You do not need Ads Manager or another app for this. Ad Creative Diagnosis is already in JeremyAI. You give it one existing ad, the offer, where the ad sends people, and the numbers for a clearly defined period. It says whether that ad can be the control, what to keep, and three next versions that each change one thing.
This sample is not a proven winner. One period of spend and purchases is a starting point. It is not proof the ad is working.
Watch the walkthrough
The walkthrough is one run inside JeremyAI. The worker opens, the sample ad goes in, and the review and variations come back in the same chat. If the player does not load, open the walkthrough here.
This example shows a first-pass review. Check the variations against the real offer, the page, and any proof you can substantiate before you launch them. Your answer may differ.
What this does
Takes one existing ad, not a request for ten new hooks.
Separates the facts in the numbers from the assumptions about why the ad may be working.
Names what to preserve: the offer, delivery, audience, destination, and substantiated claims. Does not add unapproved claims.
Writes three variations. Each changes one major variable and labels that change.
Gives a test plan with a deciding metric tied to purchases or lead quality, not clicks alone.
Stays in JeremyAI. It does not change campaigns.
When to use it
Use it when you have an ad you want to build on, and you want the next versions to test one change at a time. Also use it when performance is softening and you need to know whether the next move is a new hook, or a problem in the page, the offer, or the proof.
If clicks or leads look fine but purchases, qualified calls, or customer quality are weak, ask for that mismatch before you treat the ad as the control.
Before you start
Signed in at
jeremy.utari.ai.Ad Creative Diagnosis visible in Workers. It is a shared JeremyAI worker, so you do not build it.
The offer, the price, who it is for, and what the customer actually receives.
The current ad: the copy, the script, or a short description of the visual.
Where the ad sends people, and the action you want.
Numbers for a clearly defined period: spend, results, and the attribution window you use. Mark unknown fields as unknown.
Only the proof you can substantiate. If you have none, say no proof yet.
Do it
Open Workers and select Ad Creative Diagnosis. Click Start Chat. Do not open settings.
Paste the ad and the context. The walkthrough uses the sample prompt below. For your own ad, keep the shape and replace the fields.
Read the review before you use the variations. Write down whether the data supports using this ad as the control, what to keep, and what is assumed.
Read the three variations. Each should change one thing. If the reply says the constraint is the page, the offer, or missing proof, that is the result. Do not paste new hooks over that answer.
Read the test plan. The deciding metric should be the purchase or the qualified outcome, not CTR alone.
How to read the sample
The walkthrough uses a sample, labeled as sample. The offer is a $97 training for coaches and agencies. It teaches an ad-auditing framework and gives prompts. It is a training, not software and not done-for-you. The ad goes to cold Meta traffic and sends people to the training checkout.
The current ad says: Before writing ten more hooks, work out where your ad loses the buyer. Learn a framework for reviewing the creative, the offer, and the next step, with prompts you can apply to your own campaign. See the training.
For one completed 14-day period, the supplied data is $970 in spend and 20 attributed purchases. That is $48.50 per attributed purchase. Frequency, CTR, CPC, refunds, and a comparison period are unknown. No proof is supplied.
Those numbers describe one window. They do not establish profit, and they do not explain why people bought. The worker should keep the diagnosis provisional and should not invent testimonials, guarantees, scarcity, or delivery details.
The prompt from the walkthrough
Copy the block below for the same run as the video. For your own business, replace the offer, the ad, the period, and the counts. Mark anything you do not know as unknown.
I have an ad I want to build on. This is sample data, not a proven winner. Review why it may be working, then create three controlled variations.
Business and offer: a $97 training for coaches and agencies. It teaches an ad-auditing framework and provides prompts.
Delivery model: a training. Not software. Not done-for-you.
Platform and audience: cold Meta traffic.
Next step: the training checkout. The action we want is a purchase.
Current ad:
Before writing ten more hooks, work out where your ad loses the buyer. Learn a framework for reviewing the creative, the offer, and the next step, with prompts you can apply to your own campaign. See the training.
Performance:
- Reporting dates and timezone: one completed 14-day period. Timezone not supplied.
- Attribution window and result definition: the team's stated attribution window. Purchases are attributed purchases.
- Spend: $970
- Frequency, CTR, and CPC: unknown
- Leads, purchases, or booked calls: 20 attributed purchases
- Downstream quality, shows, sales, or refunds: unknown
- Comparison period or control: none
Proof and claims we can use: no proof yet.
First, tell me whether the available data supports using this ad as a control. Separate facts from assumptions. Identify what to keep and what data would change your diagnosis.
Then write three variations. Change one major variable per variation and label the change. Preserve the offer, delivery, audience, destination, and claims unless I explicitly ask otherwise.
Do not invent results, testimonials, time commitments, guarantees, scarcity, or delivery details. Leave missing claim slots in brackets.
For each variation, explain the hypothesis and what result would support it. Give me a test plan with a deciding metric, downstream quality checks, and a review point based on the data available.
Keep this to analysis and draft assets. Do not change campaigns.
A prompt for your own ad
Use this when the sample is not your offer. Leave unknown fields as unknown.
I have an ad I want to build on. Review it before you write variations.
Business and offer: [what we sell, price, and target customer]
Delivery model: [what the customer receives]
Platform and audience: [platform, cold/warm/retargeting]
Next step: [destination and the action we want]
Current ad:
[paste the copy, script, or screenshot description]
Performance:
- Reporting dates and timezone: [period]
- Attribution window and result definition: [details]
- Spend: [amount]
- Frequency, CTR, and CPC: [values or unknown]
- Leads, purchases, or booked calls: [counts or unknown]
- Downstream quality, shows, sales, or refunds: [details or unknown]
- Comparison period or control: [details or none]
Proof and claims we can use: [verified results, or no proof yet]
First, say whether this data supports using the ad as a control. Separate facts from assumptions. Say what to keep.
Then write three variations. Change one major variable per variation and label it.
Preserve the offer, delivery model, audience, destination, and approved claims. Do not invent results, testimonials, guarantees, scarcity, time commitments, or delivery details. Leave missing claims in brackets.
For each variation, give the hypothesis and the result that would support it. Deciding metric must be the purchase or qualified outcome, not clicks alone.
Do not change campaigns.
Write your own in the prompt builder
If you do not want to start from the sample, open the prompt builder. Pick Ad performance, then Find the cause first. Paste the ad, the period, and the counts. Leave proof blank if you do not have it.
In What I need back now, paste this so the builder does not stop at a general diagnosis:
Review the existing ad, explain what to preserve, then write three controlled variations. Change one major variable per variation, label the change, and give a hypothesis and test plan.
Copy the result into Ad Creative Diagnosis. The builder marks empty boxes as [MISSING] and tells Jeremy not to assume the fix. Nothing you type there is sent until you paste it into JeremyAI.
Bring the test back
When the test has enough data to review, paste this into the same chat.
Here are the results for the control and each variation: [data].
The reporting dates, attribution window, and result definitions are [details].
Compare the actual customer outcome, not just clicks. Account for conversion lag and sample size.
What does this support, what is still uncertain, and what should we keep or test next?
If the worker is missing
Refresh the worker list. If it is still not there, email [email protected] and include the account email. Do not build a replacement for this walkthrough.
The how-to for this worker is Ad Creative Diagnosis Worker.
Support
Stuck on a step? Email [email protected]. Include the account email, the worker name, and a screenshot of the chat.