The Ad Ran and Nobody Approved It

Most coverage of this change treats it as a settings problem. You move some things, your negative keyword lists survive, your budgets carry over. That is not the real loss.

Google is migrating Search campaigns using campaign-level broad match or standalone automatically created assets to AI Max during September 2026, with Dynamic Search Ads to follow in early 2027. Under AI Max, text customization writes new headlines and descriptions from your landing pages and your existing ads, then sends them into the auction. Nobody approves the specific promise that runs under your brand's name before it serves.

For about seventeen years at Full Sail University, I built and led creative teams. You learned fast that approval is not decoration. A producer, director, or brand-systems designer does not sign off on work because a checklist says to. They sign off because they have looked at the specific thing, understood what it claims, and verified that the claim is true.

Approval is a judgment step. Someone has to look at what goes out the door and own it.

An ad is a promise. If you have ever stood in a property and watched someone tour it because an ad brought them there, you know this viscerally. In the attainable housing communities we build across the country, someone signs a lease on a place because an ad promised something about it. The promise had better be true. When they tour and find the ad overstated the finish or mislocated the amenity, the damage is done.

What actually changes, given that machines have shuffled ad copy for years?

Not the shuffling. Responsive search ads have mixed and matched approved headlines for years, and every brand team made peace with that. What changes is authorship. Text customization writes new copy from your landing pages and existing ads, so the sentence a customer reads may be one nobody on your team ever wrote.

The other change is quieter and worse. Pinning was how a brand team locked a required claim into a required position. Google says that with final URL expansion on, pinning is not respected. The guardrail is still in the interface. It just stops holding.

Why is human approval no longer available at the finish line?

When an algorithm tests combinations continuously and humans run one meeting about next month's campaign, approval at the finished ad becomes theater. The combination running at three PM on Wednesday might not have existed at three PM on Tuesday. A new winner emerged from testing, and by the time you see it, it is already live. You cannot give approval to something you did not know was being requested.

You can still turn these settings off. The platform has simply priced that choice so you will not make it: opt out and watch performance fall, leave it on and take the growth. The real alternative is to move approval upstream, before the algorithm writes anything.

What does approval look like when the ingredients are everything?

You stop approving finished ads. You approve building blocks and constrain how they combine. Every headline fragment has to be defensible on its own. "Move-in ready" is true of the delivered building and false of the phase still under construction, and the algorithm reading those pages cannot tell you which one it is quoting.

Every description has to survive the same test. So does every claim on the landing pages, because those pages are now source material for the ad copy, not just the place the click lands. A page that oversells one floor plan becomes a headline that oversells the whole community.

Then the pairings themselves become a constraint, and that part cannot live in a document. Google Ads has no setting that reads your brand guidelines. Assets that must never run together do not share an ad group. Text customization stays off for pages you cannot vouch for. Two true claims that make one false promise never get the chance to meet.

I came to brand systems from production. The discipline was about governance: these type sizes go here, this color means that, this icon works with these words. The system kept outputs true to strategy because humans followed the rules and spotted when something was off before it shipped.

An algorithm does not spot the edge cases. It finds them and serves them if performance is better. Your system has to be tight enough that no edge case can exist. No combination can accidentally promise something false.

Who is responsible when the algorithm runs something that should not have run?

You are. The platform designed this on purpose. Its argument is that the performance was too good to leave to human judgment: more conversions, lower cost per acquisition, faster adaptation. From the platform's standpoint, the answer to every objection is data. But that design moved responsibility for truth from platform to brand.

If the promise does not survive the tour, if someone felt lied to, your name is on the ad, not the platform's. The answer is not to reclaim control over every keyword. What you need is a brand system tight enough, and built into the account structure rather than a slide deck, that an algorithm cannot write its way into a lie on your behalf.

The approval meeting moves from after the build to before it. You are signing off on the boundaries, not the output. That is a smaller job on the calendar and a much larger one to get right.