20 prompts to analyze your campaigns with AI (and the data you need)

What should I actually ask ChatGPT or Claude about my ad campaigns, and what data does it need to answer well?

An AI assistant will answer any question about your campaigns, including the ones it has no data for. That is the trap in every list of marketing prompts: a well-worded question sent to an assistant that does not have the numbers produces a fluent, confident guess. So each prompt below comes with the data it needs. Copy the prompt, check that the data is there, and replace the bracketed parts with your own.

Before the prompts: four rules

• Name the window and the clock. "Last week" means different days in different time zones. Say "Monday September 14 to Sunday September 20, Mexico City time".
• Name the attribution setting. "7-day click, 1-day view" and "7-day click" give different numbers for the same campaign. Say which one the data uses.
• Give the goal. A target cost per sale, a target ROAS or a margin lets the assistant judge instead of describe.
• Ask for the table first. "Show the figures in a table, then your conclusion, and list any data you did not have."

If your accounts are not connected yet, see how to connect Meta Ads and Google Ads to Claude; if the idea of connecting them is new, start with what MCP is.

Performance and budget

1. The weekly summary

"Summarize [dates] across Meta, Google and TikTok: spend, purchases, revenue and cost per purchase by campaign, compared with the previous week. Table first, then the three changes that matter most." Data: campaign-level spend, purchases and revenue per platform for both weeks.

2. Spend without sales

"Which campaigns increased spend by more than 20% in [dates] without a matching increase in purchases? Show spend and purchases for both periods." Data: daily campaign spend and purchases for both periods.

3. What changed between two periods

"Compare [period A] with [period B]. For the five campaigns whose cost per purchase moved most, list what changed: budget, bid strategy, audience, creative or placement." Data: campaign performance for both periods plus change history, if the connection exposes it.

4. Budget pacing

"At today's pace, which campaigns will overspend or underspend their monthly budget of [amount]? Show spend to date, daily average and projection." Data: month-to-date spend by campaign and each campaign's budget.

5. Cost per purchase against target

"My target cost per purchase is [amount]. Rank campaigns from furthest above to furthest below target for [dates], with spend and purchases." Data: campaign spend and purchases, and your target.

Creative and fatigue

6. Ad fatigue

"Which ads show fatigue in [dates]: frequency rising while click-through rate falls for at least seven days? Show both series." Data: daily frequency and CTR at ad level. Without frequency, the assistant cannot answer this and should say so.

7. The best creatives by cost per purchase

"Rank ads with at least [N] purchases in [dates] by cost per purchase, and describe what the top five have in common: format, length, first line, offer." Data: ad-level spend and purchases, plus creative names or thumbnails.

8. Video versus static

"Compare video and image ads in [dates] on CTR, cost per purchase and purchase rate after the click." Data: ad-level results with the ad format.

9. Copy that converts

"Group my ads by their first line of text and compare cost per purchase by group for [dates]." Data: ad-level results with the primary text.

Audiences and targeting

10. Prospecting versus retargeting

"Split [dates] into prospecting and retargeting campaigns and compare cost per purchase and the share of purchases from new customers." Data: campaign results with the audience type, and a new-customer flag per purchase.

11. Breakdown by placement and device

"Break down Meta results for [dates] by placement and device. Which combinations spend without purchases?" Data: results with placement and device breakdowns.

12. Geography

"Which states or cities have the lowest cost per purchase in [dates], and which spend the most without purchases?" Data: results with a region breakdown and, ideally, your store's orders by shipping state.

Checking claims against real sales

These four are the ones that change budget decisions, and they need your store's orders connected or attached, not only the ad accounts.

13. Claimed versus recorded

"For [dates], compare the purchases Meta and Google each claim with the orders my store recorded for the same days, in [time zone]. Table first." Data: platform-reported purchases and the store's orders for the same window.

14. The ceiling test

"Add up every platform's claimed purchases for [week] and compare the total with the orders my store settled that week. If the total is higher, how much is it over?" Data: each platform's claims with default settings, and settled orders. The excess is double counting, because no platform can claim an order that does not exist.

15. Refunds by channel

"For [month], what share of revenue credited to each channel was later refunded or cancelled? Show gross and settled revenue." Data: orders with refund and cancellation status and the channel each was credited to.

16. New versus returning

"For [dates], how many purchases credited to each campaign came from first-time customers according to my store's full order history?" Data: orders with customer history, not only the platform's new-customer column.

Search campaigns

17. Search terms that spend without converting

"List the search terms that spent more than [amount] in [dates] with no conversions, and suggest negative keywords." Data: the search terms report with cost and conversions.

18. Brand versus non-brand

"Split Google Search results for [dates] into brand and non-brand terms using [brand words], and compare cost per purchase." Data: search terms or keywords with cost and conversions, plus your brand terms.

Reporting and next steps

19. The leadership summary

"Write a one-page summary of [month] for leadership: spend, verified sales, cost per sale by channel, what changed and what we will do next. Use only figures from the data and mark any estimate as an estimate." Data: everything above, ideally reconciled against orders. We cover the format in how to present marketing results to leadership.

20. A prioritized action list

"Based on [dates], list the five changes with the largest expected impact on profit, with the evidence for each and what data would confirm or reject it." Data: performance plus margin, or at least your target ROAS.

Make them repeatable

Keep the prompts that work in a shared document, run them on the same weekday with the same settings, and compare this week's answer with last week's. The value of AI analysis compounds when the questions stay the same and only the data changes.