Claude and AI
The 15 questions worth asking AI about your ad account
With data connected, AI answers in seconds what used to take an afternoon of spreadsheets. The questions that produce decisions and the ones that produce nice text.
Also available in: Português · Español
Asking an AI "how do I improve my campaigns" returns text that is correct, generic and useless — because the question has no data inside it.
With the account connected, that changes: the questions start having specific, verifiable, numbered answers. These are the ones that produce the most decisions.
Quick diagnosis (the Monday-morning ones)
1. "What changed last week versus the week before?" Instead of hunting for the change, ask for the list. It is the question that saves the most time across the whole week.
2. "Which campaign spent the most without producing a sale in the last 30 days?" Straight to the idle money. It is the question that most often finds something actionable on first use.
3. "Is any ad above frequency 3 with falling CTR?" Both fatigue signals together, without building the comparison by hand.
4. "What is the real cost per sale by campaign this month, using the CRM?" The metric that decides budget, not the platform's conversion count.
5. "Was there any day in the last 15 when conversions dropped to zero?" Catches broken tracking before it becomes a lost month.
Quality and funnel
6. "What is the close rate by campaign, not just the CPL?" Separates campaigns that bring volume from campaigns that bring customers.
7. "What is the average time between a lead arriving and first contact?" The number that is not media's and pays media's bill.
8. "Which leads from the last 90 days have no source filled, and how much is that?" Coverage. Without it, any ranking that follows is opinion.
9. "What is the gap between what the platform claims and what the CRM recorded?" Your account's discount factor, which calibrates every future reading.
Creative
10. "Which creatives brought the most revenue in the last 60 days, not the most leads?" The inversion that changes which piece gets budget.
11. "Which angle is working best: testimonial, price or pain?" Groups by type instead of by piece, and it is what gives direction to the next production run.
12. "Which creative is taking more than 50% of the ad set's delivery?" Concentration is risk: when it tires, the whole ad set drops at once.
Planning
13. "Compare spend, sales and revenue by channel over the last 90 days." The table that decides the budget split and exists in no ads manager.
14. "Which audience had the best close rate, counting actual sales?" Not the best CPL — the best close rate.
15. "What should I be looking at that I have not asked about?" The open question, at the end. It works better after the specific ones, when context is already in the conversation.
The questions that are not worth asking
"Are my campaigns good?" With no criterion, the answer will be polite and empty. Ask against a target: "is cost per sale above the $600 we agreed?"
"Write 10 headlines for my ad." That is text generation, not analysis — and an AI with no context on your audience produces generic headlines. If you are going to do it, give it the piece that worked and ask for variations on its angle.
"What is the best time of day to advertise?" The generic answer exists and is useless. The useful version is "which hour had the best cost per sale in my account over the last 90 days".
"Should I raise the budget?" The AI does not know your margin, your cash flow or your quarterly target. Ask what it can know — whether the channel is saturating — and decide yourself.
How to ask so the answer is trustworthy
Ask for the number with the conclusion. "Tell me which campaign is worst and show spend, leads and sales for each" beats "which campaign is worst". You verify, and verifying is what makes the tool trustworthy.
Give the period. Without one, the answer picks an arbitrary window and you compare different things without noticing.
Ask for coverage whenever the answer is a ranking. "How many of those sales have an identified source?" The answer changes how much the ranking is worth.
Be skeptical of a conclusion with no number. If the answer says something improved and shows no series, ask for the series.
The 10-minute routine
Monday morning, in this order: question 1, question 2, question 5, question 3.
Changes from the week, idle money, tracking alive, creative tiring. Four questions, ten minutes, and you enter the week knowing where to act — instead of opening four dashboards and forming an impression.
For the full routine that follows these questions, see a weekly optimization routine.
Frequently asked questions
Can AI analyze my ad account?
It can, if the data is reachable through a connector. Without access to the real numbers, any answer is a generality written with confidence.
What is the difference between asking AI and opening the dashboard?
The dashboard shows what it was configured to show. AI answers the question you just had, including cross-cuts nobody built in advance.
Can I trust the AI's analysis?
Trust the numbers it read from the source and be skeptical of conclusions you could not verify. Always ask for the raw figure beside the interpretation.
Does AI decide for me?
It should not. It shortens reading and cross-referencing; moving money stays your decision, with the business context it does not have.