# Creative fatigue rules

Claude uses these rules to give every ad exactly one verdict. These are starting lines: tune them after a few weeks on your own account.

## Step 1: Maturity gate

Score an ad only when it has all three:

- 10,000 impressions or more
- 3 days of delivery or more
- spend of at least 1x your target cost per result

Below the gate, the verdict is **WAIT**. Do not judge it yet.

## Step 2: Find the ad's best week

Use a weekly series for the last 8 weeks. The best week is the week with the lowest cost per result. Every comparison below is against the ad's own best week, never against a fixed calendar window.

## Step 3: One verdict per ad

Cost per result is the only trigger. Frequency and CTR only decide which action it gets.

| Situation | Verdict |
|---|---|
| Spending this week, 0 results | **RETIRE** |
| Cost per result up 20% or more vs best week, AND (frequency 2.5 or more OR link CTR down 25% or more vs its first healthy week) | **RETIRE** |
| Cost per result up 20% or more vs best week, frequency below 2.5, link CTR not down 25% | **REFRESH** (the concept works; make a new cut) |
| Cost per result up 10-20% vs best week | **MONITOR** (prepare a replacement) |
| Cost per result holding, but CTR falling and frequency 2.5 or more | **MONITOR** |
| Cost per result holding | **HEALTHY** (leave it) |

## Step 4: Show your working

For every ad, show: impressions, days live, spend, cost per result this week, cost per result in its best week, % change, frequency, link CTR change, and the verdict. Label any number that is an estimate.

## Frequency trap

A 7-day frequency can look low while the same people have seen the ad many times over 90 days. Use 90-day (or lifetime) frequency for the saturation check, and say which window you used.