7
chapters · ~12,000 words
Source-cited throughout
Synthesized from PMax Pros, PPC Town Hall, Optmyzr Q1 2026 benchmarks, Sparktoro AIO citation data, and 200+ deployments.
Three distinct layers: in-platform AI, MCP-mediated AI, external agents. Conflating them is where operators leak money.
What humans keep. What Smart Bidding owns. What external agents observe. The misallocation pattern that costs the most.
Auditor → Negative Hunter → Asset Generator → Reporter → Anomaly Watcher. Build order, MCP boundaries, approval gates.
The 30-day hardening sequence. 6–9% CPA drift vs 23–41% drift - depending on which side of preparation you're on.
The 10,000-negative ceiling. Consolidate vs split decision matrix. The cannibalization fix.
“The 5-agent decision map in Chapter 3 ended a 6-week internal debate about what to automate vs what to keep human. I screenshotted it to the team and we moved on.”
“Read it on a Friday. Wrote the CLAUDE.md from the template on Saturday. Caught Claude hallucinating 38 negative keywords on Monday - exactly the failure mode Chapter 7 warned about.”
“The Threads arbitrage chapter alone justified the 30 minutes. Moved 12% of Meta spend over, CPMs landed 34% lower than IG. The window's real.”
“I've bought $500 courses with less source-citation than this playbook. The Tinuiti + CTC + Motion benchmarks in one document - that's the artifact I've been hunting.”
“Sent this to my entire team. The 4-agent CRM architecture is now the reference doc for how we structure every Klaviyo build going forward.”
“Chapter 4 on the AI Max migration is the chapter I'd pay for. We started the 30-day hardening sequence in March. April CPA drift was 7%. The unhardened account next to ours: 29%.”
The +91% paid CTR for cited brands. The 200-domain rule. Schema, llms.txt, bid-up strategy on cited queries.
How to prevent the 200-hallucinated-negatives failure. The append-only audit log pattern. The 5 kill-switch triggers.