---
name: meta-ad-teardown
description: Diagnoses a losing Meta ad against its metrics and outputs a structured teardown (hook / body / CTA), then proposes 3 net-new angles to test. Use when the user pastes an underperforming Meta ad and wants to know what's wrong.
---

# Meta Ad Teardown

You teardown an underperforming Meta ad into its component parts, diagnose each,
then propose 3 net-new angles for the next test. The deliverable is a thinking
artifact, not a copy artifact.

## Inputs you need

1. **The full ad** — primary text, headline, link description, CTA, and a short
   description of the creative (image or video, what's shown)
2. **Performance metrics** — spend, impressions, reach, frequency, CTR (link or
   all-action), CPC, CPM, conversion rate, ROAS or CPA
3. **Audience** — cold / warm / retargeting + targeting setup
4. **Benchmark** — what "good" looks like in this account (if known)
5. **What you've tried** — past iterations and their results
6. **Hypothesis the user already has** — so you can confirm or contradict

## Output format

### Diagnosis matrix

A table:

| Component | Metric signal | Diagnosis | Confidence |
|---|---|---|---|
| **Hook (first 125 chars)** | Low CTR + low frequency | (your read) | 1–5 |
| **Body** | High CTR + low CVR | (your read) | 1–5 |
| **Visual** | Low hook rate (3-sec views / impressions) | (your read) | 1–5 |
| **CTA** | High CTR + low conv | (your read) | 1–5 |
| **Audience** | High CPM + low CTR | (your read) | 1–5 |
| **Landing page** | High CTR + low CVR + low time on page | (your read) | 1–5 |
| **Offer** | High intent + low conv | (your read) | 1–5 |

For each row, the **diagnosis** must reference the actual metric value the user
provided. "Low CTR" is not enough — say "0.6% CTR vs the 1.2% account average."

The **confidence** column forces you to be honest. If you don't have the metric
to support the diagnosis, score it 1 or 2 and say so.

### Primary diagnosis (one paragraph)

Single sentence stating the single most likely cause. Then 2–3 sentences explaining
the chain of reasoning.

### Three angles to test next

Each angle must:
- Address a different component of the diagnosis (don't propose 3 hook rewrites
  if the diagnosis is "audience")
- Be a hypothesis, not a copy variant ("test brand-led vs benefit-led hook for
  cold prospecting" not "use headline X")
- Include a one-paragraph rationale tied to the diagnosis
- Include a one-paragraph test design (budget split, success metric, decision rule)

Format per angle:

```
### Angle [N]: [SHORT_NAME]

**Hypothesis**: (one sentence)

**Why this fits the diagnosis**: (one paragraph)

**Test design**:
- Budget: X% of campaign / $X for Y days
- Success metric: CTR > X%, CPA < $X, etc.
- Decision rule: at Z conversions, kill or scale
```

## Hard rules

- **Diagnosis must reference numbers.** No metric-free claims.
- **Separate diagnosis from prescription.** Don't write the new copy — propose
  what to test. Hand off to `meta-primary-text` or `meta-headline-descriptions`
  for the actual copy.
- **Be willing to say "audience" or "offer" is the problem.** Often the ad is
  fine and the targeting or price point is broken.
- **No "test more creative" as the only answer.** That's the default. Be specific.
- **If the user disagrees with the diagnosis**, engage with their reasoning. Don't
  rubber-stamp.

## What to do when data is missing

If the user gives only "this ad isn't working", ask for at least:
- Spend
- CTR
- CVR or CPA
- Audience temperature

Without those four, a teardown is guesswork. Don't fake confidence.

## Common diagnosis patterns

- **Low CTR + low CPM** → hook problem (audience saw it, didn't click)
- **Low CTR + high CPM** → audience problem (Meta couldn't find good prospects)
- **High CTR + low CVR** → landing page mismatch or offer mismatch
- **High CTR + high CVR + low ROAS** → AOV / pricing problem, not creative
- **Decaying CTR + high frequency** → creative fatigue, refresh needed
- **Strong day-1, dies by day-7** → audience too narrow or wrong learning signal

Use these as starting points, not conclusions.