ChatGPT Ads manual · The account audit and its score

📗 This is the user manual for the Suite for ChatGPT Ads, the tool in the Ninja Scripts panel. Here you can search it, listen to it, highlight it and save it to favourites. It is updated with every release of the tool.

A health snapshot of the whole account, with a score from 0 to 100 and the list of what to fix in order of severity. It's recalculated on every visit and after every read: it costs nothing, it doesn't go out to the internet and it writes nothing to your account.

The six areas and their weight

Each area starts at 100 and loses points per finding: 40 for a critical, 18 for an important, 7 for an improvement. The score is the weighted mean. An area with no data doesn't score: its weight is shared out among the others, rather than penalising you for something that can't be measured.

≥ 80 good · ≥ 55 needs work · below that, poor.

Two caps that prevent a misleading score

1. With critical findings the score cannot say «good». With one it won't go above 79; with two or more, above 54. Without this, five areas at 100 masked a collapsed delivery and the account came out «good» with its campaign paused and a rejected ad.
2. Spending with no target cost per conversion caps you at «needs work». It isn't an accusation — nobody knows whether it's wrong: it's refusing to hand out a pass when nobody can say whether that money is well spent. It lifts as soon as you set the target, which is one field.

The score card tells you when the score is capped and why.

The fourth state: «can't be judged yet»

Besides critical, important and improvement, there are two categories that exist to avoid noise:

• Not a worry (no points lost): for example, features that need another Ninja tool.
• Can't be judged yet: not enough volume, no relevance index calculated yet, an account just connected, or campaigns created less than 24 hours ago — which the panel creates paused on purpose. Also when the ⏰ Schedule has campaigns paused because it should: that isn't a finding, that's the schedule working.

Those things are said rather than kept quiet. On a new channel, without this state half the report would be noise and the other half silence.

What it checks

It doesn't repeat the 💡 Recommendations rules: it absorbs them, with their own wording, so if their thresholds ever change the audit changes with them. On top of that, its own checks:

• Delivery: brand and integrity review, delivery issues, rejected or in-review ads, an account with not one impression, nothing active.
• Structure: campaigns with no ad groups, groups with no ads or no hints, too few (under 4) or too many (over 40) hints, everything in one bag (a single group with ≥ 12 hints and ≥ 3 ads), the same hint in several groups (they compete with each other).
• Ads: text that gets cut off or that wastes the card, ads with no image, twin ads (same headline or line within the group), a single ad in a group that has traffic.
• Measurement: a campaign with no conversion event, ads with no utm_content tag or with a repeated tag, a landing page that looks like a bare sign-up form, spend with no conversions at all.
• Money: spend with no conversions against your target, cost per conversion above the target (important from 1.2×, critical from 2×), missing budget, low bid.

Thresholds in plain sight

Every finding carries its figure and its threshold in the wording itself, and the screen publishes the full threshold table in a collapsible panel. That way you can argue with a threshold instead of believing a score.

The history

One row per day is stored, with the score and its areas, and the screen draws the 30-day trend. It's the only thing that can't be reconstructed backwards: it starts counting the day you connect the account.