PPC & Ads · Blog post · source Mar 19, 2026
Ads Reports, New Ad Targets and Campaign Creation with Claude Code
Weekly ads reports per brand, an audit that finds new search terms and ASINs, and campaigns created after approval
He reports a doubling of revenue for one brand in the week after Claude's changes versus the week before, and three brands had their best months of revenue and profit afterwards.
Per source Self-reported by the source, not independently verified.
Copy-paste prompt
Pull my Sponsored Products search term report and the Brand Analytics search terms report for [ASIN] since [date of last bid change; avoid Prime Day/BFCM]. 1. Search terms in broad, phrase or ASIN campaigns with at least [2] orders that I don't target as exact. 2. Organic search terms where my product was purchased or added to cart that I don't advertise on (include misspellings). 3. Targets whose ROAS is well above or below [4], ignoring anything with less than $[10] spend. Check every suggestion against my existing targets. Don't change anything; list proposed campaigns with settings and bids for my approval.
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
Alex Willen, who owns about a dozen Amazon brands, finally got an Amazon Ads API key and put Claude Code on his ads in a three-part series. It sends a weekly report per brand via GitHub Actions. An /ads-audit skill finds search terms that already convert in broad, phrase and ASIN campaigns, organic terms from the Brand Analytics search terms report that he isn't advertising on, and similar products from the catalog. An /ads-review skill rechecks bids every 2-4 weeks, and a campaign-creation skill lists every setting, target and bid for his approval before creating anything. A small skill raises bids about 10% on new targets that haven't spent. History is stored in a database so it can look back further than the API's 90 days.
How it went
He reports a doubling of revenue for one brand in the week after Claude's changes versus the week before, and three brands had their best months of revenue and profit afterwards. Misspellings turned up as converting terms on every brand. Claude's off-the-wall targets worked too: music-playing mobiles were the best campaign for a mobile that doesn't play music, and gardening tools paid off for a seed-packet binder. The bid-raising skill saves him an hour every couple of days. But Claude can't yet handle day-to-day bid management: the instructions get too long for it to follow reliably.
We only report what the source says. Results are self-reported unless noted.
Worth knowing
Claude Code blamed Amazon for its own malformed report requests, used 90 days of data that included Black Friday, chased tiny-spend keywords with a ROAS of 55, and froze his computer by downloading the whole Brand Analytics report. His fixes, written into the skills: check why a report failed, ignore low-data targets, and filter reports at the source. Splitting one big skill into smaller ones made it more consistent.