Google Ads in the AI era: automation you control vs automation that controls you

The short answer: Google Ads has become an AI product — Smart Bidding, Performance Max and AI-generated creative now do work that account managers used to do by hand. The machine is genuinely good at bidding; it is structurally bad at telling you the truth about it. Running Google Ads well in the AI era means feeding the automation clean conversion data, fencing it with guardrails, and keeping your own measurement outside Google's homework-grading loop.

We run paid accounts for booking businesses and webshops, and the job has changed shape: less bid-fiddling, more data engineering and auditing. Here's what that looks like in practice.

What did AI actually change inside Google Ads?

Three things, each bigger than the last. Bidding went first — manual CPCs lost to Smart Bidding years ago, and today target-ROAS and target-CPA strategies read signals no human can see. Campaign structure went second: Performance Max collapsed search, shopping, display and YouTube into one black-box campaign where Google decides placement and creative mix. The query itself is going third: AI Overviews and conversational search change what a "search term" even is, and advertisers see less and less of the raw queries they're paying for.

The pattern across all three: more performance, less visibility. Google's automation optimizes toward whatever you measure — which makes your measurement the last lever you fully own.

Where the black box earns its keep — and where it eats budget

Smart Bidding with solid conversion data genuinely outperforms manual management; we've stopped pretending otherwise. Performance Max can find converting inventory a human wouldn't have tested. That's the earn side.

The eat side is just as real. PMax will happily "convert" people who were already coming — brand traffic dressed up as performance. Broad match plus automated bidding will spend confidently on queries you'd never approve if you saw them. And AI-generated ad copy optimizes for clicks, which is not the same as optimizing for buyers; a headline that promises what your business doesn't deliver converts twice — once in the auction, once into a refund.

The defense is structural, not heroic: exclude brand terms from PMax so it can't grade its own homework, feed real conversion values (not form-fills weighted like revenue), review search term and placement reports on a schedule, and write the creative yourself even when the machine offers to. This is how we run accounts — the machine bids, humans audit.

The measurement problem nobody mentions in the keynote

Google grades Google. When the platform reports conversions it also attributes them, models them, and de-duplicates them by rules you didn't write. In the AI era the honest setup is layered: platform conversions for the bidding algorithm (it needs them), server-side or first-party tracking for your books, and a monthly reconciliation against actual revenue — bookings made, orders shipped. Where the numbers diverge, the divergence is the report. Our rule from the four habits applies doubly here: numbers you can audit, or you stop paying.

What to do this quarter

Audit where PMax spends (brand vs non-brand), fix conversion values so the algorithm optimizes toward margin rather than volume, move remarketing lists onto first-party data, and test one thing the AI can't do: an offer worth advertising. Automation amplifies the account it's given — clean inputs and honest measurement beat any bidding trick you'll read about this year.

Want an account audit with the uncomfortable numbers included? Tell us what you're running.