How We Use Claude + Make.com to Run Weekly Google Ads Audits Automatically

Last updated: March 2026 | By Second Step

At Second Step, we manage Google Ads accounts ranging from $4K to $32K per month. Running manual audits across all of them every week would eat 20+ hours of senior analyst time. Instead, we built an automated system that does it in under 30 minutes total, across every account, every Monday morning.

This is not a generic “AI can help marketers” think piece. This is how we actually do it, what the system looks like, what it catches, and the results we have seen after six months of running it.

Table of Contents

The Problem We Solved

Before automation, our audit workflow looked like this:

  1. Export data from Google Ads (30 min per account)
  2. Review key metrics against benchmarks (45 min)
  3. Check search terms and negative keywords (30 min)
  4. Evaluate bid strategies and budget pacing (20 min)
  5. Write up findings and recommendations (30 min)

That is 2.5 hours per account. With 8 active accounts, we are looking at 20 hours per week just on audits. At senior analyst rates, that is $3,000/week in labor.

Our automated system does the same work for roughly $25/week in tool costs.

How Our System Works

The architecture is straightforward:

  1. Monday 6 AM: Make.com triggers a Google Ads Script via webhook
  2. 6:05 AM: Script exports last 7 days of data to Google Sheets (campaign metrics, search terms, Quality Scores, conversion data, change history)
  3. 6:10 AM: Make.com reads the sheet, formats the data, and sends it to Claude via API
  4. 6:11 AM: Claude analyzes the data against our 50+ point audit checklist and returns structured findings
  5. 6:12 AM: Make.com parses the findings, categorizes by severity, and delivers via Slack

Total execution time: under 15 minutes per account. All accounts run in parallel.

The Audit Prompt (Overview)

Our Claude prompt is the most important piece. We spent weeks refining it. At a high level, it includes:

  • Account context: industry, monthly budget, target CPA/ROAS, primary conversion actions
  • Checklist: 50+ specific items to check, each with a pass/fail threshold
  • Benchmarks: account historical averages plus industry benchmarks
  • Output format: severity-rated findings with specific numbers and recommended actions

We are not sharing the full prompt here (it is our competitive advantage), but the structure above is enough to build your own version.

What We Check Every Week

Spend Efficiency

  • Search terms with spend > $50 and 0 conversions
  • Campaigns pacing more than 15% above or below daily budget
  • Cost per conversion vs 30-day rolling average (flag if >20% increase)

Quality Metrics

  • Keywords with Quality Score below 5
  • QS changes week-over-week (flag any drop of 2+ points)
  • Ad relevance and expected CTR components

Conversion Health

  • Conversion volume changes (flag >30% drop)
  • Conversion action status (active, recently added, or removed)
  • Conversion lag analysis (are conversions still coming in from older clicks?)

Competitive Signals

  • Impression share changes (flag >10% drop)
  • Average position shifts
  • Auction insights trends

Account Health

  • Bid strategy learning phase status
  • RSA ad strength and variant coverage
  • Extension coverage and performance
  • Recent changes that could impact performance
Want us to set up automated audits for your accounts?
We build custom AI audit systems for agencies and in-house teams. Every account, every week, fully automated.

Book Your Free Strategy Call →

Real Issues the System Has Caught

Here are actual findings from the last three months (anonymized):

  • Broken conversion tracking: A client’s form submission tag stopped firing after a website update. The system caught it within 48 hours based on the conversion volume drop. Manual review would not have caught this until the next monthly report.
  • $4,200/month in wasted spend: 23 search terms across one account with significant spend and zero conversions. These had accumulated gradually over months and were invisible in the day-to-day.
  • Quality Score decay: 15 keywords dropped from QS 7-8 to QS 4-5 over 8 weeks. The system flagged the trend before it significantly impacted CPCs, saving an estimated $1,800/month.
  • Budget-limited top performer: A campaign with 4.2x ROAS was consistently hitting its daily budget cap while a 1.1x ROAS campaign had unused budget. Simple reallocation increased account-level ROAS by 18%.

Results After 6 Months

  • Time saved: 18+ hours per week (from 20 hours manual to ~2 hours reviewing AI output)
  • Issues caught faster: Average detection time went from 30 days (quarterly audit) to 7 days (weekly automated)
  • Cost savings identified: $11,400/month in aggregate wasted spend across all accounts
  • System cost: ~$100/month (Make.com Pro + Claude API)
  • ROI: 114x return on tool investment

What We Learned Building This

  1. Prompt engineering is 80% of the work. Getting Claude to consistently produce accurate, actionable findings took weeks of iteration. The data pipeline was easy by comparison.
  2. False positives are the real enemy. An audit system that flags everything is worse than no system. We spent significant effort calibrating thresholds to minimize noise.
  3. Human review is still essential. The system generates findings; a senior analyst reviews them before any action is taken. This takes 15-20 minutes per account instead of 2.5 hours.
  4. Start simple, expand gradually. Our V1 checked 12 things. Now it checks 50+. Each addition was validated against manual audit results before going live.
  5. Data quality matters more than AI quality. Garbage in, garbage out. Most of our debugging time was spent on the data extraction and formatting steps, not on the AI analysis.

Frequently Asked Questions

Can I build this for my own agency?

Yes. The technical setup is straightforward if you are comfortable with Make.com and Google Ads Scripts. The hard part is the audit prompt — that takes iteration and Google Ads expertise to get right. Our media buyer playbooks cover the foundations.

Why Claude and not ChatGPT?

We tested both extensively. Claude handles structured data analysis more reliably and produces more consistent output format when given detailed instructions. For ad copy generation, they are roughly equal. For analysis, Claude wins.

What happens when the AI gets it wrong?

It happens — roughly 10-15% of findings need human correction. That is why we never automate actions directly from AI output. Every finding goes through human review. Over time, we refine thresholds to reduce false positives.

How much Google Ads experience do I need to build this?

You need solid Google Ads knowledge to write effective audit prompts and calibrate thresholds. The AI is smart, but it needs a knowledgeable human to tell it what to look for and what matters. If you can do a manual audit, you can teach the AI to do one.

Want to see our automated audit system in action? Book a free strategy call and we will walk you through a live demo. Or download our free AI playbooks for media buyers to start building your own.

Published by Second Step — a performance marketing agency powered by AI.