPitchId (optional)
• status (new, accepted, rejected, produced)

Thresholds

Start with:

• Similarity > 0.85: reject
• Similarity 0.75 to 0.85: allow only if angleTags differ significantly

---

Operations: how to use it day to day

Your morning routine (10 minutes)

1. Open Telegram “Video ideas”
2. Mark 1-2 ideas as “accepted” (simple message or reaction)
3. Ask OpenClaw to generate outlines for accepted ideas
4. Record or schedule research tasks in Asana

Your afternoon routine (5 minutes)

• Drop 1-3 links in Slack
• Let OpenClaw generate Asana tasks

Weekly review (30 minutes)

Ask OpenClaw:

Analyze my pitches and production outcomes from the last 14 days.
Show:
- top pitch categories
- which scores predicted performance best
- duplicates that slipped through
- recommendations to improve "youtube-pitch-rules"

---

Common failure modes and fixes

“Ideas are too generic”

Fix: add constraints.

• Require 2 primary sources
• Require a specific audience
• Require a clear on-screen demo

Prompt tweak:

Only pitch ideas that include a concrete demo I can show on screen in under 2 minutes.
If there is no demo, reject the idea.

“Too many ideas”

Fix: limit output per run and raise the score threshold.

Post at most 2 ideas per hour.
Only include ideas with score >= 80.

“It keeps repeating the same theme”

Fix: add diversity rules.

Ensure the top 5 pitches of the day span at least 3 distinct topic clusters.
If not, down-rank repetitive clusters.

---

FAQ

1) Do I need to run this every hour?

No. Hourly is useful for fast-moving niches. Many creators prefer 2 to 4 runs per day (morning, lunch, late afternoon). Start slower and scale up.

2) What if I do not want to use Telegram?

Use any destination that is fast and low-friction for review: Discord channel, email digest, or even a local Markdown file. Telegram is just a good “inbox” for ideas.

3) Is X/Twitter required?

Not strictly. You can scout via web sources only. X is valuable because it surfaces early commentary, hot takes, and experts disagreeing, which often becomes your angle.

4) How do I prevent the pipeline from pitching low-quality rumors?

Add source rules:

• Require at least one primary source (official blog, docs, paper, repo)
• Down-rank screenshots and anonymous posts
• Ask for counterpoints and uncertainty

5) What is the minimum viable version of this pipeline?

MVP:

• Web search + a simple “seen topics” file
• A daily run that posts 3 ideas
• Manual acceptance and manual outlining

Then add the SQLite database and embeddings once the flow is useful.

6) How does this connect to other OpenClaw use cases?

This pipeline pairs well with:

• A multi-agent content system that drafts scripts overnight: /blog/openclaw-use-cases
• A skills-first setup so you can add integrations safely: /blog/openclaw-skills-guide

---

Next steps

If you want, you can extend this use case in two powerful ways:

1. Auto-build a research folder per accepted idea (sources, screenshots, transcript snippets)
2. Add a publishing checklist so the agent can help with scheduling, descriptions, and cross-posting

Once the scaffolding is in place, you stop asking “what should I publish?” and start asking “which of these already-scored opportunities do I want to ship next?”