I've been staring at Advantage+ creative previews for the better part of a year now, and I'll say the quiet part: I can't always tell anymore. Not reliably. Not fast. A background is swapped, a product shot gets a new angle, a headline gets rewritten in a voice that sounds suspiciously like the account's tone-of-voice doc — and I have to actually stop and check the AI label to know what happened. That's new. Two years ago the tells were obvious: warped hands, nonsense text in the background, lighting that didn't match. Now the tells are gone in a huge percentage of outputs, and the industry is still operating like it's 2023.
What actually changed
Meta didn't sneak this in. It's been building toward this systematically. Advantage+ creative tools reach customers with exactly the right creative that will make them convert by instantly producing and enhancing diverse ads in multiple formats — image, video, text and audio. That's not a features list, that's a mission statement, and Meta has been shipping against it aggressively.
The generative pieces specifically: background generation creates new AI backgrounds behind your product, image expansion uses generative fill to extend creative into ratios like 9:16, text generation produces AI variations of your text, and image animation turns a still into motion. Layer on top of that AI-generated music tracks matched to the emotional tone of the creative, new for 2026, plus an Image-to-Video tool that converts up to 20 product images into polished multi-scene video ads. This isn't "enhance what you shot." This is "build the whole thing from a folder of product photos." And it's not a fringe experiment. More than 4 million advertisers now use Meta's generative AI tools, up from 1 million six months prior. Over one million advertisers used Meta's AI tools to create more than 15 million ads in a single month in 2025. This is production infrastructure now, not a sandbox.
Why "just turn it off" isn't the safe move you think it is
I get the instinct. A brand's visual identity is the thing you spent years and real money building, and handing pieces of it to a black box feels reckless. Fair. But blanket-banning AI creative enhancements has real costs, and pretending otherwise is its own kind of malpractice.
- Meta has structurally shifted the game toward creative volume. Andromeda, Meta's updated ad ranking system, evaluates thousands of times more ad variants in parallel than its predecessor, shifting the primary performance lever from audience targeting to creative diversity. If you're feeding the system three static images and refusing every variation tool, you're competing with one hand tied.
- These enhancements are increasingly the default, not an opt-in extra. Since February 2026, all new Sales, Leads, and App Promotion campaigns launch with every Advantage+ Creative enhancement turned on by default. Ignoring the review question doesn't mean AI creative isn't running on your account. It almost certainly is.
- Duplication silently resets your settings. Every time you duplicate a campaign or ad set, AI features can silently re-enable — the most frequently reported issue in 2026 — so audit AI toggles immediately after duplication rather than assuming settings carry over.
So "we don't use AI creative" is often just wrong, technically, on a decent chunk of accounts that think they've opted out.
Why blind trust is worse
The failure mode on the other end is just as real, and it's not hypothetical. There's a well-documented case of a DTC apparel brand that had AI enhancements turned on without intending to, and Meta generated images featuring a completely wrong product for a target market the brand doesn't even sell to. That's the nightmare scenario in a sentence: your ad account quietly generating creative for a product line that doesn't exist, running against real budget, until someone happens to notice. Even media buyers who work with this daily are cautious. Big brands remain skittish about AI-generated creative, and one media buyer said most of her clients want to retain control because they've put so much time and effort into crafting their brand. Another agency operator put it plainly: he's found Meta's creative AI tools perform worse than the third-party systems he uses, and experimenting regularly still consistently produces worse results. That's not luddism. That's people who touch this every day telling you the tool is uneven.
And "uneven" is the actual state of things right now — not universally bad, not universally good. Not every enhancement carries the same risk. Standard adjustments (cropping, brightness, contrast) modify what you already gave the system. The generative ones invent things that never existed in your original asset. That distinction is the whole ballgame, and most review processes I see don't even acknowledge it exists.
The workflow that actually makes sense
Ban nothing outright. Trust nothing outright. Build a review layer that matches the actual risk profile of each enhancement type, because they are not equivalent:
- Split your enhancements into two buckets before you touch a toggle. Bucket one: crops, resizing, brightness, image expansion into new aspect ratios — modifications of an asset you already approved. Bucket two: background generation, full text rewrites, image animation, anything that creates content that didn't exist in your source file. Bucket two gets a human set of eyes every single time, no exceptions, no matter how good the last twenty outputs looked.
- Use the preview before you publish, every time. The functionality exists specifically so you don't have to find out what shipped after it's already live. You can see a preview of some samples of your AI-generated and enhanced creative before you publish, though previews are examples of possible versions and don't represent an exhaustive list of all possible variations. Translation: the preview shows you some of what could happen, not everything that will. Treat it as a sample audit, not a guarantee.
- Assign one person to own the "does this still look like us" check. Not a committee, not "the team will catch it." A named person, on a cadence, whose job includes literally opening the Ads Manager AI preview panel and asking whether the output would embarrass the brand if a customer's aunt saw it. This is unglamorous work. It's also the work that prevents the wrong-product, wrong-audience disasters.
- Audit after every duplication, migration, or account restructure. This is the single most common way brands end up running AI creative they never consciously approved. Build it into your launch checklist, not your "if we remember" list.
- Know Meta's own disclosure rules and don't skip them. Meta requires disclosure on ads containing AI-generated or AI-modified content, and skipping this step is now one of the most common reasons for ad rejections. This isn't just a compliance footnote — it's actively costing accounts approval time.
- Reserve full manual control for your hero creative, and let AI loose on the long tail. Your flagship campaign, the one built on research and testing and a brand story you can defend — keep that human-authored end to end. Your volume plays, your fifteenth variant testing a different hook, your placement-specific resizes — let the machine work. This is, illustratively, roughly the split I'd recommend for most mid-size accounts: tight control on the handful of assets carrying the brand, wide latitude on the dozens of variants that exist purely to find a winner.
The uncomfortable part
The reason this needs a new workflow and not an old rule is that the quality bar moved faster than most teams' processes did. Nobody built a "spot the fake" muscle for creative that doesn't have obvious fakes anymore. Media buyers who are good at their jobs got good at catching warped hands and garbled text — skills that are becoming less useful by the month, while the actual risk moved somewhere else entirely: not "does this look fake," but "does this say something true about our product, to the right person, in our voice." That's a judgment call, not a pattern-recognition task, and it's exactly the kind of thing that still needs a human in the loop, just not the human you had checking for it a year ago.
Meta is going to keep shipping. The tools will keep getting better and the outputs will keep getting harder to distinguish from what a human made on purpose. That's not a reason to panic or to opt out. It's a reason to build a review process that assumes the AI is good — good enough to fool you — and checks it anyway.