Turn ad performance data into fresh variants

A growth team used Claude Code to spot weak ads, draft new copy, and move it into review-ready workflows.

Done with Claude Code· Intermediate

What they did
Anthropic’s growth marketing team gave Claude Code a CSV with past ad results. The agent picked out ads with low CTR and enough impressions, then two sub-agents produced new Google Ads headlines and descriptions that matched the platform’s length rules. Those drafts were written back into CSV form, labeled for staged testing and human QA, and then reused in a Figma plugin; a small Meta Ads MCP server also supplied campaign data to Claude Desktop.
How it went
The team said a campaign refresh fell from about two hours of work to around 15 minutes. They also said creative output went up by roughly 10x, and the Figma plugin could make as many as 100 versions at once. Human review still mattered, especially because the model could invent claims or metrics.
Tips
Check every line for brand voice, legal wording, character count, and line breaks before anything is uploaded. Keep an experiment ID attached to each version so you can trace results later.

Try it yourself with Claude Code

Look at the ad data in [file path]. Find the entries that are weak on [metric] and have at least [minimum impressions] impressions. For each one, draft [number] fresh [platform] headline ideas and [number] description ideas that fit the platform’s size rules, then put the results in a separate CSV alongside the source ad and a unique experiment ID. Stop and ask me to approve before any upload, submit, send, delete, or payment step. After approval, [next approved step or stop].

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on dev.to ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.