I’ve tested every AI setup for Google Ads. ChatGPT, Gemini, custom GPTs, MCP integrations, you name it. Claude Code is the one that actually changed how I work.
It runs inside VS Code (or your terminal). It reads your files, writes Python scripts, queries the Google Ads API directly, and chains multi-step tasks together. No copy-pasting data into chat windows. No “sorry, I can’t access external services.” You just tell it what you want, and it does the work. That includes writing changes directly to your Google Ads account.
And you don’t need to be technical. I’ll show you the exact setup I use, and at the end I’ll link to a free guide so you can replicate it yourself.
👉 Want ready-made PPC prompts that work with Claude Code? Grab our free collection . 50+ expert-grade prompts for audits, scripts, and analysis.
What Makes Claude Code Different
If you’ve used Google Ads MCP (which connects Claude’s web interface to your account via Zapier), think of it this way: MCP gives Claude a walkie-talkie to your Google Ads data. Claude Code sits Claude down at your desk with full access to your computer.
The practical difference is enormous:
| ChatGPT / Claude Web | MCP (Zapier) | Claude Code | |
|---|---|---|---|
| Query Google Ads data | ❌ Copy-paste only | ✅ Limited actions | ✅ Full API access |
| Write & run scripts | ❌ | ❌ | ✅ |
| Push changes to Google Ads | ❌ | ❌ | ✅ As draft |
| Visit & analyze landing pages | ❌ | ❌ | ✅ Via Playwright |
| Access local files | ❌ | ❌ | ✅ |
| Persistent project context | ❌ | ❌ | ✅ |
| Reusable skills & automations | ❌ | ❌ | ✅ |
With Claude Code, you’re not limited to conversation. You have an AI assistant that can actually do the work, and remember how you like it done.
My Setup Inside VS Code
I run Claude Code inside Visual Studio Code using the official Claude Code extension. This is the best way to use it. You get a proper workspace with your files, skills, and MCP servers all visible in the sidebar.
Here’s what my setup looks like:
On the left sidebar, I’ve got my skills (reusable workflows I’ve built: deep audit, landing page audit, search term cleanup, ad copy optimizer). Below that, my MCP servers: Playwright for browser automation, Google Ads connection, Firecrawl for web scraping. And then client context files for each account I manage.
The power of this setup is that Claude can use all of these tools together. When I run a landing page audit, Claude queries the Google Ads API to find all the landing pages in the account, then uses Playwright to actually navigate to each page, screenshot it, and analyze whether there’s a message match between the ads and the landing page. It does all of that from a single prompt.
If you’re an agency, you connect your MCC (manager) account once and Claude has access to all your client accounts. No connecting them one by one.
Getting Set Up (15 Minutes)
You don’t need coding skills for this. The setup is mostly getting Google’s API credentials, which you only do once.
Install Claude Code. Open your terminal and run npm install -g @anthropic-ai/claude-code. Or install the VS Code extension directly from the marketplace (search “Claude Code”). Two minutes.
Get your API credentials. Go to console.cloud.google.com , create a project, enable the Google Ads API. Then in your Google Ads account, go to Tools & Settings → API Center and apply for a developer token. Choose “internal use” since you’re only accessing your own accounts. Google will ask you to describe what you’re using it for. This is the longest step but it’s straightforward.
Set up OAuth and store credentials. Create OAuth credentials (Desktop app) in Google Cloud, then ask Claude Code to help generate your refresh token. Seriously, just say “Help me generate a Google Ads API refresh token” and it writes and runs the OAuth script for you. Add the credentials as environment variables in your ~/.zshrc and you’re done.
Test the connection. Ask Claude to install the Google Ads Python library and pull your campaign names. When you see your campaigns appear, you’re live.
Three Things That Blew Me Away
Here are the use cases from my own workflow that convinced me this is a genuine step change for PPC management.
Account audits in seconds, not hours. I asked Claude to pull a full performance overview for a client account, and it ran multiple GAQL queries in parallel, analyzing quality scores, search term waste, budget utilization, device performance, and hour-of-day patterns. The whole thing took maybe 30 seconds. What would normally take someone the best part of a day to put together, Claude just… did.
The real surprise was when I asked it to look deeper at the ad copy and landing pages. It pulled every RSA, analyzed individual asset performance (which headlines were actually working), then used Playwright to navigate to each landing page and check for message match. It found problems I’d missed: heavy pinning on generic headlines, missing trust signals, and a location mismatch where Scotland traffic was landing on the homepage instead of a dedicated page.
It can draft ads directly in your Google Ads account. This is the part that really got me. After Claude analyzed the account, I said “Can you draft some better headlines and save them to Google Ads as a draft? I want to review before anything goes live.” And it just… did it. Wrote new RSA headlines based on actual review data and search terms, saved them as a draft ad in Google Ads, and gave me a direct link to review.
The original headlines were generic stuff. “Luxury Packages,” “For All Budgets,” “Affordable Prices.” Claude replaced them with data-driven alternatives like “No Chasing Friends for Money” (a real pain point from reviews), “500+ Happy Hen Parties” (social proof from the account), and “2026 Dates Booking Fast” (urgency based on actual booking data). Much stronger.
I always say: don’t publish anything until you’ve reviewed it in Google Ads. But the ability to generate hypothesis-driven ad tests from real data and save them as drafts? That alone saves hours per account.
Landing page audits that actually visit the pages. I built a skill that runs a full landing page audit. Claude queries Google Ads for all unique landing pages, then uses Playwright (a browser automation tool) to visit each one and analyze it. Does it pass a 3-second test? Is there message match with the ad copy? Are there trust signals? Mobile-friendly?
For this account, it caught that the Bath and Cotswolds location pages had perfect message match, but Scotland traffic was hitting the homepage. A mismatch nobody had noticed. That’s the kind of insight that actually moves the needle.
Build Reusable Skills
This is where Claude Code becomes something much more powerful than a chatbot. You can save your best workflows as skills: reusable prompts with instructions that Claude executes on command.
I’ve built skills for deep audits (pulls every available metric from the API, runs 10-12 minutes, generates a full HTML report), landing page audits, search term cleanup, ad copy optimization, weekly reporting, and competitor research. Each one is a saved prompt with specific instructions, like “always output in this format” or “use these benchmarks for scoring.”
The beauty is you build them once and they’re always there. Next time you onboard a new client or need to run a quick audit, it’s one command.
Saving context between sessions. At the end of every session, I tell Claude something like “Update our documentation for this client. I don’t want us to lose any of the context we built today.” Claude saves business context, audit findings, and account notes to a local file. Next time I work on that account, Claude reads that file automatically and picks up where we left off.
You can also create a CLAUDE.md file in your project folder with account details, targets, and conventions:
# Google Ads Setup
- Manager Account: 123-456-7890
- Client accounts: Account A (111-222-3333), Account B (444-555-6666)
- Target CPA: $35 for lead gen, $25 for ecommerce
- Always use GAQL (not the REST API)Claude reads this at the start of every session. No repeating yourself.
Workflow Automation
The real power is chaining multi-step tasks. Here’s an example. A single prompt that runs 5 steps automatically:
You say “Find wasted spend in my search terms and fix it” and Claude pulls 14,000+ search terms, identifies $3,800 in wasted spend, clusters irrelevant terms into themes, generates a Google Ads Editor CSV, and gives you a summary with projected savings. One prompt, five steps, zero manual work.
Other workflows I run regularly:
- “Run my deep audit skill on Account X”: 180-day analysis with competitive positioning, quality score breakdown, and device/time-of-day insights
- “Set up an A/B test for this campaign”: Claude drafts variant ads and saves them as experiments in Google Ads
- “Compare our ads against the top 5 competitors for this keyword”: Claude scrapes competitor ads and landing pages, then compares against your current creative
Honest Limitations
- Always review before publishing. Claude can draft ads and save them to your account, but I never let it push anything live without my review. This is a feature, not a bug
- API rate limits. For very large accounts (millions of keywords), batch your requests
- Context window fills up. Long sessions hit a limit (you’ll see something like “98% context used”). Start a new chat and ask Claude to save context first
- Output formatting needs guidance. Claude’s default audit reports look very AI-generated. Build that into your skills: “always format like this template”
- Not 24/7 monitoring. It’s on-demand. Schedule scripts with cron jobs for automated checks
For a fully automated, always-on solution, you’ll want purpose-built AI agents for Google Ads , which is exactly what we’re building at PPC.io.
Get Started
If you want to try the terminal approach:
npm install -g @anthropic-ai/claude-code- Set up Google Ads API credentials (~15 min, one-time)
- Run your first audit
If you prefer VS Code (recommended):
- Install the Claude Code extension from VS Code marketplace
- Same API credential setup
- Start building skills in the sidebar
If you’re not sure yet, start with Google Ads MCP . It’s a 5-minute no-code setup that lets you chat with your data in a browser. Graduate to Claude Code when you’re ready for full control.
I’ve used this setup to build complete audit frameworks, automate weekly reporting, generate hundreds of ad variations, and catch thousands in wasted spend that I’d have missed manually. It’s not a toy. For anyone managing Google Ads professionally, this is genuinely the biggest productivity shift since smart bidding.
👉 Want the full setup guide + pre-built PPC prompts? Grab our free collection . Everything you need to replicate my exact workflow.
What We’re Building at PPC.io
Claude Code is incredible for technical PPC professionals. But not everyone wants to live in a terminal, or even VS Code.
That’s why we’re building PPC.io, a platform that packages this power into a clean interface with specialized AI agents that handle everything from audits to bid management to creative optimization. All the capabilities you’ve seen here, but accessible to everyone.
👉 Get early access: PPC.io/waitlist