Agents In Use
What’s new todayLast update
1005 additions

Claude Code exports reviewed Google Ads copy

Lau reports a Claude Code RSA workflow that checks campaign inputs against brand and product skills, refines ad copy with him and exports a manually reviewed Google Ads CSV.

FROM THE SOURCE

Original prompt

Shared by the source author. See source notes for context.

/rsa

What happened

Lau reports a Claude Code RSA workflow that checks campaign inputs against brand and product skills, refines ad copy with him and exports a manually reviewed Google Ads CSV.

Reported by Anthropic interview with Austin Lau

How the agent did it

  1. Start the custom responsive-search-ad command in Claude Code.
  2. Supply campaign data, existing copy and keywords.
  3. Apply brand, product-accuracy and ad-writing skills.

Where a person stepped in

Lau reviews and refines the copy, with human-written brand and product references. The source explicitly retains manual review before Google Ads upload.

Austin Lau, a growth marketer at Anthropic, builds a Claude Code workflow for responsive search ads. It starts with his custom slash command and asks for campaign data, existing copy and keywords. The assistant cross-references those inputs with skills covering brand voice, product accuracy and ad-writing practices.

Lau then refines the proposed headlines and descriptions conversationally. He evaluates whether the value proposition fits the audience, whether the tone is right and whether the message stands out. The source emphasises that approved examples and copy were created with product-marketing and copywriting colleagues before the agent used them as references.

Once the copy is ready, Claude Code assembles an upload-ready CSV containing campaign and ad-group fields, fifteen headlines and four descriptions per ad. Manual review precedes upload. The article is Anthropic's interview with its own marketer, so reported time savings are not an independent benchmark or evidence of improved ad conversion. The useful workflow gives the agent both performance context and explicit product-and-voice constraints, then preserves a human creative judgment step. The original prompt shown below is the actual command used to begin this process, not the complete skill definition.

Prompt
Original prompt

Source: Blog · Anthropic interview with Austin Lau

Frequently asked questions

What inputs shape the drafts?

Campaign data, existing copy and keywords, checked against brand voice, product accuracy and RSA-practice skills.

What does the export contain?

Campaign and ad-group fields, fifteen headlines and four descriptions per ad in a CSV file.

Does the agent upload without review?

No. The source says the export undergoes manual review before being uploaded to Google Ads.

More details

Remaining steps

  1. Refine proposed headlines and descriptions with the marketer.
  2. Export the CSV and manually review it before upload.

About this account

Claude Code retrieves campaign inputs, applies reusable skills, iterates on copy and assembles a structured upload-ready file under the marketer’s direction.

Firsthand account by Anthropic interview with Austin Lau; original publication date verified against Primary article date January 26, 2026. Outcomes are the author’s reports, not independently reproduced. Original custom command shared by Austin Lau; the full skill and campaign inputs are not reproduced.

Credit: Anthropic interview with Austin Lau

Source published 26 Jan 2026

Source checked 14 Oct 2026

Tools and topics

Claude Code · Agent Skills · Campaign data · Google Ads CSV

Responsive search ads · Brand accuracy · Reviewed export

Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.