Skip to content
agentusehub

Generate ad variations and pull campaign data for growth marketing

Anthropic's growth marketing team used Claude Code to flag underperforming ads from a CSV and generate hundreds of Google Ads variations, plus a Figma plugin and Meta Ads MCP server.

Time saved
Not reported
Setup time
Not reported
Running cost
Not reported
Cost range
Not reported

Use case overview

Anthropic's growth marketing team used Claude Code to flag underperforming ads from a CSV and generate hundreds of Google Ads variations, plus a Figma plugin and Meta Ads MCP server.

Reported experience

  1. What they did

    This is a third-party summary of Anthropic's own document. Claude Code reads a CSV of past ads and metrics and filters for low-CTR ads with enough impressions. Two sub-agents then write new headlines (30 characters) and descriptions (90 characters). The variants go back to a CSV, tagged for staged rollout and human QA. A Figma plugin swaps the copy into templates, and a small Meta Ads MCP server feeds campaign data to Claude Desktop.

  2. How it went

    Ad copy creation for a campaign refresh dropped from about two hours to about 15 minutes. Creative output rose roughly 10x. The Figma plugin produces up to 100 variations at once. Outputs still need human QA for brand voice and legal compliance.

  3. Worth knowing

    Models can fabricate claims or metrics, so check character counts, line breaks and legal wording before upload. Attribution only works if each variant is tied to an experiment ID.

Example prompts

Try with Claude Code
Read the ad performance CSV at [file path] and flag the ads that are underperforming on [metric, e.g. CTR or CPA]. For each one, write [number] new headline and description variations that fit [platform] character limits, and save them to a new CSV with the original ad referenced. Stop when every flagged ad has variations; do not upload anything to the ad platform.