# Top Picks — REVISED after Cross-LLM Round 3

⚠ **Round 3 (focused on 8 Top Picks) significantly re-ranked round 2's findings.** This doc supersedes the original tiering. Full review: `cross-llm-review-round3.md`.

---

## Headline finding

Two definitive top picks (different by selection criterion):

- **Strategic / Concept top pick:** `ad-neighbor-v1` — Gemini's clear #1 at 57/60. *"Your block already did the math"* called the strongest single line in the batch. Bypasses rational argument, hits social-proof / fear-of-being-left-behind. Pure-CSS-art at its best.
- **Consensus top average:** `ad-loan-v1` — synthesis-weighted 51.7. Grok and GPT both rank #1. "Homeowner Math" eyebrow + two-column ledger = math-first pillar in one image.

**Recommended primary launch:** A/B these two inside a SINGLE ad set (per `[[meta-ab-single-adset]]`). Same targeting, two creative.

---

## Tier 1 — Primary launch cohort (REVISED)

| Pick | File | Round-3 score | Why |
|---|---|---|---|
| 1 | `ad-neighbor-v1` | 49.3 avg, **Gemini 57** | Strategic top pick. Best line in batch. Social-proof masterclass. |
| 2 | `ad-loan-v1` | **51.7 avg** | Consensus highest. Math-first concept-execution match. ⚠ Legal sign-off on illustrative numbers recommended. |

---

## Tier 2 — Audience-specific launch

| Pick | File | Round-3 score | Audience |
|---|---|---|---|
| 3 | `ad-ev-v1` | 45.7 (Grok 34, GPT 52 — 18pt spread) | **EV-owner lookalikes ONLY.** Grok called it "preachy." Gemini called it "the smartest niche-targeted ad." Don't broad-target. |
| 4 | `ad-firsthome-v1` | 45.3 | First-time homeowners. Quietly strong. "Ownership pillar." |
| 5 | `ad-retire-v1` | 44.0 | 55-65 demographic. ⚠ Legal review on "Lock the bill" + 20+ year framing (implied-guarantee risk per Grok). |

---

## Tier 3 — Hold or rework

| File | Why hold | Action |
|---|---|---|
| `ad-skeptic-v1` | "Wrong format for feed" per Gemini + Grok | **Repurpose as carousel ad or LP content** — don't ship as static feed. ⚠ Compliance: "State programs still pay" + "won't bring it back" need per-claim verification. |
| `ad-network-v1` | **2 of 3 reviewers want killed** ("placeholder that never got finished") | Hold from launch. Could rework with stronger visual execution. |
| `ad-spam-v1` | **2 of 3 reviewers want killed** in round 3. Round 2 said ship-grade; round 3 says kill. Concept clever but visual "cluttered mess." Doesn't hit core pillars. | Hold. **Major reversal from round-2 ranking.** Don't lead with this. |

---

## Tier 4 — Niche / situational (unchanged from original ranking)

Same as before. Use as supplements:
- `ad-cold-v1`, `ad-local-v1`, `ad-tariff-v1`, `ad-insurance-v1`, `ad-aesthetic-v1`, `ad-diy-v1`, `ad-meter-v1` (+ animated), `ad-arbitrage-v1`, `ad-electrify-v1`, `ad-inflation-v1`, `ad-billcompare-v1`, `ad-meterpov-v1`, `ad-disproportion-v1` (+ animated), `ad-paidfor-v1` (+ animated), `ad-yearone-v1`, `ad-toldyou-v1` (+ animated), `ad-math-v1`, `ad-subscription-v1` (+ animated), `ad-powerplant-v1`, `ad-literacy-v1`, `ad-boring-v1`, `ad-warranty-v1`, `ad-summer-v1`, `ad-homevalue-v1`

---

## ⚠ Cross-cutting risks raised (round 3)

1. **Visual register monotony.** Gemini called out `ad-loan` + `ad-retire` + `ad-firsthome` feeling like siblings (dark bg + chart/list registers). Grok separately flagged `ad-loan` + `ad-retire` + `ad-skeptic` clustering. **Mitigation:** when picking 4-ad rotations, pair chart-led ads with split-frame/document/map registers. Don't stack all chart-style in one ad set.

2. **Compliance flags worth legal pre-ship review:**
   - `ad-loan-v1` — illustrative numbers (Gemini + Grok)
   - `ad-retire-v1` — "Lock the bill" implied-guarantee territory (Grok)
   - `ad-skeptic-v1` — forward-looking + unsubstantiated framings (Grok)

3. **Visual execution weakness.** All three reviewers agreed `ad-spam-v1` and `ad-network-v1` look "cheap," "templated," "placeholder." Pure-CSS-art pays off only when concept is strong. **Don't rush to ship these two** without a visual-craft pass.

---

## Recommended A/B set construction (per ad-set per Meta canon)

**Cold-traffic primary set (Tier 1):**
- `ad-neighbor-v1` (map / social-proof) + `ad-loan-v1` (ledger / math)

**Audience-segment sets (parallel ad sets, NOT mixed with primary):**
- EV-owner: `ad-ev-v1` solo (or paired with second EV-targeted creative — but `ad-electrify-v1` is the only adjacent one and it's visually similar)
- First-time-buyer: `ad-firsthome-v1` solo or paired with `ad-warranty-v1` (different register — checklist vs warranty bars)
- Retirement: `ad-retire-v1` + maybe `ad-yearone-v1` (different register — chart vs diary)

**Don't launch without legal review:**
- All Tier 1 picks once at scale spend > test budget

---

## What changed from original TOP_PICKS

| Pick | Original tier | Round-3 tier |
|---|---|---|
| `ad-spam-v1` | **Tier 1 #1** (8.83 round 2) | **Tier 3 — Hold** (38.7 round 3, "kill" from 2/3 reviewers) |
| `ad-network-v1` | Tier 1 #4 | **Tier 3 — Hold** (40.7, "kill" from 2/3) |
| `ad-loan-v1` | Tier 1 #2 | **Tier 1 #2** (consensus best, 51.7) |
| `ad-neighbor-v1` | Tier 1 #3 | **Tier 1 #1** (strategic top, Gemini 57) |
| `ad-skeptic-v1` | Tier 2 | **Tier 3 — repurpose to carousel** |
| `ad-monopoly-v1` | Tier 3 hold | **Still on hold + Option-A restructure built** (`ad-monopoly-v2-restructured`) |

---

— Clone (ad31a131), revised 2026-05-15 post-round-3