Back to Projects

Voice-to-Multi-Platform Content Pipeline

An n8n automation that converts a 60-second voice note into LinkedIn, Facebook, and X drafts via Telegram → Whisper → Claude → Typefully.

2026 Automation Engineer Automation

n8nTelegram BotWhisperClaude APITypefullyMeta Graph API

taimoorakhtar.com/projects/content-automation-pipeline

Voice-to-Multi-Platform Content Pipeline

Introduction

Most operators have ideas walking the dog, driving, or right before sleep — moments where opening a laptop and writing a LinkedIn post isn't realistic. I built a pipeline that turns those voice moments into platform-ready content in under 90 seconds, with a human approval gate before anything publishes.

The Challenge

Voice-to-text dictation alone doesn't produce social-ready content — it produces transcripts. And consumer AI rewriters spit out generic, Hormozi-cosplay drivel with em-dashes and 'It's not just X — it's Y' patterns. The challenge was building a pipeline that preserved the operator's actual voice, refused to hallucinate, and produced platform-formatted output that didn't sound like every other AI-generated post.

The Solution

Built a 7-node n8n workflow: Telegram bot captures voice → Whisper transcribes with forced English → Claude rewrites against a strict prompt with banned phrases, vantage markers, and a DM-worthiness test → output lands in a Google Sheet for approval → on approval, Typefully publishes to LinkedIn and Meta Graph API publishes to Facebook.

Technical Deep Dive

Telegram capture node. Custom bot (@DrovinIdeasBot) receives voice messages. n8n uses a separate Get File node to fetch the .ogg payload via Telegram's file API — voice files don't come through in the main update payload.

Whisper transcription. OpenAI Whisper API call with language='en' forced. Without forcing, English-with-Urdu-accent input occasionally gets transcribed as Urdu. Forced language solved this.

Claude rewrite with strict prompt. Custom system prompt enforcing: no Hormozi-style phrases, no 'It's not just X — it's Y', concrete vantage markers required, hallucination refusal protocol, platform-specific length targets (LI 600-1300, FB 250-500, IG 100-250).

Approval gate in Google Sheets. Output writes to a single row with columns for each platform variant + an approval checkbox. Nothing publishes until I check the box. Catches misfires before they hit my audience.

Multi-platform dispatch. Typefully for LinkedIn (auto-publish with social_set_id 301702), Meta Graph API for Facebook Page (HTTP Request with query parameters, not form-urlencoded body). Instagram deferred per Meta's use-case restrictions.

Key Features

60-second voice note → publishable draft in 90 seconds
Banned-phrase enforcement keeps output un-AI-sounding
Vantage marker requirement forces concrete specifics
Approval gate prevents hallucinated content from publishing
Cross-platform formatting tailored per channel

Results & Impact

Personal posting cadence committed to 3-5 posts/week sustainably
Voice-to-publish cycle compressed from ~25 min to ~3 min
Zero hallucinated facts published — approval gate caught all misfires
Reusable template for client deployments at $600+ starting price

Lessons Learned

"Voice is the ideal input modality for content — typing introduces friction that kills ideation."

"AI rewriters need explicit anti-pattern lists, not just style guides. Tell the model what NOT to do."

"Always build a human approval gate when AI is touching your public-facing channels."

Related Work

Have a similar build in mind?

I'm available for engagement on GoHighLevel implementations, A2P 10DLC compliance, AI automation pipelines, and CRM migrations. Most projects start at $300–$1,200 depending on scope.

Book a 30-min call WhatsApp View More Projects