If AI runs the campaign, who owns the mistake?

A marketer opens an advertising platform with a product to sell, a budget to spend and a customer to find.

They set the objective, upload creative assets and establish some boundaries. Increasingly, the machine does the rest: adjusting bids, expanding audiences, selecting placements, combining creative and even generating new variations.

If the machine is right, conversions go up and acquisition costs come down. But suppose it is wrong.

An ad reaches an audience the brand never intended to target. It appears somewhere the marketer would never have chosen. Or an AI-generated variation makes a claim nobody in the marketing department actually wrote.

Whose mistake is it?

Google's updated advertising terms, which took effect on July 1, put that question into sharper focus. Advertisers can authorise Google's automated tools to format, select or generate targets, ads and destinations on their behalf, while remaining responsible for reviewing and approving what those systems produce.

In other words, the machine can increasingly make the decision. The advertiser still puts its name on it.

“No brand should be fully comfortable here, and that discomfort is doing its job,” said Megha Agarwal, Chief Marketing Officer at Table Space.

For Agarwal, responsibility does not disappear simply because an algorithm has entered the process.

“Responsibility for an ad, its audience, or its placement stays with the brand regardless of how the decision got made,” she said. “A platform's algorithm choosing badly doesn't make the outcome any less ours to answer for with our customers.”

The question is becoming harder because marketers are delegating more decisions to these systems.

AI and Agencies

According to Fluency's Trends and Benchmarks Report, Google Performance Max (PMax) adoption among surveyed advertisers rose from 60% in 2024 to 71% in 2025, driving a 19% increase in platform-wide PMax spend alongside a massive 192% year-over-year surge in Google Demand Gen ad investment.

During the same period, Meta expanded its generative AI capabilities across Ads Manager, reporting that over 1 million advertisers were actively generating AI video and image creative monthly (with its video-generation tools seeing rapid adoption into 2025) while feed-based formats like Meta Collection ads also saw a 15% bump in advertiser usage.

Marketers are not surrendering control for the sake of automation. They are trading some of it for performance.

Agarwal's answer is therefore not to manually approve every decision a machine makes. Brands should instead establish firm boundaries around whom they will target, where they will appear and what they will be associated with, then allow automation to operate freely within them.

“Efficiency is the platform's job. Where brand trust is at stake, the brand still decides,” she says.

A blank cheque for AI?

Prasun Kumar, CMO at Magicbricks, agrees that brands ultimately remain accountable, but draws a distinction between accountability and what he calls “blind accountability”.

“There is an important distinction between delegating execution and delegating responsibility,” he said.

Kumar is comfortable allowing AI to make high-volume decisions around bids, audience expansion and creative combinations. But the more autonomy a platform receives, he argues, the more transparency it should provide.

An advertiser should be able to establish what an AI system generated, where it ran, how much was spent behind it and, at least meaningfully, why the system made the decision.

Kumar asserts, “Brands should not be expected to sign a blank cheque simply because they clicked ‘Enable AI’.”

For Amit Mathur, President - Sales and Marketing at Finolex Cables, that means responsibility cannot rest entirely with advertisers when platforms themselves are making more decisions on their behalf.

The answer, however, is not human micromanagement. “The objective is not to approve every individual decision made by AI, as that would undermine the value of automation,” Mathur said.

Instead, brands need visibility into audience expansion, placements, creative modifications and budget allocation, alongside the ability to set exclusions and intervene. According to Mathur, they need to know “what was changed, why it was changed and what impact those changes had.”

Kumar similarly identifies four requirements for greater AI autonomy: pre-launch guardrails, meaningful visibility into decisions, an asset- and placement-level audit trail, and the ability to override or roll back problematic changes.

The emerging model of human oversight may therefore be less about keeping a person inside every decision loop than ensuring somebody can define the loop, inspect it and stop it.

That becomes particularly significant as automation moves beyond media optimisation into producing the advertisement itself. Google's AI Max can customise text and select final URLs; Performance Max can assemble advertising assets; Meta's automated products can influence audiences, placements and creative while its generative tools produce advertising variations.

Yet responsibility in the platforms' terms remains extensive. Meta's Self-Serve Ad Terms, for instance, state that advertisers are “solely responsible” for their ad content, targeting decisions and placements, while its separate generative-AI advertising terms make advertisers responsible for AI-generated ad content.

That contractual allocation may not settle the legal question, however. Ankit Sahni, Partner at Ajay Sahni Associates, said platforms may place responsibility on advertisers through their terms, but cannot necessarily contract themselves out of statutory liability. As AI assumes more control over creative, audiences and placements, he said, the legal question may increasingly turn on who actually exercised control and contributed to the offending outcome.

That could depend on who supplied the underlying claim or data, configured the AI tools, could review or override the output, and materially participated in creating or targeting the ad. While advertisers could remain exposed for conceiving and approving a campaign, Sahni said a platform's position becomes harder to defend where its proprietary system generated the problematic claim or targeting decision with little meaningful advertiser control.

The scale of that transition is substantial. Google generated $81.6 billion in advertising revenue in Q2 2026, Meta $59.4 billion and Amazon's advertising-services business $19.8 billion, roughly $161 billion between them in three months.

Big Tech in Q2

Automation is not changing advertising at its edges. It is being built into its largest distribution systems.

What happens to the agency?

There is another participant caught in this redistribution of control.

Agencies traditionally selected audiences, planned placements, allocated budgets, evaluated creative and optimised campaigns. Platform automation is steadily absorbing parts of that work.

“Accountability should follow control,” said Jacob Joseph, VP - Data Science at CleverTap. He argues that agencies should remain responsible for the objectives, inputs and boundaries they establish, but cannot reasonably be held responsible for an opaque algorithmic decision they had no ability to see or influence.

That does not necessarily diminish the agency's role. It changes it.

Meher Patel, Founder of Hector, a Wondrlab company, expects agencies to spend more time defining strategy, challenging platform recommendations and independently measuring whether automation actually delivers the business outcomes it claims.

“The ability to understand when automation is working, and when to override it, will become a key agency capability,” Patel said.

Platform dynamics

Joseph describes the shift more neatly: agencies are moving from “pulling every lever to making sure the machine is pulling the right ones.”

The marketers we spoke to were therefore not arguing against AI taking greater control of campaigns. They were drawing conditions around that control.

Brands appear willing to surrender execution. They are less willing to surrender the ability to establish boundaries, understand what happened and intervene when something goes wrong.

Which leaves the industry with a question that Google's July 1 terms bring into increasingly sharp relief.

If platforms are going to make, target, place and optimise more of the advertising, who draws the boundaries within which their machines operate, and who carries responsibility when the machine crosses them?

For now, the brand's name remains on the ad.

But as Kumar puts it, clicking “Enable AI” should not amount to signing a blank cheque.