Content Engine

Content Engine brings a creator’s videos, articles and courses into one library. It looks at the questions people actually search for and studies the channel’s thumbnails. An agent can search that library and write from it, with the sources showing.

Here’s the kind of job it’s for. Say a creator has years of videos, articles and a paid course on email marketing, and people keep searching for how to write a welcome email. Content Engine gathers the creator’s existing answer into a research packet, with links back to the video and the relevant moment. It can point to a paid lesson without turning that lesson into public copy; a person decides what gets published.

• Search everything at once: articles and video transcripts, cleaned up and split into timestamped sections.
• Start from real questions: actual search questions are joined to the content that answers them. What was observed is kept separate from what the source says.
• Research packets with sources: every packet points back to where each fact came from.
• Agents can use it directly: it offers seven tools over MCP, so an AI agent can search, pull a packet, and save a draft that's logged.
• Paid stays paid: search applies visibility rules, so paid lessons aren't served up as public posts.
• One piece becomes many: one source-backed master draft can become up to eight related pieces, and the citations carry forward.
• It stages; people publish. Outside connections are for WordPress drafts and YouTube titles, descriptions and tags—not automatic publication. Without the required credentials, staging returns "not configured" and changes nothing outside the app.

Coming next

• Video frames. Content Engine reads what a video says (the transcript, with timestamps) and studies its thumbnails. Pulling frames from inside a video is the job of its video-specific part, Video Evidence Engine. A recorded demo of the two working together doesn't exist yet.
• Outside staging connections still need credentials and a live connection check. Publishing remains a separate step outside Content Engine.

Since 2003, I’ve made more than 450 courses on email marketing, websites, WordPress, YouTube, Facebook and other social media, sales funnels, one-time offers, upsells and conversion optimization. I also built campaign tracking with coded landing pages and individual tracking links, so I could see how people moved from one step to the next. Content Engine carries that same habit into agent work: start with what a creator already made, look at what people are asking, and keep every claim tied to its source.

The private example is Watercolor Misfit: bringing its existing lessons and articles together so an agent can find the answer without losing the source. My part is the system behind that workflow.

The phase-by-phase build reports describe what was implemented and checked. The counts below come from those records; they are not a fresh uptime check or a measurement of content performance.

• Private by design: the admin area sits behind its own sign-in, changes are logged, and visibility rules keep paid lessons out of public results.
• Library: 439 cleaned items, 1,341 timestamped sections and 249 cross-links.
• Transcripts: 73 caption files became 73 transcripts, split into searchable, timestamped pieces.
• First live research subject: 10 question clusters from 517 real search questions, 3,198 links to evidence, and 8 cards waiting for a person to review.
• Agents can use it directly through a small set of tools. Every packet comes with its sources, and every saved draft is logged.
• Outside connections: the July 19, 2026 build report tested the unconfigured WordPress-draft and YouTube-metadata staging paths. It did not exercise them against live accounts.
• Not compiled here: the full build timeline, model and lane history, and cost.