Careful — Meta AI Can Change Your Ad Creative, But Without Your Permission

Imagine a typical Tuesday at a digital agency. Your team has just finished a month of preparation for the launch of a large summer campaign for a premium brand. The team checked every word, approved the way the discount is presented, and polished the static banners to perfection in the brand's signature pastel tones. You upload everything into Ads Manager, pass moderation successfully, and the campaign goes live.

Some time passes. Suddenly you get a call from a furious client who has seen their ad on Instagram. Instead of the approved creative, they see a chaotic 3D animation. The carefully agreed headline has vanished without a trace, and in its place sits a template phrase generated by a neural network.

The client demands an explanation, the budget has been "burned" on irrelevant clicks, and you stare in horror at the settings and realize: you didn't do this. Meta's artificial intelligence did. And the scariest part — it did this automatically, without any direct permission from you.

This scenario is not an exaggeration and not an isolated system bug. Today this is the new reality of working with the Meta ad account, where the Advantage+ Creative algorithms have gained unprecedented power over the visual and text communication of brands. The platform decided that the machine knows better than the specialist what the creative should look like, and rolled out an aggressive "improvements" policy that is now causing a real rebellion among marketing agencies around the world.

Andromeda and the illusion of automation

In recent years Meta has systematically moved toward the full automation of ad buying. The peak came at the end of 2024 — the beginning of 2026 with the Andromeda algorithm.

First, the artificial intelligence took control of targeting (Advantage+ Audience), then of budget distribution (CBO), and finally it reached the creative.

Advantage+ Creative Enhancements is when Meta's AI automatically changes your images, videos, and texts right at the moment the ad is shown. Officially: the system adapts the creative to each user in order to lower the cost per action (CPA) and raise the click-through rate (CTR). Meta claims an average saving of 4%.

In practice, things are different. The head of paid media at Edge Marketing, Susan Harangi, says it plainly: this shift serves Meta's internal goals, not the interests of advertisers. Instead of saving time — three problems:

• All AI features are enabled by default. To turn them off, you have to consciously search for hidden toggles.

• Ads that have already been approved and launched can update themselves right in the middle of running.

• The AI generates a hundred variations of a banner in a second, but if none of them can be trusted, all the saved time goes into quality checking. As David Halter of Dentsu notes, bad automation doesn't reduce work — it just moves it to another stage.

How AI destroys brand communication

Advantage+ Creative is a "black box" where the brand's original message can change beyond recognition. Meta offers more than 20 optimizations, and their number keeps growing. Here are the most aggressive ones.

• Visual enhancements. The algorithm itself changes brightness, contrast, and color, applying filters. The result — "acid" HDR banners: white becomes cream, shadows unnaturally deep, the brand style disappears.

• Image animation. Static photos turn into videos with artificial panning or zoom. Carefully crafted hero shots start to look like cheap motion graphics from the 2000s.

• Text variations. The AI rewrites headlines and texts or invents its own. Wordings that passed A/B tests are replaced with trite clichés, and legal disclaimers can simply disappear.

• Music overlay. The algorithm adds background music to the video on its own. The problem is that it doesn't understand the context: a serious B2B campaign can suddenly get an inappropriate entertainment or New Year track.

• Image expansion. Generative AI "draws in" the background to adapt the banner for Stories (9:16). This is where hallucinations appear: deformed logos, blurred edges of packaging.

• Optimize text per person. The system swaps the headline, primary text, and description depending on the user's reaction. Because of this, a strong call to action can hide inside the text, while a technical description can jump to the front.

• Additional links. The algorithm adds links to other sections of the site under the main CTA. People lose focus, click on secondary pages instead of the target action — and conversion drops.

Legal and reputational consequences of AI hallucinations

For e-commerce, a changed dress color in a photo is at most a product return. But for regulated industries, the algorithm's text improvisations lead straight to court. Medical clinics, financial and insurance companies, and law firms work within strict compliance frameworks: any promise in an ad for a loan or a medical procedure must be accompanied by a disclaimer. When "Text variations" or "Optimize text per person" hides this text or paraphrases it — the company violates advertising law.

As David Halter notes, an extra limb in a model can raise a smile on marketers' forums. But when a client sees an inaccurate price, an invented product feature, or an unlawful claim about healing properties, it is not Meta's AI that pays for it, but the brand itself and the agency running the account.

The death of classic A/B testing and chaos in metrics

For performance marketers, Advantage+ Creative is an analytical nightmare. The foundation of quality media buying is clean A/B testing: we take two different "hooks" (for example, one pushes on benefit, the other on the fear of missing out) and launch them under identical conditions to see what resonates better with the audience.

But if AI enhancements are enabled, this testing breaks. Imagine: Creative A brought in more leads than Creative B. You think your text worked. But in reality the system might have quietly added a bright filter to Creative A, enlarged the photo, and overlaid a cheerful melody — and you don't even know about it. In other words, you no longer understand what exactly produced the result.

Drawing correct conclusions in such conditions is almost impossible — especially when campaigns run on automatic strategies (Cost Cap, CBO), where the system itself decides where to pour the traffic. That is exactly why experienced media buyers turn these features off first of all, to preserve the purity of measurements.

Creative fatigue

AI masks real ad "burnout." Every creative has its own life cycle: when a narrow audience sees the same banner too often (Frequency above 2.5–3.0 for cold traffic), banner blindness sets in — CTR drops, CPA rises.

In a normal process this is a signal: you need a new creative angle, fresh text, a change of format from photo to Reels. But with Advantage+ enabled, the algorithm tries to "rescue" the dead creative with cosmetics — it draws in a background, shuffles words, applies zoom. This creates the illusion of work and wastes budget (each variation costs money to test), but it doesn't solve the main thing: a weak offer or a boring visual.

The algorithm is not capable of coming up with a new idea — it only endlessly chews over what it was given.

What is really happening in the online space

In the spaces of the internet, on specialized threads and media buyers' forums, discontent is reaching a boiling point. Specialists complain not only about the features themselves, but also about how cunningly Meta tries to keep them enabled.

The standard scaling practice is cloning an ad set. But specialists are massively recording a bug (or a deliberate feature): even if all enhancements were manually turned off in the original, when duplicating, the system "forgets" this and turns Advantage+ Creative back on. Some options, such as "Enhance CTA," can activate on their own even in the middle of an active run.

A separate story is "Optimize text per person." It cannot be turned off with an ordinary toggle: Meta hid the function deep in the "Advanced preview" menu, where you have to manually uncheck "All optimisations" and also check hidden lists. Such an interface labyrinth strongly slows down the work — manually turning off the AI for 20–30 creatives is exhausting and leads to human errors.

Survival guide. How to remove AI from your campaigns

Since Meta is not going to stop rolling out automation, every SMM specialist and targetologist needs to implement a strict procedure (checklist) for launching ads in order to protect the brand. This process covers several levels.

1. Global level (Business Manager)

Before creating an ad, go to:

Business Settings → Advertising Settings (or Ad Account Settings)

Here you'll find a non-obvious option about testing new AI features (AI creative testing). Turn it off. This will stop the platform from using your campaigns and the client's money as a testing ground for raw generative tools.

2. Catalog level (For online stores)

If you launch dynamic retargeting or an Advantage+ Shopping Campaign based on a product feed, go to:

Catalog → Catalog management → Settings → Auto-enhancements

This is exactly where you need to turn off the generation of AI backgrounds (Generate Background) and other automatic image processing, so that your product on the banner matches what a person will see on the online store's site.

3. Ad creation level (Ad Level)

This is the most important and most tedious stage.

At the level of the specific creative (Ad), find the Advantage+ Creative section — in some interfaces, it may be called Optimize Media.

Click the "Edit" button.

Manually set each toggle to the OFF position:

• Turn off brightness changes.

• Turn off adding music.

• Turn off 3D animation.

• Turn off image expansion.

In the text creation block Text & headlines, be sure to deactivate the Generate variations (or Text variations) toggle.

The copywriter must write 3–5 headline options themselves for honest testing, rather than relying on generation.

To turn off text shuffling, use the Advanced Preview lifehack described above:

Advanced preview → Advantage+ Creative → Turn off All optimisations

A tip for pros

When scaling campaigns through duplication, always recheck all these settings anew. Never trust the system to save your refusals of Advantage+.

To automate this process, some advanced agencies use external bulk-launch tools (for example, TheOptimizer), which allow you to apply a template with AI settings firmly turned off to hundreds of ads in one click.

When AI is actually useful

Rejecting innovations entirely would be unprofessional. Some algorithms really do improve results without harming reputation — the main thing is to keep them within strict limits.

Susan Harangi gives an example: her agency allowed the AI to generate a lifestyle background around a photo of a product with no background, but strictly forbade changing the product itself and the text. Within a month this creative received twice as much budget from the algorithm and brought in 5 purchases, while the original static banner — none.

The main rule for working with AI in Meta in 2026: use it for distribution and adaptation, but never trust it to form meaning.

Why the machine won't replace manual work

AI can produce a hundred versions of an idea in seconds — but a human still has to come up with the idea itself. Meta simplifies launching an ad down to a single "Advantage+" button, but for a premium, medical, or financial product this is unacceptable: there, the tone, the aesthetics, and the precision of wording decide everything.

AI should take away the routine — adapting banners, analyzing data, optimizing bids — not forming meaning. And practice confirms this: well-thought-out manual campaigns with clean testing still deliver a more stable result than blind trust in automation.