📗 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.
Three sister tables, one per level, all behaving the same: the date selector, filters, columns that sort when you click them and a totals row at the bottom.
🏗️ Campaigns
Status · daily budget · objective and bid · how many ad groups and ads it has (live out of total) · impressions · clicks · CTR · conversions · spend · CPC. Each row links to its ad groups.
🧩 Ad groups
Their campaign · status · strategy and bid (with a ⚠️ badge if OpenAI flags a low bid) · how many context hints it has · ads · relevance index with its colour · the metrics. Plus the three ways into what you do with an ad group: 🧭 Hints, 🧪 Lab and 🖼️ Ads.
🖼️ Ads
The full card (headline, line and where it goes: domain plus its utm_content tag), its ad group, its campaign, the status and OpenAI's review status. If an ad is rejected, its badge links straight to 🚫 Rejected.
Ads are also created and edited from here: above the table you have ✍️ «New ad by hand» and 🧪 «New ad with AI», and every live row carries its ✏️ «Edit». What stays in the ad group's 🧪 Lab is what needs its context: the official preview and archiving.
There is no conversions column per ad: OpenAI's API doesn't provide them at that level, and a column of zeros would look like a measurement. Real cost per ad lives in 📈 Report.
Two badges you need to read
«No longer there» · a row with this badge is something that spent money during the period you picked but no longer exists in the account (archived or deleted). It appears on purpose: without it the table's totals wouldn't match the account's, and that spend is real.
«Not serving · its ad group is archived» · an ad can be active and never show, because what's switched off is its ad group or its campaign. OpenAI's API only reports each object's own status, without looking at its parent; the table does look, and tells you the reason alongside. That's the explanation for the «not serving» you see in the Ads Manager.
What you can change from the table
The idea is that you enter the Ads Manager as little as possible:
| From | You can |
|---|---|
| 🏗️ Campaigns | ⏸ pause · ▶️ activate · 💶 change the daily budget |
| 🧩 Ad groups | ⏸ pause · ▶️ activate · 🎯 change strategy and bid |
| 🖼️ Ads | ⏸ pause · ▶️ activate · ➕ create a new one · ✏️ edit its text |
Always with confirmation before anything is touched. Underneath it's the same engine as 🎛️ Bids, so it inherits its guarantees: the value is re-read live before writing (the panel's snapshot can lag behind what you did in the Ads Manager), it's confirmed by reading afterwards, it goes into the history and it can be undone.
Two details with their reasons:
- Archiving is not offered here. In OpenAI, archiving is irreversible; it's done from the 🧪 Lab, which saves a copy of the creative first.
- A change you make by hand is not quarantined and not capped: you set whatever figure you want and can touch it again straight away. Quarantine and caps protect against repeated decisions the panel takes (applying a recommendation, a bulk batch), not against your judgement.
➕ Creating an ad without leaving this screen
An ad hangs from an ad group, never from the account, so the first thing asked is which one. If you came with an ad group in the table's filters, it is already filled in and you do not have to pick it again.
Then you choose the path. Each button takes you to its own: if you pick «with AI» you will not see the by-hand form underneath, and the other way round. They are genuinely two different paths:
| What it does | What it needs | |
|---|---|---|
| ✍️ Write it myself | you write the title, the line and the landing page, and the ad is created exactly like that | nothing else: it uses no AI and not your Claude key |
| 🧪 Let Claude suggest the ideas | it takes you straight to the 🧪 Lab of that ad group, where Claude drafts several ideas and you pick which ones get created | your Claude key (cents per proposal) |
The AI path is not a separate screen: it is the usual lab, with the ad group already picked (if the group is empty, ➕ The first ad). That way the variant logic —the angle map, the verdict, the test each one opens— lives in a single place, and there is no in-between click that decides nothing.
🖼️ The image, by hand too
On the by-hand path the image can come from four places:
| When it suits | |
|---|---|
| From another ad in the account | the quickest: its image is already uploaded to OpenAI |
| Upload it yourself | PNG, JPG or WEBP, up to 5 MB |
| 🎨 Draw it with AI | when you have none at hand. You write the texts and the AI only makes the image |
| By URL | OpenAI downloads it when the ad is created |
The AI's drawing comes from what you have just written: your title, your line, the domain it leads to and the ad group's context hints —which are the ones that say in which conversation the ad will appear—. That is why you have to write the title and the line first: if they are empty, the button asks you for them.
It uses your Gemini or Ideogram key (the one you set once in «My accounts»), it costs cents and takes 10 to 30 seconds; while it draws, the page reloads on its own. If you have no key, the button is replaced by the link to set one.
The image never carries text, logos or screenshots: the card already puts its title and line on top, so an image with letters competes with them.
And what you write is not lost: uploading an image or asking the AI for one reloads the page, so the title, the line and the URL are kept as a draft and are still there when you come back. They clear on their own once the ad is created.
Three things that always happen, and are worth knowing:
- It is created paused unless you tick «create it already active». After that it goes through OpenAI's review.
- The landing page is checked BEFORE anything is created. OpenAI's API is not atomic: an ad pointing at a page that is down stays created and fails review.
- The URL is tagged on its own with its own
utm_content, so that ad can be measured separately. If you copy the image from another ad, the tag is NOT inherited: two ads sharing a tag mix their visits and their leads, and then the comparison is worthless.
✏️ Editing an ad
The ✏️ on each row opens the title, the line and its landing page. On saving, the ad goes back to OpenAI's review: it stops serving for a few minutes and, if it passes, keeps its history (no second ad is created).
- The image is not changed here: it keeps the one it has. For a different image, the honest thing is to create a new ad with it — that way comparing the two still means something.
- The URL is only touched if you change it. If you do, we check that it responds and tag it again.
- An archived ad cannot be edited: in OpenAI that cannot be undone.
- And a rejected ad is not edited here: it sends you to 🚫 Rejected, which is where OpenAI's reason and Claude's suggested fix are. Editing it behind its back would leave its case saying something that is no longer true.
Everything is logged in 🕰️ Changes, with the text it had before.
The minimum budget
OpenAI requires a daily minimum that depends on the account and that it publishes nowhere: we have measured 25 in a dollar account —in clicks campaigns too— and 15 accepted in a euro account. If your figure falls below it, their API replies with the exact minimum, the panel learns it and from then on warns you before even trying.
⚙️ Settings: all of a campaign's settings, together
As in Google Ads, 🏗️ Campaigns has a fifth tab, ⚙️ Settings: one card per campaign, and inside it their settings stacked vertically, one per line — objective, daily budget, its ad groups' bidding strategy, where it runs, platforms, attached conversion event, whether the ⏰ Schedule governs it and since when it serves. It answers "what strategy did this campaign have?" without going through three tabs, and reads the same on a phone: nothing runs off the screen. Those figures come from the panel's last read; the "View and edit" button opens the campaign sheet, which is read live from OpenAI at that moment (if the read fails, the snapshot is shown and you are told).
On the sheet, every setting sits where it is changed, and what cannot be changed says why:
| Setting | From the sheet | Otherwise, where |
|---|---|---|
| Status (⏸ / ▶) | yes, with confirmation | — |
| Daily budget | yes (subject to the minimum OpenAI requires in your account) | — |
| Strategy and bid of each ad group | yes, group by group | — |
| Platforms | yes (the same lever as 📱 Devices) | — |
| Attached conversion event | yes: chosen among the account's events | — |
| ⏰ Schedule (what hours it runs) | yes: the campaign is scheduled from here | — |
| ⏸️ Paused on purpose | yes: declared here, with the campaign paused | — |
| Objective (clicks / conversions) | no: OpenAI does not let you change it | 🧬 Clone with the other objective, linked right there |
| Where it runs (countries, regions, postal codes) | yes, in the 🌍 block: search and add, add a whole group of postal codes, remove and undo | — |
| Name and dates | no | Ads Manager |
Three things worth knowing about that sheet:
- The bidding strategy is not the campaign's, it is each ad group's. OpenAI sets it there. That is why the table says "mixed" when a campaign's groups do not all run the same, and the sheet shows it group by group, with its form alongside.
- The conversion event is the quality lever. With none attached the campaign measures nothing; with two, they add up and the same lead may count twice, and the sheet warns you. One gets attached. If the campaign optimises for conversions, changing the event restarts its learning, and the change goes into 🕰️ Changes.
- Where it runs is changed here, and it belongs to the campaign. In the 🌍 Where it runs block you see the truth in one line — not «Spain», but «1,542 postal codes · Spain» if that is what it has — and you change it with operations: search and add, add a whole province of postal codes, remove a level at once and undo the last change. It is never left with no locations (with none it would serve worldwide). What lies behind it — the three levels that exist, that there are no exclusions or group-level targeting, and that their reports only break down by country — is in section 19 of this manual.
- The schedule is set here, and it is in charge: in the ⏰ When it runs block you choose whether this campaign is scheduled, on which days and from what hour to what hour. If you activate it by hand outside its slot, it will pause it again. You can set up to six slots on the same campaign —9:00 to 14:00 and 17:00 to 21:00, for example— and it is active inside any of them; to remove one, untick all of its days. Until you give it slots of its own it uses the account's window, and the sheet says so with «(inherited from the account)». To decide them with data, ⏰ Schedule.
The status, budget and bid changes you make from the sheet are the same forms as in the tables: no cap or quarantine, with history in 🎛️ Bids and with undo.
⏸️ When you pause a campaign on purpose
To the panel, a paused campaign is a campaign that isn’t serving — and it has no way of knowing whether that is your decision or a breakdown. So it flags it: the 🩺 Audit opens the «No active campaign» critical, the 🤖 Agent makes it the first thing to do this week, and the Monday report keeps arriving. If you paused it because the numbers didn’t add up, those three alerts are noise — and an alert you know is false teaches you to ignore the ones that aren’t.
That is what the ⏸️ Paused on purpose row of the sheet is for. It only appears with the campaign paused —declaring «paused on purpose» on one that is serving means nothing— and you can write why you paused it, which is what you’ll want to read three months from now.
While ALL the account’s live campaigns are paused and declared this way:
- the Monday report and the watchdog alerts for that account are not sent;
- the «no active campaign» critical becomes «can’t be judged yet», with its reason and its date;
- the 🚦 Delivery and 💶 Money areas stop scoring: they measure what happened when the account was serving, and flagging spend you have already acted on —by pausing it— helps nobody;
- and the 🩺 Audit gives no score while it stays paused: a health figure for an account that isn’t serving would say nothing. In its place you get the ⏸️ paused on purpose badge;
- the 🤖 Agent doesn’t spend your week on that account.
Three things worth being clear about:
- Nothing is hidden. The findings are still on their screens; what is silenced is the email. And the 🏗️ campaigns table marks every declared campaign with its ⏸️ on purpose badge.
- With a single undeclared campaign, the emails keep going out. That is on purpose: a pause nobody has declared may be exactly the breakdown you need to hear about. The row itself tells you how many are left.
- You don’t have to remember to clean it up. When you activate the campaign its declaration is removed on its own; and if you activate it from the Ads Manager, the mark stops counting until it is paused again.
And what this does not do: it doesn’t touch anything in ChatGPT Ads. The campaign is already paused — this only declares why.