When AI Eats Its Tail
Using AI to Fix Google’s AI-Powered Ad Waste
TL;DR
Performance Max (PMax) campaigns are marketed as an easy, set-it-and-forget-it solution for advertisers, especially small businesses. However, after running a PMax campaign for a local construction business, we found that the majority of our budget was being spent on irrelevant mobile games and low-quality websites, rather than targeting high-intent local users. By building a clean reporting system using Google Looker and auditing our campaign placements with Gemini (Google’s own AI), we found that over 60% of impressions came from mobile apps, and only 30% landed on Google-owned platforms. We used this data to shift our strategy from PMax to a more effective manual SEM approach, and in doing so, uncovered lessons and a framework that could benefit advertisers of any size.
PMax: A Promise That Doesn’t Hold Without Oversight
Google’s Performance Max campaigns offer a compelling pitch: minimal setup, broad reach, and machine learning that optimizes your ads across channels. For small businesses with limited time and expertise, it’s easy to see the appeal.
We ran a PMax campaign for my dad’s construction and property management business, hoping to generate quality leads within our local service area. On paper, the results looked fine, a decent number of leads, solid clickthrough rates, and steady spend.
But in practice, something felt off. Many leads (90%) were outside our region, and most of them weren’t responding. We weren’t just getting bad traffic, we were getting irrelevant traffic and maybe even fraud.
Step One: Build Clarity Before Making Decisions
To understand what was happening, we needed visibility — not the cluttered, siloed dashboards inside Google Ads or GA4, but something clean and decision-ready and practical.
We built a custom reporting dashboard in Google Looker, focused on core questions: where is our budget going, what’s converting, and is it helping the business? The result was an automated report that stripped away vanity metrics and zeroed in on ad placements and outcomes.
Step Two: Let AI Audit AI
Armed with our data, we ran it through Gemini, Google’s own AI assistant, and asked it to categorize the placement data and highlight problem areas.
Here’s what we found:
60.4% of impressions came from mobile apps like Solitaire – Card Game, My Talking Angela 2, Which Dress? Left or Right, and Beat Maker Pro. These were apps where user intent was non-existent or accidental, classic examples of “fat-finger” clicks and incentivized ad engagement.
14.8% of impressions came from questionable websites like dealday.today and monkey.app, which showed all the hallmarks of low-quality “Made for AdSense” content or non-relevant platforms.
Only 30.2% of impressions came from Google-owned properties — Search, YouTube, Gmail — where we would expect to see higher user intent and better lead quality.
Despite this mix, PMax still displayed the message: “Your campaign is limited by budget.” In reality, the budget was just being routed into low-value placements that generated noise, not business.
Step Three: Shift the Strategy — and Keep the AI
We used the insights to pivot. Instead of pouring more money into PMax, we moved our budget into manual SEM. We tightened geographic controls, focused on high-intent keywords, and used Looker to monitor ongoing performance.
The change wasn’t just in impressions, it was in lead quality. Calls came from within our market. Form fills were real. And most importantly, we had full visibility into what was working and why. AI didn’t just help us uncover the problem, it helped us build a repeatable system for better decisions.
This Isn’t Just an SMB Problem
Our campaign spend was small, but the problem is systemic. Larger advertisers with hundreds of campaigns and broader budgets face the same lack of transparency, just at scale.
There’s a clear opportunity here for others:
Build a lightweight AI tool that automatically audits PMax and other campaigns for performance
Custom alerts when spend shifts into mobile games or low-quality web domains
Smart dashboards that connect ad spend to actual business impact
If we can build this for a local construction company, there’s no reason larger brands and agencies can’t implement similar AI-powered controls to protect spend and improve outcomes.
Conclusion: Build With AI, Don’t Blindly Trust It
Performance Max can be powerful, but it’s not foolproof. In our case, automation was optimized for conversions, not real leads. Engagement instead of intent. Spending money, not necessarily achieving results. By using AI internally to audit, analyze, and automate reporting, we transformed a black-box system into one we could understand and act on. That’s the future: AI not as your autopilot but as your co-pilot. Whether you’re running ads for a family business or a global brand, visibility is crucial. AI can assist, but only if you ask the right questions.
