Meta's Ad Assistant Can Now Build Campaigns From a Chat Prompt — Here's the Approval Playbook

On October 6, at Advertising Week New York, Meta announced something more consequential than another AI feature drop. The AI business assistant inside Meta Ads Manager can now plan and create a campaign, generate creative imagery and text, update targeting, and change a budget — conversationally, from a chat prompt. It drafts everything and stages it for your sign-off.

Read that again: the assistant executes. For years, “AI in ads” meant a side panel suggesting a headline variant you could accept with one click. This is different. You describe the change in plain language; it builds the campaign, rewrites the creative, shifts the budget, and hands you one job: approve, or don’t.

Meta says advertisers who act on its recommendations are seeing 5% more conversions at the same cost. Early partner tests, reported via Common Thread Collective, compressed campaign launch times by roughly 60%. That’s the gap between a tool that advises and a subordinate that acts. We’ve watched a lot of automation waves roll through paid media, and this is the one that finally changes the job description instead of just the workflow.

What the assistant actually does now

The announcement had several moving parts. Worth knowing each one:

• Conversational campaign building — describe what you want (“launch a prospecting campaign for the holiday sale, $50/day, broad targeting”) and the assistant drafts the campaign structure, creative, and settings inside Ads Manager.
• Creative generation and edits — it generates ad imagery and text, and the expanded Ads Creative Studio can now create new video concepts and variations from your existing assets: change the hook, adjust audio and clip length, add text overlays and captions, or change the end scene. Over 60% of time on Facebook and Instagram is spent with video, so this is aimed squarely at the Q4 creative crunch.
• Audience discovery via Customer Lifecycle Strategy — you specify a campaign is for acquisition, and the assistant recommends how to exclude existing customers instead of you assembling those exclusions manually. Anyone who has maintained suppression lists knows how much fiddly work that saves.
• Conversational budget and targeting changes — “shift $20/day from this ad set to that one” now happens through a prompt instead of a five-click drill-down.

This lands two months after Meta’s August AI creative suite, which was mostly about generating on-brand creative with brand memory and an in-platform approval workflow. October’s announcement is the other half: the assistant now also runs the media side. Creative generation plus campaign execution in one chat box.

Three things worth letting it do

First drafts of video variations from your statics. The mistake we see most often: brands sitting on 40 product photos while spending $2,000+ on a video shoot for Q4, when Meta’s Advantage+ can now generate video from those static assets in minutes. The output isn’t a masterpiece — it’s a volume play. You’ll get five hook variations to test by Friday instead of one polished video by November. Review the first three seconds of each one yourself, though. The hook is still a human decision.

Acquisition exclusions through Customer Lifecycle Strategy. Excluding existing customers from acquisition campaigns is exactly the kind of maintenance work advertisers postpone — and then pay for in wasted spend. This one is a genuine improvement. Set it up once, verify the exclusions are correct, and stop babysitting suppression lists.

Creative diagnosis inside Creative Studio. The expanded Studio “helps them understand what is working and why,” per Meta’s announcement. The why is the useful part. We tried this workflow on a client store and the analysis correctly flagged that hook variations with on-screen text in the first two seconds were driving the majority of that account’s creative wins. We’d been guessing. The data wasn’t. That said — treat the diagnosis as a starting point for your next brief, not gospel. (And if your creatives are dying from repetition rather than poor hooks, that’s a creative refresh cadence problem, not a targeting one.)

Two things you never hand over

Budget authority. Set your daily cap before you prompt the assistant to do anything. The assistant doesn’t know your cash flow, your inventory position, or your margin. We learned this the expensive way on a client account: an auto-applied “recommended” budget change doubled a test budget the night before a big inventory drop, and the traffic went nowhere profitable. Your rule should be simple — the assistant proposes budget changes, a human approves every single one, and it never touches spend while you’re asleep.

The offer and the brand voice. AI-generated copy converts when the underlying offer is real and dies when it isn’t. No assistant can fix a weak offer; it can only dress one up in sharper words. Same with voice: the generated variations are fine, but if your brand has a distinct personality, the final pass belongs to a human who knows where the line is.

Here’s the bad practice, named plainly: the approve-everything button-pusher. If your entire media-buying workflow becomes clicking “approve” on whatever the assistant suggests, you’re not using AI — you’ve outsourced your judgment to Meta’s incentives. Meta’s incentives are ad spend. Yours are margin. Those are not the same thing, and they never will be.

The approval workflow we actually recommend

Don’t wing it. Run a cadence:

• Monday: Prompt the assistant with hard constraints — daily budget cap, audience definition, offer terms. Then review every draft campaign against a ten-minute checklist: budget correct, audience correct, offer correct, brand voice intact. Nothing goes live unchecked.
• Midweek: Let it generate the week’s video and hook variations. Watch the first three seconds of each. Keep what lands, kill what doesn’t.
• Friday: Run the fatigue check. If CTR has dropped 30% or more over two weeks, the creative is tired — refresh it before next week. The heavy lifting of an Advantage+ setup still depends on feeding it fresh creative; the algorithm can’t save stale assets.
• Always: Keep a human kill rule. Any ad running 50% over your target CPA for three days gets paused by a human. Automation finds the winners. Humans bury the losers.

This takes about ninety minutes a week once it’s routine. The advertisers who’ll see the biggest gains from this assistant are the ones currently drowning in manual setup — the ones spending four hours building what the assistant drafts in four minutes. If your operation is already tight, expect a smaller lift.

Treat the headline numbers with respect, not reverence

The 5% more conversions figure is Meta’s own, measured on advertisers who act on recommendations. The 60% faster launch time comes from early partner tests, not a randomized benchmark. Neither is a lie; neither is a promise either. Your mileage will vary — especially if your account is small, your pixel is young, or your creative is weak. The assistant amplifies what you feed it. Feed it a sharp offer and clean data and it moves fast. Feed it mush and it generates mush at machine speed.

What we’re still unsure about

Two honest caveats. First, rollout: Meta ships in waves, and like the August suite, this won’t be visible in every Ads Manager account on day one. If the assistant features aren’t there yet, that’s a staged rollout, not an account problem. Second, edge cases are untested in public: how the assistant handles restricted ad categories, and whether conversational budget changes respect account-level spend limits in every scenario. We wouldn’t find out the answer to the second one with real money on a Friday night.

Q4 is the wrong season to learn a new tool by trial and error. Set the guardrails first — budget caps, a review checklist, a kill rule — then let the machine do the boring parts. The approval step is the whole job now. Do it properly and the assistant is the best junior media buyer you’ve never had to train. Do it on autopilot and it’s just a faster way to spend money.