AI went from suggesting your marketing to doing it. Who's in control?

Most weeks, AI news for marketers is just a list of new features. Last week was different. From Oct. 30, the tools that make, place, and run marketing gained a new category: they act autonomously, and you don't sign off on each action. Before that, some agents acted by default but still left room for your influence (with Google, for example, you can update your own content to shape the summary). Some act only after a human approves. Some act only on the nodes or applications you approve. Now that there's a new category, it's a good time to review where the approve button sits and whose finger is on it.

Here's what happened, and what I think it means.

Four kinds of "yes"

1. Yes by default. OpenAI said on Oct. 21 that ChatGPT ads run in 89 countries and that more than 100,000 advertisers have used them.¹ Its help center set creative enhancements to turn on automatically for new and existing campaigns on Friday, Oct. 30, with an opt-out in Ads Manager; existing account- and campaign-level opt-outs stay in place.² Search Engine Journal reported the same day that the default was on.³ That's AI that can rewrite headlines and descriptions, translate text, resize images, and send clicks to a different page on your site. When the setting is on, ads may run "without requiring you to review or approve each version."² Advertisers can switch it off. For those who did not opt out, the starting point is on.

2. Yes without asking. Google published a help page for "Review summaries," which use AI to condense reviews of your business into a short line inside your Search ad. The summaries are "positive only," and advertisers "cannot directly edit or customize the generated text."⁴ They run in English in the US, UK, Australia, and India, and PPC Land notes Google has tested versions since December 2025.⁵ Part of your paid ad is a Google AI summary of customer reviews that you can't edit. For the wider shift away from the click, see Your next customer may never visit your website.

3. Yes, after a human checks. On Oct. 27, Runway made Runway Ads available on every paid plan. It connects to Meta, Google, and TikTok, studies what's working, and makes on-brand variants. Runway says every ad needs a team member's approval before it goes live.⁶ The same day, Sprinklr opened a beta of six marketing agents, from ideas to budget moves. Its compliance agent checks content against brand and regulatory rules and flags exceptions for human review.⁷ Notice the difference: one model reviews everything, the other reviews what the machine flags.

4. Yes, inside your business systems. Also on Oct. 27, IBM launched tools that give each AI agent a governed identity, limit its access, and let companies find, sandbox, and shut down agents across vendors.⁸ And Oracle and OpenAI announced Oracle AI Workspace, which lets employees use ChatGPT to "complete approved business workflows" across Oracle's business applications, data, and cloud services.⁹ The word that keeps showing up is "approved."

“Why are we giving decisions over to AI? The honest reason is volume.”

Why this is happening

The honest reason is volume. At the Dell Technologies Forum in Denmark in October, Said Akar said token usage has grown 320-fold. Runway says its own team went from under 100 ads a week to roughly 1,000 with the same people.⁶ That's the company's own figure, not an audit, but the direction is clear. When output grows tenfold, the old review process, where someone reads every version, breaks. So vendors are redesigning review. Some move it to the exceptions, some to a default you can opt out of, and some remove it.

None of this is bad on its face. Most marketing teams are stretched, and nobody wants to sign off on the 40th resized banner by hand. But "fewer approvals" is a business decision, and platforms are making it for you through their defaults.

What I think it means

The approve button is the new brand guideline. For years, brand control lived in a PDF: logo rules, tone of voice, approved claims. It lives in settings, permissions, and approval flows across a dozen tools. If your guidelines aren't turned into rules a machine can follow, and a named person owns them, the machine will fill the gaps.

Data readiness decides how good the autopilot is. Every one of these tools learns from what you feed it: your brand kit, your landing pages, your reviews, your performance data. Messy inputs give you confident, scaled mistakes. In my view, getting data and workflows in order is still the hardest and most valuable part of AI in marketing, and these launches make it more urgent, not less.

Use AI as the first line of quality control, not the last. Sprinklr's compliance agent points at something useful: AI is good at systematic, repeatable checks, like whether the language matches the market or a claim matches the approved list. Let it catch the obvious, every time. Then keep human judgment for what's new, risky, or truly on-brand.

Be clear about who does what, and why. AI can break down silos between creative, media, legal, and data teams, because one agent can work across all of them. But that only helps when everyone knows which decisions belong to the machine, which belong to a person, and which person. Without that, you don't get fewer silos. You get a new one nobody owns.

What it means for you

• Get your data and brand rules ready first. The tools above only work as well as the brand kit, claims list, and performance data you feed them.
• Write down the AI–human handoff. For each platform, name what the machine can change, what a person must approve, and who that person is.
• Put AI on first-line QC, not final sign-off. Let it catch the repeatable mismatches; keep people for judgment, risk, and anything that carries your name.

Your checklist

You don't need a new platform to act on this. You need an hour and the right people in the room.

• Map your yeses. For each ad platform and AI tool you use, write down what can go live without a human click. Include defaults like ChatGPT's.
• Name the untouchables. Prices, legal lines, regulated claims, and core brand promises. Make sure nothing rewrites them without sign-off.
• Turn guidelines into rules. If a tool asks for a brand kit, give it a real one and keep it current.
• Check what customers see. Search for your own ads in each market and screenshot anything odd, including AI review summaries you didn't write.
• Keep an agent list. Know which agents you run, what they can touch, who owns them, and how to pause them.
• Test in pairs. Run AI-made work next to human-made work on one campaign before you scale it.

AI will keep getting better at making marketing. The part that stays ours is deciding what's good enough to carry our name. Late October was a reminder to set that up on purpose, before the defaults set it up for us.