Data Dump Friday — Gold Nugget Report

1.Executive Summary

Six videos, ~3 hours of source material, processed in under 30 minutes of reading. Zero dollars spent; four assets added to the stack, one workflow formalized. The night meta-finding: three of six dumps — from three unrelated creators — independently delivered the same message: AI search visibility for home services is the money lane, Gemini/Google wins it, and only Google measures it. That is the Answers AI pitch, validated by strangers, for free.

The single most critical risk: our two biggest assets are both one step behind the market right now. The 200 AEO city pages sitting in production match — spec-for-spec — the exact mass-templated pattern Google is actively de-indexing in batches (Caleb Ulku, dump #3). They're undeployed, so nothing is lost yet; but the planned programmatic directory makes page 30+ inevitable, and the uniqueness pass must happen before deploy, not after. Simultaneously, a named competitor (Phlash Consulting, dump #4) is selling AI-search customer acquisition to $1M–$10M home service businesses — our exact restoration/plumber ICP — and speaks to 600 of them in Austin next week. Every month we run undeployed costs us the 3–6 month AI trust tax on new domains (Julian Goldie, dump #1).

2.Sources Processed Tonight

Video / Channel	Length	Verdict
Union Alpha: New FREE AI API — Julian Goldie SEO	33 min	Headline dead in 21 hours. Kept the content flywheel + AI trust tax data.
The Fastest Way to Get Agency Clients — Nick Ponte	87 min	$97 offer skipped — owned in-stack. Kept Call 1/Call 2 system + opener + stats.
Google Is Killing Pages — Caleb Ulku	15 min	Torpedo warning for our 200 swarm pages. Kept the 4-layer uniqueness fix + Mueller test.
Best AI Platform to Get Customers in 2027 — Phlash Consulting	16 min	Direct competitor recon. Kept Gemini-first proof stack + intent stats.
NotebookLM Trend Tracker — Tool Drop	9 min	Reusable intel engine. Kept relative-baseline architecture + vocabulary migration.
NotebookLM Human Writing — Tool Drop	9 min	Confirms in-stack humanizer skill. Kept fresh-context Auditor loop.

3.Workflow Diagnostic — Gold Nuggets Added to Stack

Video Intel Pipeline is now a standing stack asset. Filter: kill dead ideas first → extract mechanics → tie to an active lane → one go/no-go offer. Running as designed; tonight was its live-fire test.
Union Alpha (Julian Goldie headline): dead on arrival — pulled from OpenRouter free tier after ~21 hours. Kill-fast rule confirmed working on first live target.
Nick Ponte's $97 course + $117/mo Inner Circle: every component already owned in-stack (Hermes workers, Sam dialer, our own AI audit tool). Nuggets kept: Call 1/Call 2 split, the "Do you know if tools like ChatGPT are recommending your business?" opener, cherry-picking = disqualification discipline, Constant Contact stats (73% / 60% / 56%). Fix: fold Call 1/Call 2 into restoration first-call sheet (offer ①).
Julian Goldie's flywheel: auto-suggest-only keyword research, 3–5 posts/day, 4 CTAs + schema per post, 93K AI mentions / 418 clicks/day as proof, 3–6 month maturity tax. Fix: cadence plan for Answers AI once pages pass audit.
Caleb Ulku's de-indexing data: pages graded against each other on-domain; page-30 threshold; fix = 4-layer uniqueness (GSC query grounding + call transcripts + local Reddit + brand narrative) rebuilding indexation to ~97%; track rejection rate per template. Fix: Mueller competitor-test + uniqueness pass on swarm pages before deploy (offer ②).
Phlash Consulting competitive recon: Gemini wins local on ecosystem, intent (80% start on Google; only 18% act on AI recs unverified), and measurability (GSC Generative AI report = only native AI visibility metric). Fix: fold stats into restoration pitch as the "why now" block (offer ③).
Tool Drop trend tracker: relative baselines beat absolute volume; vocabulary migration (GitHub → HN → news) catches trends early; mainstream tier = closing bell, not warning. Fix: build ProPlum Trend Radar on our sources (offer ④).
Tool Drop humanizer build: fingerprint = predictability + burstiness; self-grading is structurally broken — writer must never audit its own draft; fresh-context Auditor with hard gates. Fix: upgrade humanizer skill, fold into swarm-page audit (part of offer ②). Flag: detector-evasion ≠ ranking — sell "sounds human," never "beats detectors."

4.Threat Matrix

Severity	Risk	Component	Remediation
High	200 swarm pages match the mass-template pattern Google is de-indexing in batches; same-spec generation = pages compete against each other	Swarm content store + planned programmatic directory	Run Mueller competitor-test + 4-layer uniqueness pass before any deploy (offer ②)
High	Named competitor (Phlash) pitching AI search to $1M–$10M home services — our exact ICP; 600-owner talk in Austin	Restoration vertex lane; Answers AI positioning	Fold Phlash stats into restoration outreach pitch (offer ③); lead with Gemini-first measurability
Med	3–6 month AI trust tax on new domains — every undeployed month delays compounding	Answers AI content cadence	Start 3–5 posts/day flywheel once pages pass audit
Med	Self-grading blind spot: unchecked AI content excuses its own clichés	Humanizer skill; all outward copy (swarm pages, restoration SMS/email)	Mandatory fresh-context Auditor pass before anything ships (folded into offer ②)
Low	Dead tools confirmed: Union Alpha (21-hour lifespan), paid humanizer tools (skill already owned in-stack)	—	No action; kill-fast rule validated

5.Offers Awaiting Ruling

#	Project	Status	Next Action on "Go"
①	Ponte Call 1/Call 2 → restoration first-call sheet	Specced	Merge framework + AI-search hook into first-call sheet
②	Mueller test + humanizer audit → swarm pages	Specced	Competitor-clone flags + 4-layer uniqueness spec on all 200 pages
③	Phlash stats → restoration outreach pitch	Specced	Add "why now" stats block to outreach pitch doc
④	ProPlum Trend Radar	✅ BUILT — Friday Night Build (see §6)	Live; daily 06:23 UTC; first verdict filed

6.Friday Night Build — ProPlum Trend Radar

Picked from Jeff's GitHub shortlist: AKzar1el/mcp-trendpulse — the only zero-paid-dependency option (no X API bills, no scraping service), covering the two tiers that matter most: Google News (mainstream) + Google Trends (relative baselines), speaking agent-native into the same stack as the Hermes workers. Runner-up ericciarla/trendFinder rejected for X/Firecrawl monthly-bill fragility.

What it does: every day at 06:23 UTC, for 12 keywords across our three money lanes — answers-ai (AI SEO · generative engine optimization · answer engine optimization · AI search visibility · llms.txt) · proplum-restoration (plumber marketing · water damage restoration · home services AI) · sam-voice (AI voice agent · AI receptionist small business) — it scores each term against its own trailing 30-day norm ("is this bigger than usual for this topic?"), checks today's Google News count vs its own prior-week median, and scans rising related queries for vocabulary migration. Output: a staged verdict — EARLY SIGNAL / BUILDING / MAINSTREAM — filed to reports/trend-radar-latest.md. ~4-minute run, $0.

First live verdict (night one):

🔴 MAINSTREAM — AI receptionist small business at 4.29× its own baseline, accelerating ▲ — Sam's exact category is spiking right now
🔴 MAINSTREAM — llms.txt at 3.56× — Answers AI's spec term going mainstream
🟡 BUILDING — plumber marketing at 1.87× — the restoration lane's search demand is warming

The radar validated itself on its first run: it lit up on our own product category the night we switched it on. First-week caveat: news-count baselines need 7 days of history before the "bigger than usual" math is meaningful.

7.Prioritized Action Plan — Next 48 Hours

Jeff (tonight, 2 minutes): batch-rule remaining offers ①–③ go/no-go. Done = three yes/no answers in one message.
Jenni (Saturday): execute approved items in value order — ③ pitch stats (fastest) → ① first-call sheet → ② swarm audit spec. Done = updated files delivered.
Jenni (Sunday): deliver one-page "stack changes" memo covering tonight's additions + approved builds, so Monday's dial block runs on the new material. Done = memo delivered + filed.

— Report Ends —