# YouTube Full-Auto Research Memo
*Barry ask 2026-09-07 · Author: Grok (executor) · Audience: CoS / Sage / YouTube bot*  
*Companion to [`STRATEGY.md`](STRATEGY.md) (Sage 2026-09-06). This memo optimizes for **max automation** (pipeline without Barry in the loop), not for max theoretical RPM alone.*

**Skeptical baseline:** “Set and forget → $200K” claims are mostly affiliate bait. Realistic $200K/yr ad run-rate still needs ~2–4M monetized long-form views/mo at blended ~$5–8 RPM, or fewer views in higher-RPM niches **that are still policy-safe**. That is top-few-percent execution over **M24–30**, not month 6. Most faceless channels never clear YPP.

---

## 1. Honest split: what CANNOT be zero-Barry vs what CAN be automated

### Cannot be zero-Barry (identity / legal / platform gates)
These require a real human with legal standing. Automating them = ToS violation, account loss, or tax fraud risk. Do not attempt.

| Gate | Why Barry-only |
|---|---|
| Google account + phone verification | Carrier SMS / identity; Google bans automation |
| 2-Step Verification / recovery | Account ownership |
| Brand channel creation / ownership transfer | Tied to Google identity |
| AdSense for YouTube link + tax forms (W-9/W-8) | Legal identity + IRS |
| Bank / payment profile | KYC; payouts |
| YPP apply click + policy review responses | Human review; appeals need owner |
| Community Guideline / copyright strike appeals | Channel owner only |
| Content ID / rights disputes that need “I own this” | Legal assertion |
| AdSense PIN / address verification (if mailed) | Physical |
| Final **public upload / go-live** (current Apex rule) | Barry yes until he changes it |
| Kids licensing / brand deals / merch contracts | Negotiation + signatures |

**Implication:** “Whole line without him helping” is **false for the business shell**. True ceiling = Barry does a **one-time setup weekend** (~2–4 hrs for 3–4 accounts) + rare appeals + YPP apply days; Sage/bot runs production forever with **draft-only** uploads until Barry flips the publish gate.

### Can be automated (production + ops loop)
| Stage | Automation fit | Apex stack |
|---|---|---|
| Niche research / topic queue | High | Claude + YouTube Data / search scrape |
| Script (hook, chapters, visual tags) | High | Claude (4-rule script prompt) |
| Fact-check / originality pass | Med–High | Claude checklist + human-in-loop optional |
| TTS | High | Local Piper (already in Sage path) or ElevenLabs if Barry buys |
| Stills / B-roll / scenes | High | OpenArt (+ stock only if licensed) |
| Motion / lip-sync clips (sparingly) | Med | Higgsfield / Veo via OpenArt credits |
| Assemble / captions / loudness | High | ffmpeg (+ Whisper for SRT) |
| Thumbnail variants | High | OpenArt + Claude title set |
| SEO metadata | High | Claude |
| Schedule as **private/unlisted draft** | High | YouTube Data API (OAuth once by Barry) |
| Analytics → next-topic iterate | High | YouTube Analytics API → Claude weekly |
| Kill/feed channel decisions | Med | Bot recommends; CoS/Barry confirms kills |

**Closed loop that is real:**  
`topic → script → TTS → visuals → ffmpeg → thumbnail → draft schedule → analytics → topic`  
**Publish** stays gated. That is “as automated as possible” without lying.

---

## 2. Niche ranking (automation × RPM × policy survival × time-to-YPP)

Scoring 1–5 each; **composite = product** (harsh on any weak leg). RPM figures are **creator-reported ranges** (OutlierKit, DepthHQ, ReelPilot, Faceless.my, STRATEGY.md) — not YouTube-published. Prefer mid of range for planning; US/UK geo skews high.

| Rank | Niche (faceless) | Auto | RPM* | Policy | YPP speed | Comp | Notes |
|---|---|---|---|---|---|---|---|
| **1** | **History / dark documentary** (narrow: disasters, empires, Cold War ops) | 5 | 4 ($5–12; some claim $10–22) | 4 | 4 | **320** | Best Apex fit. Original narration + maps/AI stills. Evergreen watch time. |
| **2** | **Space / science curiosity** (long-form explainers, mysteries) | 5 | 3–4 ($4–10) | 4 | 5 | **~300** | High median demand in 2026 samples; long runtime → watch hours fast. Public-domain NASA + AI viz. |
| **3** | **Mythology / folklore retellings** (original scripted retellings) | 5 | 3 ($3–8) | 4 | 4 | **240** | Small-channel friendly; script-led; avoid scrapes of copyrighted adaptations. |
| **4** | **Original scary / creepypasta-style fiction** (own stories, not Reddit dumps) | 5 | 4 ($4–13) | 3 | 3 | **180** | High RPM *if* original. **Reddit read-alouds** fail reused-content (“readings of other materials”). |
| **5** | **English-learning long-form** (podcast-style intermediate; NOT “AI tutor medical”) | 4 | 4 ($8–15 claimed) | 3 | 3 | **144** | Strong automation + advertiser demand; crowded “vocab slideshow” dies as inauthentic. Need distinct curriculum arcs. |
| **6** | **Sleep / ambient** (varied original scenes + distinct soundscapes) | 5 | 2 ($1–5; skeptics $0.50–2.50) | **2** | 5 | **100** | Watch-hour machine but **highest YPP rejection** under inauthentic. Demote from STRATEGY core. |
| **7** | **True crime (doc narration)** | 3 | 4 ($5–12) | 2 | 2 | **48** | RPM good; editorial load, advertiser-friendliness, “off-putting” risk. Not max-auto. |
| **8** | **Benny / MFK kids Shorts** | 4 | **1** ($0.02–0.10 Shorts; $1–3 long) | 2 | 3 | **24** | Structural COPPA tax; kids quality principles hit AI mass-prod; licensing needs Barry. |
| **✗** | **AI-persona finance / health / legal / politics advice** | 5 | 5 | **0** | 0 | **0** | **Official YPP ban** (AI personas on sensitive topics). Do not chase high RPM into a brick wall. |
| **✗** | **Template Top-10 slideshow / scraped compilations** | 5 | 2 | **0–1** | 1 | **0** | Inauthentic + reused. “Viral and easy” is dead by design (Jul 2025). |

\*RPM = approximate creator-reported long-form AdSense RPM after YouTube’s cut; treat as directional.

### Policy sources (primary)
- [YouTube channel monetization policies](https://support.google.com/youtube/answer/1311392) — Jul 15, 2025 rename to **inauthentic content**; reused content; **AI personas on finance/health/legal/politics** ineligible; kids quality principles.
- [YPP overview](https://support.google.com/youtube/answer/72851) — 1K subs + 4K watch hrs **or** 10M Shorts/90d (current).
- [YouTube Blog Aug 10, 2026](https://blog.youtube/news-and-events/youtube-partner-program-updates-2027-new-opportunities-earn/) — **Feb 1, 2027**: new applicants need **8K watch hrs or 20M Shorts**; Shorts ad share needs recurring 10M/90d. **Get into YPP in 2026.**

---

## 3. Revised portfolio (max automation) — differs from STRATEGY.md

STRATEGY kept ambient + Benny as portfolio legs. For **Barry-hands-off production**, revise:

| # | Channel | Role | Cadence | Steady-state (honest M18–24) | vs STRATEGY |
|---|---|---|---|---|---|
| **A** | **History/doc** (narrow vertical) | **Core earner** | 2–3×/wk, 10–18 min | 0.8–1.2M views/mo × $6–10 → **$5–12K/mo** if it hits | Keep / promote |
| **B** | **Space/curiosity science** | **Watch-hour + second lottery** | 2×/wk, 12–25 min | 0.6–1.0M × $4–8 → **$2.5–8K/mo** | **New** (replaces ambient as #2) |
| **C** | **Original scary fiction** | Gated earner | Start M4–6 if A pipeline stable | 0.5–0.8M × $5–10 → **$2.5–8K/mo** | Keep but **no Reddit scrape** |
| **D** | **Ambient sleep** | Optional watch-hour helper only | 1–2×/wk if capacity free | Low $; helps hrs toward YPP | **Demote** — do not count in $200K math |
| **E** | **Benny Bunny** | Brand lottery only | Maintain light Shorts **or freeze** | Ads ≈ noise; upside = licensing (Barry) | **Demote hard** — not automation path to $200K |

**Portfolio math for ~$200K/yr (~$16.7K/mo):** need **A + B (or A + C) in top-few-percent view territory**, not four mediocre channels. Ambient + Benny no longer pad the spreadsheet.

**Kill / do not start**
- On-camera Barry personality channels.
- AI finance/health “expert” hosts (policy zero).
- Compilations of other creators’ clips.
- Kids music licensing hustle as a *required* revenue line (Barry time).
- Anything whose weekly quality depends on Barry’s creative taste.

**Still true from STRATEGY:** $200K = **M24–30 run-rate**, not calendar-year-1; ~50/50 even with good execution; kill losers fast.

---

## 4. Concrete hands-off production stack (Apex-native)

```
Claude (script + titles + SEO + weekly analytics brief)
    ↓
Local Piper TTS (en_GB-jenny or dedicated channel voices)  [already Sage]
    ↓
OpenArt (keyframes / scenes / thumbs) + sparse Higgsfield/Veo motion
    ↓
ffmpeg (concat, ducking, loudnorm, burn captions from Whisper SRT)
    ↓
YouTube Data API → upload as private/unlisted + schedule
    ↓
YouTube Analytics API → Claude iterate topics / hooks / length
```

| Piece | Tool | Notes |
|---|---|---|
| Brain | Claude | Hook + open loop + visual tags + soft CTA; originality checklist |
| Voice | Piper local | $0 marginal; swap voice per channel for brand separation |
| Stills | OpenArt | Budget ~$150–300/mo at full cadence (per STRATEGY) |
| Motion | Higgsfield / Veo sparingly | Don’t burn credits on every beat — stills+pan is enough for history/space |
| Edit | ffmpeg | Deterministic; CI-friendly on mini-PC/VPS |
| Captions | Whisper → SRT → burn or upload | Accessibility + retention |
| Thumbs | OpenArt + 2–3 Claude title variants | A/B via TubeBuddy/VidIQ later if Barry enables |
| Upload | YT Data API | **OAuth once by Barry**; bot only drafts until publish gate lifted |
| Ops | Sage / YouTube bot | Daily queue; weekly CoS report; never Fuller |

**Originality checklist (gate before draft upload)** — reject if any fail:
1. Script is original prose (not paraphrase of one source / not Reddit dump).
2. ≥N distinct visual beats (not one template slideshow).
3. Story/argument varies outcome across episodes (no identical storyline cookie-cutter).
4. No AI “expert” advice on finance/health/legal/politics.
5. Music/SFX: owned or properly licensed (not random TikTok audio).
6. Synthetic-media disclosure if required for realistic depictions.

---

## 5. Risks (ranked)

1. **Inauthentic content (Jul 15, 2025)** — #1 existential. Mass-produced / templated / minimal variation → YPP reject or demonetize channel-wide. Mitigation: variation + editorial checklist + capped cadence (quality > flood).
2. **Reused content** — compilations, scraped social, **verbatim readings of others’ text** (explicit example in Help). Mitigation: original scripts only; transformative commentary if citing.
3. **AI personas × sensitive topics** — finance/health/legal/politics with AI host = monetization ban. Do not “pivot to finance for RPM.”
4. **MFK / kids quality principles** — Benny: no personalized ads; “hard to follow” / mass-produced kids content can suspend YPP. Licensing upside needs Barry → demote for full-auto goal.
5. **Copyright / Content ID** — music, footage, character lookalikes. Prefer AI-original + public domain (NASA, archive.org) + licensed libraries.
6. **Feb 1, 2027 YPP bar double** — 8K hrs / 20M Shorts for *new* applicants. Clock is real; prioritize watch-hour formats (history/space long-form) in 2026.
7. **“Set and forget $200K” myth** — ignore gurus. Plan for library compounding + 2/wk discipline for 18+ months.

---

## 6. Next actions — Sage / YouTube bot

**Hard rules (unchanged)**  
- **No public upload without Barry yes** (draft/private/unlisted OK once accounts + OAuth exist).  
- **Never Fuller** (`+1 (650) 460-1036`). YouTube work has zero reason to touch voice production.

### This week
1. Barry (human, scheduled): create **2 brand channels** (History + Space) — phone verify, 2SV, advanced features. Ambient/Benny optional later.
2. Sage: brand kits (name, banner, About, voice sample, OpenArt style refs) for A + B; store under `youtube/projects/`.
3. Sage: implement draft pipeline stub — Claude script → Piper → OpenArt batch → ffmpeg → **private** upload dry-run on one channel after Barry OAuth.
4. CoS: keep publish gate in LOCKS; do not ask Barry to auto-publish yet.

### M1–3
5. Ship History 2/wk + Space 2/wk into private→Barry-approve→public. Gate M3: ≥1 video >10K views **or** CTR >4% + retention >40% on each.
6. Chase 4K watch hours on long-form (prefer 12–20 min history/space over Shorts spam).
7. Freeze or minimal-maintain Benny; do **not** staff kids licensing until ad engine works.

### M4–6
8. YPP apply as soon as eligible (Barry clicks). Aim **before Feb 2027**.
9. If pipeline stable, soft-launch original scary fiction (C). Kill ambient if it distracts.

### Explicit non-actions
- No finance AI host channel.  
- No copying viral channels.  
- No Fuller’s number, no Retell, no outreach send.  
- No “set and forget” vendor purchase that claims guaranteed $200K.

---

## Sources
- YouTube Help — [Channel monetization policies](https://support.google.com/youtube/answer/1311392) (inauthentic, reused, AI personas, kids quality)
- YouTube Help — [YPP eligibility](https://support.google.com/youtube/answer/72851)
- YouTube Blog — [YPP updates 2027](https://blog.youtube/news-and-events/youtube-partner-program-updates-2027-new-opportunities-earn/) (Aug 10, 2026)
- Secondary RPM / niche (directional only): OutlierKit, DepthHQ, ReelPilot, Faceless.my, EchoEtch, Gyre/VidIQ (MFK), Dynamoi (ambient rejection), TechCrunch (2026 policy clarify)
- Internal: `youtube/STRATEGY.md` (Sage 2026-09-06); LEARNED transcripts on 7-agent pipelines / Claude+Higgsfield stacks

---

*End memo. Mirror: `website/projects/youtube/RESEARCH-FULL-AUTO.md` for later deploy.*