Claude Code as the PMM operating system. Automated fact-check on every page before it ships.

Most PMM functions buy AI tools and complain the tools don't know the product.
This one built its own, in the open, on top of Claude Code.
The outcome: automated fact-checking on every landing page the team ships, runnable from a single slash command, every claim traceable to a sourced quote.
Zero LLM spend on top of an existing Claude subscription.

For the AMP / GEO / agent-infra founder reading this

If you're hiring a PMM in 2026 and the thing you're actually buying is "this person won't ship a hallucinated landing page on launch day," this is the case to read. Substrate runs the pipeline and the fact-check on every page before it ships. Six slash commands keep the cadence. The gate is what you're buying, not the AI. The AI part is cheap. The gate is what I leave behind when the engagement ends.

What it was

A B2B communications platform where PMM owned messaging, landing pages, and battle cards. The help centre lived elsewhere on the team. The bottleneck was never headcount. It was that the canonical knowledge file was always a week stale, and every operator had a slightly different version pasted into their prompt.

The diff

Before, jul 2025: ChatGPT custom GPTs for one-off help. Drafts hand-checked. Numerical claims on landing pages shipped without verification. Sales-call review eating three to four days per twenty-call batch.

After, mar 2026: A pipeline (the precursor to what is now substrate) running claim → source → fact-check → render. One canonical Product-Knowledge.md file. An Allowed-Claims.md registry that grows with every verified claim. Six slash commands wired into the daily cadence. Two research workstreams in build against named programmes. 730 of 730 help articles read and gap-mapped.

What I actually did

1. Started with customGPTs scoped to single bottlenecks. Jul 2025: deal-brief GPT and n8n + Mastermind for CI extraction from calls. Aug 2025: a second customGPT for sales-call CI that fed the State of CI report the same month.
2. Shipped MessagingGPT in Nov 2025. ICP, feature grid, value-based messaging, social proof, all integrated. Fine-tuned for AIVA Custom Actions in Dec 2025.
3. Moved from one-offs to a pipeline inside Claude Code. Every output traceable back to a sourced claim. Every claim labeled direct, indirect, or contextual. This pipeline became the precursor to substrate; the same pipeline this portfolio was written inside.
4. Made Product-Knowledge.md the single source of truth. Six commands read from it: /render-pages, /detect-drift, /generate-content, /read, /discover, /digest.
5. Ran the first production fact-check on a live PPC Sales Dialer page. 42 claims verified. 5 BLOCK, 12 FLAG, 4 FIX, 1 MATH error caught before the page reached a buyer. Three new case studies added to Allowed-Claims.md the same month.
6. Shipped through justcall-staging.vercel.app, a personal Vercel project, to bypass the production handoff gate. Mar 2026 LP batch (9 live, 6 in design, 4 integration LPs) and Homepage Refresh 6 went through it.

What stayed honest

None of these were sanctioned initiatives. I built them on weekends to unstick bottlenecks the monthly report kept naming, then folded them into the rhythm once they worked. KB coverage sits at 16%, not done. The 70% call-review reduction is projected from a pilot, not measured at full volume. The 6,724 win/loss and 8,773 retention records were ingested but the 2026 ICP work that depends on them was still in flight when I resigned. Every number cites a monthly PMM report you can read on a hiring NDA.

What it became, in substrate

The fact-check pipeline with BLOCK/FLAG/FIX/MATH labels and an Allowed-Claims registry, the single canonical product-knowledge file every downstream command reads, the operator-owned Vercel staging that bypassed a slow handoff gate, and the principle that every tool is scoped to a bottleneck named in last month's report are four of the moves substrate carries forward. The skill, routine, and principle layers are open source for any client to clone and run.