AI campaign drafts in Meta Ads Manager: what changed in October 2026

Since October 6, 2026, according to Meta, work an AI agent stages through Meta's ads MCP server lands as a draft in Ads Manager. People and agents can co-author the same campaign, either side can publish, and every action is gated by advertiser-set permissions and explicit approval before anything goes live. Meta also named ten companies that have built the server into their products. For anyone running accounts, this is good news: it separates preparing from publishing. But a draft isn't a complete guardrail. Here's what changed, what to check before you approve, and where the draft-plus-approval pattern stops.

What Meta announced

The news came at Advertising Week New York on October 6, 2026, in the section aimed at large advertisers and agencies. Three points matter:

• Drafts in Ads Manager. What the agent builds through the MCP server shows up as a draft in the account, and people and agents can work on the same campaign.
• Explicit approval. Each action runs through the permissions the advertiser set and needs a yes before it goes live.
• Ten integrations. AdManage.ai, Birch, Blip, Databricks, Innovid, Kochava, Perplexity, Reportei, Snowflake and Triple Whale have integrated the server.

This builds on earlier steps. Meta's ads MCP server has been in open beta since April 29, 2026, and on July 16, 2026 it gained rules: anyone with full control of a business portfolio can govern what an agent may do on the ad account. For the bigger picture on Meta's server versus running or renting one, see Meta Ads MCP server.

What's still in testing

On the same day, Meta talked about the AI business assistant inside Ads Manager. Keep these apart:

So the path "an outside agent prepares, you approve in Ads Manager" exists today. The path "Meta's own assistant does it all by conversation" is still in testing. How Meta's built-in automation differs from an AI reading your account from outside is covered in Meta Ads automation vs an AI assistant.

Why draft plus approval is the right guardrail

Anyone who has watched an AI read "raise it to 80" as "raise it 80%" knows the value of a pause between instruction and execution. A draft puts that pause in the right place: inside Ads Manager, with the full campaign on screen, not in a chat summary.

It fixes three problems at once:

1. Misreads. You see what was built, not what the AI says it built.
2. Accountability. It's clear who approved. At an agency, that ends the "the AI did it" conversation.
3. Division of labor. The agent does the grunt work (structure, naming, duplicating ad sets) and you handle the judgment calls.

What to check before approving a draft

Approving without looking turns the guardrail into decoration. Run this list, in order:

• Objective and optimization event. Leads, purchases, messages: is it what you asked for?
• Budget and type. Daily or lifetime, and the right amount, decimal point included.
• Audience and location. Country, city, age. A wrong radius burns money fast.
• Placements. Automatic or manual, as the strategy calls for.
• Creative and copy. The right file, with no claim the client hasn't signed off on.
• URL and UTMs. A working, tagged destination; without UTMs, the sale disappears from your CRM. See UTM parameters for ads.
• Dates. Start and end, and not launching overnight on a promo nobody approved.

Where the draft stops

Draft plus approval beats direct execution. It still has limits worth knowing:

• Approval is binary. There's no cap: approving US$ 50 or US$ 5,000 takes the same click. A tired approver lets mistakes through.
• What Meta described may look different in your account. The announcement talks about advertiser-set permissions; how that shows up can vary. Before you rely on it, test on a small budget which actions become drafts and which don't.
• It only covers Meta. If your AI also touches Google and TikTok, each platform has its own rules, or none.
• Approving isn't measuring. Once the campaign is live, someone still has to come back and compare before and after.

The most common mistakes people make once they let AI act are in AI campaign optimization mistakes.

How Vazante does the equivalent

Vazante landed on the same principle by a different route: instead of a draft in Ads Manager, anything created starts paused, and every action asks for confirmation in the chat.

The practical difference: with Meta's drafts, you review on the Ads Manager screen; with Vazante, you confirm in the chat and check the paused campaign in Ads Manager before reactivating. Both work as long as someone actually looks.

Next step

If your AI already builds campaigns, adopt the draft rule even where it isn't enforced: nothing goes live until someone has run the checklist above. If it doesn't yet, start read-only and open up write access gradually, as laid out in AI that executes in the ads manager. To do that across Meta, Google and TikTok with the same guardrails, see connect Meta Ads to Claude: the first 7 days cost US$ 2.99 on any plan.