PPC in 2025: How to control the algorithm (before it burns your budget)

It’s 2025. Your Google Ads dashboard is a sea of green check marks. CPA looks fine. ROAS is “within target.” On paper, you’re winning.

But behind the dashboards, lead quality is collapsing, and you’ve just spent five figures teaching AI-powered PPC campaigns how to find the wrong customers—faster.

In a recent webinar with CXL, co-founders of Paid Media Pros, Michelle Morgan and Joe Martinez, put it succinctly:

“If you’re not watching what the algorithm is doing, you’re not managing campaigns, you’re just funding experiments for Google and Meta.”

This is the automation paradox: PPC automation can scale results or burn through budget at record speed. The difference? Whether you’re actively controlling automation or letting it run until your CFO pulls the plug.

In this post, we’ll break down what an effective AI PPC strategy really looks like in 2025, from controlling black-box algorithms and preventing budget burn to training automation with better data and knowing when to take the controls back.

The rise of “black box” PPC campaigns

Google’s Performance Max. Facebook’s Advantage+ Shopping. Microsoft’s AI bidding. Each year, platforms like these hand you shinier automation features while hiding more of what’s actually happening inside.

Black box PPC means limited visibility into:

• Who’s actually seeing your ads;
• Which placements are eating your budget; and
• How the algorithm is interpreting your audience signals.

As Michelle Morgan, co-founder of Paid Media Pros, put it:

“There has yet to be any sort of fully automated campaign that I have ever run that doesn’t need guardrails and regular check-ins. Left alone, they always drift further and further away from your target customers.”

Performance Max isn’t malicious; it just needs guidance. If you don’t feed it the right signals, it will happily “optimize” you into spending 70% of your budget on Display placements with sky-high bounce rates. And because you’re locked out of most reporting, you won’t notice until sales calls start drying up.

The two-week trap: Why early wins don’t last

One of the nastiest quirks of AI-driven PPC is what Joe Martinez calls the Performance Cliff.

“You’ll see really good performance in the first couple of weeks, then a clear drop-off once Google thinks it’s learned enough. We see that constantly.” — Joe Martinez

Here’s why: Early PMAX campaigns lean heavily on your remarketing audiences. Results look great because you’re retargeting warm traffic. Then the algorithm expands out, gets “confident,” and performance tanks.

This honeymoon trap is dangerous because marketers report glowing dashboards up front, only to discover two to three weeks later they’ve been training the algorithm on junk.

When automation backfires: Conversion spam in the wild

Automation’s biggest blind spot is that it optimizes based on whatever signals you give it. Bad tracking, weak creatives, mislabeled conversions; these aren’t small issues. They’re training data, and the platform will amplify whatever you give it, good or bad.

One B2B campaign saw PMAX “optimize” into a weekend flood of bot signups. Even after blocking entire countries, the system kept pushing spend to fraud sources. Another account hit ROAS targets on paper until the sales team flagged that the majority of “leads” were unqualified.

As Martinez warned:

“Without offline conversion data, the platform just keeps optimizing for conversion spam. And that’s not a great thing in general.”

This is why AI isn’t set-and-forget. Without human intervention, your campaigns will chase the path of least resistance and deliver the worst possible customers at scale.

Full PPC automation isn’t a free pass

Smart bidding and automated targeting still rely on the quality of your inputs: conversion tracking, exclusions, creative assets, and clean CRM data.

It needs human oversight. But now, instead of hand-tuning bids every day, you’re designing PPC campaign guardrails, curating inputs, and editing outputs.

“Ad copy has gone from me spending hours reading long landing pages to just pasting them into ChatGPT and editing the output. I’m no longer a copywriter, I’m an editor.” — Michelle Morgan

That’s the new PPC job description: Automation does the grunt work, humans make it usable.

Where AI in PPC actually works

Not all PPC automation is the enemy. Used correctly, AI is a power tool for what no human can do manually at scale.

Best use cases for AI in PPC in 2025:

• Automated bidding: Managing thousands of SKUs or keywords in real time;
• Predictive analytics: Spotting seasonal surges before competitors.;
• Creative testing: Running dozens of variations across formats simultaneously.

The key is clean data. 

• Lean on Google Smart Bidding only if you’ve got airtight conversion tracking;
• Feed Meta’s system with high-quality server-side CAPI events only, rather than click clutter.

Automation isn’t strategy, it’s scale.

The control levers you still have

Even inside a black box, you’re not locked out. You just need to know which levers are still yours to pull.

Guardrails marketers can still control:

• Negative keywords and placement exclusions to stop budget leaks;
• CRM loop integration so you’re optimizing for revenue, not spam;
• Audience exclusions and first-party data enrichment to sharpen targeting;
• Manual overrides for seasonal or high-value campaigns.

Sometimes the control lever is the exit door

Not every campaign is worth saving. In many cases, the smartest move isn’t tightening PPC campaign guardrails, it’s moving your budget entirely.

Morgan shared how her agency is shifting client spend:

“As Google becomes less profitable for lead gen, we’re pulling spend into LinkedIn and Microsoft, because that’s where the targeting still works better.”

Why bleed money trying to force-fit Google’s black box when LinkedIn can target job titles directly or accept inflated CPCs in Search when Microsoft offers cheaper inventory with better intent?

And don’t forget the upstarts. Roku, Disney, Reddit, Quora, TikTok—they’re not “experimental” anymore. They’re often cheaper, easier entry points with surprisingly robust targeting.

Martinez’s advice:

“Go where your target audience is and max that out, even if it’s just $50 on Quora or Reddit. Don’t waste thousands trying to force one keyword or one platform to work.”

PPC algorithm control sometimes means cutting the cord.

Feeding the algorithm high-quality inputs

Even with guardrails, bad inputs will sink you.

4 rules for training PPC algorithms in 2025:

1. Tag every conversion that matters.
2. Import offline and CRM-qualified events.
3. Build rich audience signals with first-party data.
4. Rotate strong creative variations to avoid “teaching” the algo one trick.

“The creative is the training data for the ad system. If all your assets look the same, you’re teaching it one trick and expecting it to win every race.” — Michelle Morgan

One PMAX campaign went from losing money to exceeding ROAS goals simply by restructuring the product feed, tagging high-margin SKUs, and killing deadweight assets hogging impressions.

Human skills that still matter

As automation creeps in, human judgment becomes more (not less) valuable. In 2025, the edge won’t come from PPC performance monitoring tools; it’ll come from these three skills:

• Interpreting outputs: Spotting when “good” dashboard metrics hide junk pipeline;
• Scenario planning: Knowing what to test when the data is incomplete or misleading;
• Cross-platform PPC optimization: Managing Google, Microsoft, Facebook, Amazon, and niche platforms without letting each run in a silo.

“PPC used to be hard because you had to make endless manual edits. Now it’s hard because you need to think strategically at a high level. The tactical work is easier. The thinking is harder.” — Michelle Morgan

Case snapshots: Human oversight in action

• B2B SaaS: PMAX optimized for “free trials,” but 80% of signups were students. 

The fix: Imported CRM-qualified leads and added job title exclusions. CPA flat, pipeline quality doubled.

• E-commerce: Smart Shopping shifted budget to low-margin SKUs. 

The fix: Restructured product groups, boosted bids on high-margin products, excluded loss leaders. ROAS up 35%.

• Lead Generation: Facebook Advantage+ flooded the CRM with junk forms. 

The fix: Introduced custom conversion for post-demo attendance. Lead volume fell 20%, but cost per qualified lead improved 50%.

Automation didn’t save these campaigns. Human oversight did.

Automation is a power tool, not a pilot

Platforms don’t optimize for your business; they optimize for theirs.

Google optimizes for clicks. Meta optimizes for form fills. Microsoft optimizes for reach. None of them care whether your sales team can close the lead—that’s on you.

Your job isn’t to outwork the machine, it’s to guide it.

• Keep the inputs clean.
• Keep the guardrails tight.
• Keep your hand on the brake.

Because the biggest risks you can take in your AI PPC strategy this year isn’t under-spending. It’s leaving the algorithm on autopilot and trusting it to land your campaign.

If you’re ready to stop funding Google’s experiments and start running your own, check out the full webinar with Paid Media Pros or dive into CXL’s B2B Paid Media Advertising courses.