Build a YouTube Transcript Cleaner Workflow in n8n

Raw transcripts are useful, but they are rarely pleasant to read.

They come with filler words, half-finished sentences, repeated phrases, unclear timestamps, and the occasional “what was I even saying here?” moment.

This n8n YouTube transcript cleaner workflow gives you a repeatable way to clean rough transcript text without losing the original meaning.

That last part matters. A transcript cleaner should make the source easier to use. It should not rewrite the speaker into a completely different person or quietly invent cleaner claims than the original supported.

This is Workflow 5 in the GetPrompting Free n8n Workflow Library. It teaches the difference between cleanup and rewriting.

raw transcript -> cleanup rules -> readable Markdown -> review flags

You are cleaning a rough transcript.

Source title:
[TITLE]

Transcript:
[PASTE TRANSCRIPT]

Cleanup level:
[LIGHT CLEANUP, MEDIUM CLEANUP, OR REPURPOSE READY]

Preserve:
- original meaning
- speaker intent
- important names
- technical claims

Return:
- cleaned transcript
- key takeaways
- reusable snippets
- repurposing ideas
- review flags for claims, names, and numbers

What the YouTube Transcript Cleaner Does

The workflow takes source title, source type, transcript text, content goal, audience, cleanup level, and intended output and turns them into a Google Doc with a cleaned transcript, key takeaways, repurposing ideas, cleanup notes, and review flags.

The important part is not that the workflow is complicated. It is that the workflow creates a real document you can review, edit, and use. That is what separates a practical automation from a fun demo.

Why This Workflow Matters

This workflow teaches the difference between cleaning and rewriting. The model should remove friction while preserving meaning, names, claims, and the speaker voice.

This matters because beginners often try to automate the exciting part first. They jump straight to agents, dashboards, and complicated branching logic before the core pattern is reliable. I like starting smaller. Make one useful thing work. Then make it better.

That approach is slower for about five minutes and faster for everything after that. Once the base workflow is understandable, you can change the model, destination, trigger, or output format without rebuilding from scratch.

What You Need Before You Build It

The version I built uses n8n, Ollama, a local chat model, and Google Docs. You can change those pieces later, but this setup makes the workflow easy to inspect and test.

• n8n running locally, self-hosted, or in n8n Cloud
• Ollama running locally if you want the local AI version
• a local chat model such as llama3.1:8b, or another model your machine runs reliably
• a Google account
• Google Docs credentials connected inside n8n

Download the Free n8n Workflow

I published the clean workflow export on GitHub so you can import it, inspect it, and adapt it to your own setup.

Download the YouTube Transcript Cleaner workflow on GitHub

The repo includes the n8n workflow JSON, screenshots, sample input, sample output, installation notes, customization ideas, and troubleshooting docs.

The public export does not include my private credentials, OAuth tokens, workflow IDs, API keys, or account details. After importing it, you will still need to connect your own Google Docs credential inside n8n.

How This n8n YouTube Transcript Cleaner Workflow Works

Here is the practical flow:

Manual Start -> Set Workflow Inputs -> Build AI Prompt -> Generate With AI -> Review AI Output -> Prepare Google Doc -> Create and Write the Google Doc

Let us walk through the main pieces.

1. Manual Start

Manual testing lets you compare the cleaned transcript against the original before trusting the workflow.

2. Set Workflow Inputs

This node stores the transcript, cleanup level, audience, and intended output.

3. Build AI Prompt

The prompt defines cleanup boundaries so the model improves readability without inventing content.

4. Generate With AI

The local model removes filler, improves structure, and extracts takeaways.

5. Review AI Output

This node formats the cleaned result and keeps review flags visible.

6. Prepare Google Doc

The workflow turns the cleaned transcript into a document that can be searched, edited, and reused.

7. Create and Write the Google Doc

The output becomes a practical working document for articles, notes, or social repurposing.

The New Concept This Workflow Teaches

This workflow teaches the difference between cleaning and rewriting. The model should remove friction while preserving meaning, names, claims, and the speaker voice.

That concept is the reason this article exists as its own piece instead of being a copy of the previous workflow guide. Each workflow in the library should add a useful idea you can carry into future builds.

Once you understand this pattern, you can reuse it in other workflows. The exact topic changes, but the habit stays the same: define the input, give the model a clear job, review the output, and send the result somewhere useful.

How to Customize This Workflow

The GitHub version is intentionally simple. That is a feature, not a limitation. A simple workflow is easier to understand, modify, and trust.

Change the Inputs

Open the Set Workflow Inputs node and replace the sample values with your own source title, source type, transcript text, content goal, audience, cleanup level, and intended output. If you use this often, you can replace the manual fields with a form, webhook, Google Sheet row, Obsidian note, or Notion database item.

Change the Model

The default version uses a local Ollama model. Smaller models are usually faster and cheaper to experiment with. Larger models may follow complex instructions better, but they can be slower and more memory hungry.

You can also swap the local model for a cloud model through n8n if the workflow needs stronger reasoning. I would still keep the review step, because better models are not the same thing as perfect models.

Change the Output Destination

Google Docs is a friendly first destination because it is easy to read and edit. But you can point the same pattern to Obsidian, Notion, Airtable, Google Sheets, a local Markdown file, a task manager, or a custom dashboard.

Upgrade It Later

• Add a YouTube transcript extraction step.
• Save cleaned transcripts to a searchable knowledge base.
• Generate short clips or social post ideas from key sections.
• Add a quote-verification checklist before publishing.

Common Mistakes to Avoid

Letting the model rewrite too aggressively

A transcript cleaner should not turn the speaker into someone else.

Trusting names and numbers without checking

Always verify people, tools, prices, dates, and technical claims.

Skipping cleanup levels

A light edit and a repurposed article draft are different jobs. Tell the workflow which one you want.

Where This Fits in a Bigger AI Workflow System

The YouTube Transcript Cleaner is small on purpose, but it fits into a larger practical workflow system. It can sit beside the Daily Action Brief Builder, the Search Intent Blog Outline Builder, and the rest of the free n8n workflow library as one reusable tool in a larger process.

That is the real value of building these workflows one at a time. You are not just collecting templates. You are learning patterns: cleanup, planning, triage, structure, review, repurposing, documentation, and knowledge management.

Those patterns compound. A small workflow that solves one clear problem today can become a building block for a much more useful system later.

Where this fits: transcript cleanup is a useful starter AI workflow because the AI has a focused transformation job and the human review step is obvious.

Final Thoughts

The YouTube Transcript Cleaner is not impressive because it is massive. It is useful because it gives one messy problem a clear path from input to output.

That is the kind of automation worth learning. It respects the human part of the work while using AI to handle the structure, cleanup, and first-pass organization.

If you want to experiment with it, download the free workflow from GitHub, import it into n8n, run the sample input once, and then replace the sample with something from your own work.

Start small. Make it useful. Then improve one piece at a time.

Stay sharp,
Michael
Creator of GetPrompting.com

Keep Building the Workflow Library

This guide is part of the Free n8n Workflow Library, a set of small n8n builds designed to be imported, inspected, and customized one workflow at a time. If you want the previous step in the series, read Prompt Starter Library Builder. The next build is SOP Generator, which adds another practical pattern without turning the system into one giant automation.