Meta continues rolling out AI features faster than most advertisers can evaluate them. There’s no warning, no staged rollout announcement, and one day you’re in Ads Manager and something new is just there. I recently had the opportunity to discuss this topic with my friend and fellow digital advertising expert, Lisa Raehsler of Big Click Co, as a guest on her Ai Ads & Beyond podcast. We had a thoughtful conversation about where AI is genuinely helping advertisers, where caution is still warranted, and how agencies can navigate the constant stream of new features Meta continues to release. Thank you, Lisa, for the invitation, the insightful questions, and the engaging discussion.
As someone managing accounts across a range of client types—from strict brand licensing situations to “use it however you want” relationships—I’ve had to develop a consistent framework for deciding what to test, what to disable immediately, and what I’ve actually changed my mind on over time. Here’s where I currently stand.
Advantage+ Targeting: The One I’d Tell Everyone to Test
If there’s one place I’ve completely shifted my thinking, it’s here. I used to pride myself on finding niche audiences, such as specific interest stacks, tight behavioral combinations, and strategic lookalikes. And for a while, that worked. Lookalikes especially had a rock star moment around COVID. But then Advantage+ targeting rolled out and, after testing it across enough accounts to feel confident, I can say it performs better than manual targeting more often than not.
The honest reason is simple in that Meta has data on people that I don’t have, and that my clients don’t have. When I was building manual audiences, I was very likely leaving customers and revenue on the table without knowing it and having the deeper data access to tap into.
That said, I don’t advise blowing up a working account structure to switch over. If you have a pre-existing manual setup that’s converting well, run Advantage+ targeting in a parallel test and let the data tell the story. When you have a clear winner, migrate budget gradually. If you’re starting fresh, start broad and let Meta find the people most likely to accomplish your objective. I’d also add that don’t be surprised if manual targeting quietly disappears at some point. They keep pushing it further down in the interface, taking away options, and making it less efficient to utilize. When Meta starts hiding things, it’s usually not long before they phase them out entirely.
The AI Features at the Ad Level: This Is Where It Gets Complicated
The ad-level AI features under Advantage+ creative enhancements are a different story. Some show promise, though, I disable most of them.
The core issue is when you create a new ad, Meta auto-enables every AI feature available for that unit and campaign objective. It’s like playing whack-a-mole with toggle buttons. They’re not always easy to find and you have to keep scrolling and expanding sections. And if you try to turn something off, Meta will sometimes ask “are you sure?” and flash a stat like “62% of advertisers who enabled this saw better results.” Without knowing what industry, objective, or audience they’re citing, that number means very little. It’s a pressure tactic, so pay attention to what you’re pressing.
Even more frustrating is that sometimes you’ll disable features, launch the ad, and Meta will re-enable a few of them anyway. I’ve had clients come back to me saying they’re seeing AI features live on ads I had turned off. The reason is that buried in Advertising Settings, there is also a setting feature there’s an override that essentially gives Meta permission to use a small percentage of your impressions or budget to test whatever they want, regardless of your ad-level settings.
The ones I typically disable:
- Text improvement overlay: It takes your primary copy and rearranges pieces of it above your actual ad text. The result rarely reads well or looks intentional. I haven’t seen anyone who loves this feature. Sure, it can draw more attention and maybe that’s a strategy you want to test, but for some other brands who are more rigid with what their ads look like, they may not want to enable this.
- Relevant comments: If you have a loved brand with strong organic engagement, it can work, as it pins the top comment to your ad, which creates social proof. But “top comment” means most engagement, not most positive. If there’s any controversy or price sensitivity around your product, you may not want that pinned automatically.
- AI background generation: I’ve seen a sandal paired with a snowy background. There’s potential here as the technology improves, but right now I’m not comfortable testing it on a client’s dime as there is no control or insights.
The ones I’ll test or keep on:
- Catalog value props and preview overlays: For e-commerce brands, the small detail callouts or hover previews can add useful context without hurting creative integrity. Low risk, worth trying.
- Subtle image animations: If a client only has static images and limited creative resources, the Advantage+ animation that adds a gentle zoom or rotation is worth enabling. That subtle movement does help stop the scroll, and it’s not intrusive enough to hurt brand presentation.
The Creative Question: Where I Draw the Line
The one area I’m not ready to hand over to AI fully is creative. Not yet.
Ad copy variations have gotten better. A year ago they were stuffing copy with emojis and hashtags, which nobody uses on Meta or most platforms any longer, but they’re still generic. They lack the nuance of understanding a brand’s voice, a specific audience’s language, or the strategic angle you’re trying to hit. I’ll use AI-generated variations as a starting point sometimes for ideas, but they need heavy editing before they’re ready to run.
AI-generated image creative has a similar problem. The proportions aren’t reliable. I had a client selling a toy truck, and the AI lifestyle images had a kid holding it at wildly inconsistent sizes — massive in one image, tiny in another. AI image expansion can produce uncanny results, especially of products with wrong backgrounds, people extended in physically impossible ways. That’s the kind of thing that creates customer complaints and returns, not conversions. Until the technology is more consistent, I’m keeping a human editor in the loop on anything that touches client-facing creative.
Meta AI Connectors (Claude and ChatGPT Integration): Genuine Potential, Still Early
This is one of the newer developments I’ve been exploring. Meta’s AI Connectors, now in open beta, let advertisers manage campaigns directly through AI assistants like Claude or ChatGPT — no API keys or developer credentials required for the MCP version. The CLI version takes more setup, involving Python, and I enlisted my husband (a web developer and owner of Code For Broke) to help me navigate it initially.
What I’ve found useful so far is the reporting and diagnostics angle. Being able to quickly ask “what was the performance difference between these two campaigns?” without pulling and formatting a report manually is genuinely time-saving. Similarly, checking when a pixel last fired across multiple accounts in seconds is the kind of operational efficiency that makes sense for agencies managing large portfolios. The connector supports four core areas: reporting, campaign management, catalog management, and signal diagnostics.
I’m moving carefully on the creative and campaign editing side. Ads Manager is what I know, and I don’t want to rush into making changes through a new interface before I understand the guardrails. But I do see real potential here, particularly for bulk operations and fast account diagnostics.
The Mindset That Makes All of This Work
The thing I keep coming back to, whether I’m working with agency clients or teaching inside Social Ad AKademy, is this: you are still in charge of the results.
AI did not launch your campaign. AI is not reading the account at 8am to check what happened overnight. You are the auditor. You are the decision-maker on what gets enabled or disabled. The features are tools, and some useful, some not ready yet, some actively risky for certain brands. Your job is to understand them well enough to make that call for each client and each campaign. Meta will keep rolling out AI features. They may keep auto-enabling them. They may nudging you to leave them on. The advertisers who get the best results aren’t the ones who enable everything or disable everything and they’re the ones who actually understand what they’re turning on and why. That hasn’t changed but the platforms change constantly.