Google's Ad Stack Is Becoming an Agent You Brief. Here's the Playbook.

What actually changed at Google Marketing Live

Google spent its Marketing Live 2026 keynote making one deliberate move: it placed a Gemini agent at the middle of its advertising business, where the strategy is decided, rather than at the edges where the busywork lives. The agent is called Ask Advisor, and Google describes it as a single, unified partner that spans Google Ads, Analytics, Merchant Center, and the Marketing Platform.

The rest of the announcement arranges itself around that agent. Search ads are being rebuilt for conversational, AI-era results, a Direct Offers pilot is widening, Asset Studio now runs creative generation through Gemini, and an agentic-commerce layer, complete with a Universal Cart and new payment protocols, is designed to let software discover and buy on a shopper's behalf.

We covered the news itself in a Flash the day it landed. What follows is the part that outlasts a keynote: what it means, and what a marketer should actually do about it.

None of it, if you've been paying attention, is a surprise. Ask Advisor is not a change of direction so much as the destination a decade of Google automation has been driving toward, and it's worth pausing on the sheer scale of what now sits behind the wheel.

A decade of ceding control, in one table

The pattern is old and consistent. Every few years Google hands a larger decision to the machine, and the bargain is identical each time: you trade a measure of control for a measure of performance. Smart Bidding opened the account in 2016 by taking over the bids, and everything since has only widened the remit.

Read down the right-hand column and the arc is hard to miss. The machine took the bid, then the ad, then the targeting and the budget, then the creative, until strategy was the one thing a human still clearly owned. Ask Advisor is Google reaching for that too. Each individual handover looked reasonable in the moment; stacked on top of each other, they add up to an account where a marketer sets a goal and an algorithm decides very nearly everything that follows.

What you gain, and what you quietly give up

It would be dishonest to pretend the gains aren't real, so start there. An agent that reads across Ads, Analytics, and Merchant Center at once can catch a broken product feed that's dragging down a campaign, or a budget marooned in a channel that never converts, long before a human clicking between four tabs would spot either. It also erases the grunt work, the hours a junior buyer loses to pulling reports and reconciling numbers that never quite agree with one another.

The keynote stops at the gains. The costs it skips over come in threes, and they feed on each other.

• Visibility. When the agent connects the dots and hands you a recommendation, what you see is the recommendation, not the reasoning behind it. A black box that happens to be right most of the time is still a black box.
• Granular control. Automated systems are built to reward broad goals and to punish micromanagement, so the more tightly you constrain them, the worse they report they'll perform. The pressure, always, runs in one direction: hand over a little more.
• Independent measurement. This is the loss that actually bites, because the agent optimizes toward the conversions Google itself counts, using the attribution model Google itself sets.

The measurement trap: grading Google's homework with Google's pen

Underneath everything sits a conflict of interest that agentic advertising sharpens, and it's worth stating without euphemism. Google increasingly runs the ads, owns the analytics that grade the ads, and, through agentic commerce, is edging into the checkout where the sale finally gets counted. Then it reports back on how well the ads did.

This isn't an accusation of cheating; the problem is more structural and more stubborn than that. Platform conversion counts lean on last-click logic and self-attribution, which means they flatter the platform by design. Point an agent at that number, tell it to maximize, and it will cheerfully spend more to buy conversions that would have happened anyway, because on its own scoreboard those conversions read as a straightforward win.

The only real defense is a source of truth the platform doesn't control. Incrementality experiments, geo holdouts, and media-mix modeling all exist to answer the one question a platform's dashboard is structurally incapable of answering honestly: what would have happened if the ad had never run at all?

You've already rehearsed this with Performance Max

If any of this feels familiar, that's because you've lived a smaller version of it already. Performance Max was the rehearsal, and most advertisers have been through the discomfort at least once.

The adoption figures show how completely it won. The share of advertisers running Performance Max climbed from 60% in 2024 to 71% in 2025, more than a million businesses now use it worldwide, and it already drives close to 45% of all Google Ads conversions.

The early complaints, notably, are the same ones now forming around the agent: no visibility into placements, spend leaking toward junk inventory, and a headline conversion figure that looked wonderful right up until someone checked it against reality. What the market did next is the useful part. Advertisers didn't walk away from Performance Max; they learned to hem it in, using account structure and brand exclusions to fence the algorithm, feeding it first-party audience signals so that it chased their most valuable customers rather than the cheapest available conversion, and pressing Google, together and loudly, until it surfaced far more placement and search-term reporting than it had first been willing to offer. Ask Advisor is that same story retold at the scale of the whole account instead of a single campaign type, and the moral survives the retelling: you win control back by fencing the system, feeding it your edge, and insisting on visibility, not by refusing to use it.

The playbook: how to stay in control

None of that requires swearing off automation. It requires staying the strategist while the agent does the labor, and five moves carry most of the weight.

1. Set the guardrails before the autonomy. Give the agent the brief you'd give a new buyer on their first day: hard spend caps, a CPA or ROAS floor it has to respect, brand and placement exclusions it can't override, and clear negative signals. Autonomy inside a fence is genuinely useful; autonomy without one is just exposure with a friendlier interface.
2. Keep one scoreboard Google doesn't own. Run a standing geo holdout, or a quarterly incrementality test, or a lightweight media-mix model, and own your conversions server-side while you're at it. You need at least one number the agent can neither optimize toward nor quietly flatter.
3. Test the new formats early and small. Conversational and AI Search ads have thin competition right now, and thin competition usually means cheap reach, so there's a genuine first-mover case for getting in. Just judge the results on your own incrementality read rather than on Google's reported conversions.
4. Feed the machine your edge. Whatever you give it, the agent amplifies, which makes first-party data and genuinely good creative the two levers still fully in your hands. They're also what separates your results from a competitor pointing the identical agent at the identical auction.
5. Reskill the team toward judgment. The value stops being who can build and adjust the campaign and becomes who can brief it, audit it, and overrule it when it's confidently wrong. Hire and train for people who can read an incrementality test, not just recite an in-platform report.

The verdict

For most advertisers the agent is going to win on efficiency, and it probably should, because reconciling four dashboards by hand was never a good use of anyone's afternoon. There's an honest exception at the small end, where an under-resourced marketer running a commodity product with no measurement budget is often genuinely better off letting the agent run than fighting it with resources they don't have. The playbook above is written for everyone else, which is to say everyone with enough at stake to be able to afford being wrong.

But efficiency and effectiveness were never the same thing, and only one of them shows up in Google's dashboard. The marketers who come out ahead in the agentic era won't be the ones who refused it out of principle. They'll be the ones who let it take the labor and kept the judgment for themselves, with guardrails on one side of the account and honest measurement on the other.