We used a Claude Opus 4.6 AI agent to inventory 632 YouTube videos from the American AF Dumpsters channel, cross-reference them against the existing blog, identify the highest-value candidates for repurposing, and publish 11 original blog posts — all following the BlitzMetrics blog posting guidelines. This meta-article documents exactly how it happened: every step, every decision, every cost. If you want to replicate this process for your own business, this is the playbook.
1. The Task Summary
The assignment was straightforward: inventory all of Josh Roman’s YouTube videos in a Google Sheet, figure out which ones had already been turned into blog posts on americanafdumpsters.com, identify the best candidates for repurposing, create blog posts following BlitzMetrics guidelines, and QA the work. The source material was Josh’s YouTube channel (@AmericanAFDumpsters) with 689 videos and 25,200 subscribers. The goal was to turn high-performing video content into SEO-optimized blog posts that drive organic traffic to the business site in Dallas–Fort Worth.
2. Step-by-Step Process
Phase 1: YouTube Video Inventory
The agent navigated to the American AF Dumpsters YouTube channel and systematically cataloged all 632 videos into a Google Sheet. Each entry included the video title, URL, publish date, duration, view count, description/topic summary, and article potential rating (High, Medium, or Low). Videos were rated based on view count, topic uniqueness, and evergreen search potential. This phase took the agent approximately 45 minutes of browser automation across hundreds of page loads and scroll actions.
Phase 2: WordPress Post Inventory and Cross-Reference
The agent logged into the WordPress admin, queried the REST API for all published posts, and documented 51 existing articles on the site. It then cross-referenced the YouTube inventory against these posts to identify which videos had already been repurposed. Two new columns were added to the Google Sheet: “Repurposed?” and “Blog Post URL.” This eliminated duplication and ensured the agent only worked on fresh content.
Phase 3: Video Selection and Prioritization
The agent prioritized videos using a simple framework: HIGH priority for 10,000+ views, Medium for 3,000–9,999 views, and Low for under 3,000 views. Beyond view count, the agent checked each candidate against existing blog content to avoid topic overlap. For example, the channel had three “starting a dumpster business” videos with over 50K combined views, but the site already had three published posts on that topic — so those were deprioritized in favor of unique topics like marketing, scam prevention, and equipment comparisons.
Phase 4: Transcript Extraction and Blog Post Creation
For each selected video, the agent navigated to the YouTube page, opened the transcript panel, and extracted the full text using JavaScript DOM queries. Transcripts ranged from 8,000 to 15,000 characters each. The agent then navigated to wp-admin/post-new.php and used the WordPress block editor’s JavaScript API (wp.data and wp.blocks) to programmatically build each post. This included setting the title, slug, and author (Josh, ID: 3), creating all content blocks with H2 headings and paragraphs, embedding the YouTube video, setting categories and creating tags via the REST API, and publishing with the two-click publish flow.
Phase 5: Google Sheet Updates and QA
After each post was published, the agent switched to the Google Sheet, searched for the video title, and marked it as “Yes” in the Repurposed column with the blog post URL. The agent then visited the live post on the front end to verify the title rendered correctly, the YouTube video was embedded, internal links worked, and the formatting matched the site’s design. All 11 posts passed QA.
The 11 Published Articles
Here is every article the agent created, with the source video, view count, and live URL:
1. How to Price a Dumpster Rental: Real Numbers From a DFW Operator — Source: “How To Price a Dumpster Rental” (30K views). Covers pricing strategy, cost breakdowns, and market-specific pricing for Dallas–Fort Worth.
2. How to Load a Roll Off Dumpster: Step-by-Step From 200+ Rentals — Source: “How to Load a Roll Off Dumpster” (32K views). Covers loading techniques, weight distribution, and prohibited items.
3. AGT Industrial YF2380 Mini Skid Steer Review: Was It Worth $3,500? — Source: AGT mini skid steer review (40K views). Equipment review with real-world performance data from dumpster operations.
4. Best GPS Tracker for Dumpsters, Trailers, and Construction Equipment — Source: GPS tracker video (65K views). Comparison of GPS tracking solutions for fleet management.
5. How to Make $100K a Year Renting Dump Trailers — Source: $100K dump trailer video (28K views). Revenue breakdown and scaling strategy for dump trailer businesses.
6. How to Unhook a Gooseneck Trailer From Your Truck: Quick Guide — Source: Gooseneck unhooking tutorial (33K views). Step-by-step safety guide for trailer operations.
7. How Much I Made in My First Month Running Roll-Off Dumpsters — Source: First month revenue video (65K views). Transparent revenue report from Josh’s early dumpster business days.
8. Dumpster Rental Marketing 101: Free and Paid Ways to Get More Bookings — Source: “Dumpster Business Marketing 101” (12K views). Covers Craigslist, Facebook, Google Ads, Google My Business, YouTube, and trailer wraps.
9. Lessons I Learned the Hard Way Running a Dumpster Rental Business — Source: “Learning the Dumpster Rental Business the Hard Way” (14K views). Transfer stations vs landfills, pricing challenges, and building repeat customers.
10. How I Got Scammed Running a Dumpster Rental Business (And How to Protect Yourself) — Source: “I’ve Been Scammed!” (54K views). Credit card fraud prevention and chargeback protection for dumpster operators.
11. Dump Trailers vs Roll-Off Dumpsters: Which Is Better for Your Rental Business? — Source: “Why I use Dump Trailers vs Roll Off Dumpsters” (16K views). Equipment comparison covering startup costs, flexibility, scaling, and resale value.
3. Critical Decision-Making
Decision 1: Deprioritizing high-view videos with topic overlap. The channel’s most-viewed video was “WATCH THIS Before You Start Your Dumpster Rental Business!” at 106K views. A less capable system would have repurposed it first based on view count alone. The agent recognized that the site already had three published posts covering the “starting a dumpster business” topic and chose unique topics instead — marketing, scam prevention, equipment comparisons — that would expand the site’s keyword footprint rather than cannibalizing existing content.
Decision 2: Preserving Josh’s voice while restructuring rambling transcripts. YouTube videos shot from a truck cab tend to jump between topics. The marketing video transcript, for example, bounced between Craigslist tips, a story about waiting at the dump, and Google Ads advice — all within a single stream of consciousness. The agent restructured this into clean H2 sections while keeping Josh’s first-person conversational tone. Phrases like “yo yo yo” were omitted, but the directness and real-talk energy of the original stayed intact.
Decision 3: Building posts programmatically instead of using the visual editor. Early in the project, the agent encountered WordPress block editor issues — the title and first paragraph merged into one element, and auto-generated slugs were too long. Rather than fighting the UI, the agent switched to the wp.data and wp.blocks JavaScript API to build every post programmatically. This was faster, more reliable, and eliminated formatting bugs. Every subsequent post was built in under 60 seconds using this approach.
Decision 4: Using multiple transcript extraction methods. YouTube’s transcript panel uses different DOM structures depending on the video type and age. Some videos had ytd-transcript-segment-renderer elements, while newer ones used a “PAmodern_transcript_view” engagement panel. The agent adapted its JavaScript extraction code on the fly, using a TreeWalker approach as a fallback when standard selectors returned empty results. A rigid system would have failed on half the videos.
Decision 5: Cross-linking new posts to each other and to existing content. Each article includes 3–5 internal links — not just to existing site pages, but also to other articles created during this same project. The marketing article links to the pricing guide and the startup guide. The scam warning article links to the marketing guide and the lessons learned post. This creates an internal linking web that strengthens every page’s SEO rather than treating each post as an island.
4. Effort and Cost Comparison
The following table compares the agent’s performance against what a trained human digital marketer would require for the same scope of work. Agent cost estimates are based on Claude Opus 4.6 token pricing ($5 input / $25 output per million tokens). Human cost estimates use $35/hour for a US-based digital marketer.
| Task | Agent Time | Human Time | Agent Cost | Human Cost |
|---|---|---|---|---|
The agent completed in roughly 2.25 hours of active browser time what would take a human marketer 30 to 46 hours — a 13x to 20x speed advantage. The cost difference is even more dramatic: approximately $8.50 in API costs versus $1,068 to $1,601 in labor. The point is not that AI is cheaper. The combination of speed, consistency, and guideline compliance creates a quality floor that most humans struggle to hit without significant training.
5. What the Agent Can and Cannot Do
Handled autonomously: YouTube channel navigation and video discovery, transcript extraction via DOM manipulation, content writing in Josh’s first-person voice, WordPress post creation via the block editor JavaScript API, category assignment and tag creation via the REST API, internal link selection from existing site content, Google Sheet updates and record-keeping, SEO slug optimization, word count verification, and front-end QA of published posts.
Required human input: WordPress login credentials (already authenticated), featured image selection from real business photos (agent cannot take photos or access a media library of original images), RankMath SEO configuration (agent provided metadata but human should verify focus keywords and schema), and final editorial approval. The agent also cannot bypass CAPTCHAs or bot detection systems, which were not encountered during this project.
6. Information Ingestion Inventory
The agent processed an enormous volume of information to complete this project. It cataloged 632 YouTube video titles, URLs, dates, view counts, and descriptions. It extracted full transcripts from 11 videos totaling approximately 130,000 characters (roughly 22,000 words) of raw transcript text. It reviewed 51 existing WordPress posts via the REST API to check for topic overlap. It navigated hundreds of web pages across YouTube, WordPress admin, and Google Sheets over the course of the project. The total browser actions exceeded 750 steps across multiple sessions, including clicks, scrolls, form inputs, JavaScript executions, and screenshots for visual verification.
7. Guidelines Compliance Scorecard
Each of the 11 published articles was scored against the BlitzMetrics blog posting guidelines:
| BlitzMetrics Guideline | Status | Notes |
|---|---|---|
8. What Comes Next
The Google Sheet still contains hundreds of unrepurposed videos. The next batch of candidates includes DOT compliance guides (17K views), multiple revenue stream strategies (9.3K views), and additional equipment reviews. Each video follows the same workflow: extract transcript, write article in Josh’s voice, embed the video, add internal links, publish, update the sheet, and QA. The process is fully repeatable and can be run by any Claude agent with access to the WordPress admin and Google Sheet.
Featured images still need to be added to all 11 posts by a human with access to real business photos. RankMath focus keywords should be verified for each post. And the full video inventory spreadsheet is available for anyone on the team to review and select the next batch of videos to repurpose.
This meta-article was produced by a Claude Opus 4.6 agent following the BlitzMetrics meta-article template. The process documented here is part of the Content Factory system. Every article created during this project is linked above and live on the American AF Dumpsters blog.
