# How to ask better questions about your ads

> Ask AI about your ads with four parts: the client, the time window, the metric and what to compare with. Weak vs strong questions and a worked rupee example.
- **Author**: Marxx
- **Published**: 2026-11-07
- **Category**: AI
- **URL**: https://marxx.ai/library-guides/how-to-ask-better-questions-about-your-ads

---

A good question to an AI about your ads names four things: whose account, which dates, which number and what to compare it with. "How are ads doing?" gets a vague answer. "Northwind's Meta cost per lead, last 7 days vs the 7 before" gets one you can act on.

## The four parts

Question = client + time window + metric + compare-to

| Part | Weak | Strong |
|---|---|---|
| Client | "our ads" | "Northwind Escapes, Meta only" |
| Time window | "recently" | "last 7 days, 26 Oct to 1 Nov" |
| Metric | "performance" | "cost per lead and number of leads" |
| Compare-to | nothing | "vs the 7 days before" or "vs our INR 400 target" |

In a Slack channel that belongs to one client, the channel already supplies the client. You still name the platform.

## Five weak questions, rewritten

1. "How are ads doing?" becomes "What was Meta spend, leads and cost per lead last week vs the week before?"
2. "Which ad is good?" becomes "Which three ads had the lowest cost per lead in the last 14 days, with at least 10 leads each?"
3. "Is Google working?" becomes "Which Google Ads campaigns had a cost per conversion above INR 600 this month?"
4. "Why is CPA up?" becomes "Cost per lead rose this week. Which ad sets drove the rise, and what changed in their CTR and CPM?"
5. "What should we do?" becomes "List three changes to test next week, each with the number that supports it."

The "at least 10 leads" filter stops the AI from crowning an ad that got one cheap lead. [CTR](/glossary/ctr) is the share of people who clicked after seeing the ad. [CPM](/glossary/cpm) is the cost of 1,000 views.

## A worked example

Say Priya Shah wants to know if Northwind Escapes' new reel helped. Her first try: "Is the new reel good?" The answer is a paragraph of praise with no numbers.

Her second try uses all four parts: "Northwind Meta ads, 26 Oct to 1 Nov. Compare the cost per lead of Monsoon reel 02 with the account average for the same week."

The answer: the reel got leads at INR 311, against an account average of INR 397. The gap is INR 397 − INR 311 = INR 86 a lead. On 50 leads, that is INR 86 x 50 = INR 4,300 saved. Now she has a reason to move budget, and a number to check in Ads Manager.

## Follow-ups that help

- **"Show the numbers behind that."** Ask for the figures, not just the verdict.
- **"Exclude anything with fewer than 10 results."** Small samples mislead.
- **"Split by placement" or "split by age".** Only when the first answer points there.
- **Stay in the same thread.** The AI keeps the context of the earlier question.

For more method, see [how to analyze ads with AI](/guides/how-to-analyze-ads-with-ai). For a list to copy, see [ChatGPT prompts for Facebook ads](/templates/chatgpt-prompts-for-facebook-ads). We make Marxx, which you can ask in a client's Slack channel, see [how to ask AI about your ad account in Slack](/guides/how-to-ask-ai-about-your-ad-account-in-slack).

## FAQ

### Should I ask long or short questions?

One clear question at a time works better than a long list. Ask the follow-up in the thread once the first answer arrives.

### Why does the AI give different answers to the same question?

Small wording changes shift the answer. Fix the time window and metric in the question, and the numbers stay steady.

### Can I ask "why" questions?

Yes, but ask for the numbers that support the reason. An AI can sound sure about a cause it cannot see, such as a competitor's sale.

---
- [More AI articles](https://marxx.ai/library-guides/category/ai)
- [All articles](https://marxx.ai/library-guides)