PPC & Ads · Podcast episode · source Jun 8, 2026
Scripted Ads Reports for Claude to Read Instead of Raw Data
Get reliable numbers from AI by feeding it clean reports instead of letting it fetch everything
He says that when the AI did the job end to end it drifted, forgot numbers and moved decimal points; once it only read the prepared report, the error rate 'started dropping down dramatically'.
Per source Self-reported by the source, not independently verified.
Copy-paste prompt
Read only the file ads_report_[date].csv in /reports. Don't fetch or calculate anything outside it. Summarize spend, sales, ACOS and orders by campaign versus last week's file, list the five biggest changes, and quote the exact row behind every number you mention.
The source didn't publish its exact prompt. We wrote this starter so you can try the same workflow. Replace anything in [brackets].
What they did
A seven-figure seller who has sold online for 25 years, mostly on Amazon now, stopped asking Claude to 'read everything and give me all the numbers'. Instead a script on Google Cloud pulls the ads report and drops it into a folder, and the AI is told to read that one report and say what it shows.
How it went
He says that when the AI did the job end to end it drifted, forgot numbers and moved decimal points; once it only read the prepared report, the error rate 'started dropping down dramatically'.
We only report what the source says. Results are self-reported unless noted.
Worth knowing
An interview on Helium 10's podcast with no error-rate figures. The split matches what other sellers report: let code pull and total the numbers, and let the AI read and explain them.