The Buried Lede is my media accountability channel. Every story takes one thing a news outlet got wrong, buried or covered for, and puts it next to the primary document that shows what actually happened. There are always two links, what they reported and what actually happened.
It started as a daily written post on Substack, Instagram and TikTok. In September 2026 I launched it on YouTube, which is now where the main work goes.
The first video, on Fox News and Dominion.
Chris Stirewalt called Arizona for Joe Biden on election night 2020, first in the country, and he was right. Two months later Fox News let him go. The network spent weeks airing claims about Dominion voting machines that its own hosts and executives called nonsense in private. In April 2023 it paid $787.5 million to settle before a jury could hear those messages read aloud. Fox never had to correct the record on air.
How it started
I wanted to find out if one person could run a daily news channel with a clear voice and strict sourcing rules, with AI doing the production work. So I built it on n8n, Airtable and the Claude API.
A weekly research workflow reads a set of news feeds and a bank of older stories, checks which series is running, then queues stories in Airtable. The stories are organized into 16 series, from The War Salesmen to The Pension Heist. That part still runs today. Until September a second workflow wrote each story for Substack, Instagram and TikTok, and Flux generated an illustration for it. The writing now runs through the same pipeline as the videos.
Then I sat down for a review session. Every post got checked before it went out. Did the links load? Was the source the actual primary document? Did the key fact match what the document says?
The first version sent me a Telegram message every morning asking GO or SKIP, plus a ping 14 days later asking me to rate the post. I stopped answering the rating ping pretty fast, and about 15 posts sat there unrated. So I killed both. Those calls moved into the review session, where I was looking at real numbers from TikTok and Substack anyway.
Why YouTube
A 400-word post can point at a court filing. A video can put the filing on screen and highlight the line while you hear it read out. That's a lot harder to argue with.
So I built a long-form video pipeline on top of the daily one. YouTube is the main channel now. Substack is the written version, and Instagram and TikTok bring new people in.
How a video gets made
Every video opens on a real person the story hurt, someone named in the public record. Then it shows what the outlet printed, and what the documents say.
The documents are the whole point. Every receipt on screen is a primary source, like a court filing, an official report, a transcript or the outlet's own archived page. I only use a news article about a document when I can't get the document itself. Pages come from archive.org snapshots, so nobody can quietly edit them later.
Once I approve a script, most of the production runs on my own machine. The pipeline:
- captures every receipt and highlights the key line
- generates the narration one line at a time
- pulls illustrations from a library of 42 images made in one house style, rough agitprop with torn paper and film grain
- adds music, sound effects and chapter cards, then renders the video chapter by chapter
- writes the title options, description, chapters, tags, pinned comment and full source list, then sends that copy to Airtable
- cuts two Shorts from each video for Instagram and TikTok
I write and approve every script. I watch every render before it goes up and send notes back. On video 4 the notes included the AI voice getting a name wrong, an illustration showing a man when the story is about a woman, and a glitch effect that looked like a render error. That review step is the part I won't automate.
What's AI and what isn't
The narration is an AI version of my voice for now. The illustrations are AI generated, and they show situations with no recognizable faces. When a real person appears, it's a real photo. The documents are real, and every one is linked under the video. If I get something wrong, I'll correct it on the record.
The plan is to go on camera once the style and the workload settle. If a channel names people, you should be able to see who's making the case.
What I learned
The rules came before the automation. Voice, format, sourcing standards, which outlets are fair game and which topics I won't touch were all decided first. The AI follows those rules. Give an AI a vague brief and a topic list and you get content that sounds like everyone else's.
Receipts have to be the documents themselves. A news story about a court filing is someone else's summary of it, and that's the thing this channel is checking.
Small technical choices save hours. In Airtable I use field IDs instead of column names in every workflow, so renaming a column never breaks anything.
What's next
Nobody gets a pass, left or right. The first videos cover Fox News and Dominion, then the sites that printed Ruby Freeman's name.
Want something like this?
If you run a channel, or an agency with several, and want a setup like this built for you, get in touch. Each channel gets its own database, voice rules, series plan and review step, all running on shared infrastructure. I've already made most of the mistakes, so you don't have to.
Follow The Buried Lede on YouTube, Substack, Instagram and TikTok.