Meta’s aggressive rollout of artificial intelligence features across its advertising suite was sold as a shortcut to higher click‑through rates and lower creative production costs. In theory, the Advantage+ creative suite could auto‑generate variations, suggest copy tweaks, and even redesign images to better match audience preferences. In practice, many advertisers have found the experience far from seamless. Instead of polished, brand‑safe assets, they are encountering bizarre distortions: human figures with extra limbs, product photos that morph into unrelated items, and ad copy that devolves into meaningless strings of characters. These mishaps are not isolated curiosities; they are recurring enough that agency veterans describe dealing with them as a routine part of campaign management. The promise of frictionless automation has collided with the reality of immature models that still struggle with basic visual coherence and contextual understanding.
Specific examples illustrate how the technology can stray dramatically from the advertiser’s intent. One consultant working with a female‑founded pajama line reported that Meta’s AI recommended swapping a elegant pajama dress for a completely different shirt‑and‑pants set, effectively changing the product being sold. Another case involved a women‑only networking group in Montana where the system inserted male figures into promotional imagery, undermining the group’s core messaging. These are not mere aesthetic quirks; they represent substantive alterations that could confuse customers, dilute brand identity, and even raise legal concerns about misleading advertising. When the AI re‑imagines a product or audience without explicit permission, the resulting ads risk eroding the trust that brands have painstakingly built with their communities.
Meta’s public response to these incidents has leaned heavily on its terms of service, which place the burden of verification squarely on the advertiser. A company spokesperson reiterated that “AI can make mistakes” and that it is the advertiser’s responsibility to review AI‑generated outputs before they go live. While this disclaimer protects Meta legally, it shifts a significant operational cost onto clients who must now allocate time and resources to quality‑checking every AI‑enhanced asset. For small businesses without dedicated creative teams, this extra step can be prohibitive, while larger agencies report that the need to constantly police the system has become a new, unavoidable line item in their workflow.
The operational toll is especially acute for agencies that manage hundreds or thousands of concurrent campaigns. Rok Hladnik, CEO of Flat Circle, which oversees roughly $200 million in annual Meta spend, noted that his team has “accepted” the need to manually audit AI settings as a standard operating procedure. Each campaign requires a meticulous walk‑through of toggles to ensure that features like automatic image generation or copy variation are not inadvertently activated. When a bug flips these settings on, the resulting ads can go live with unwanted alterations, forcing teams to pull campaigns, issue corrections, and sometimes seek refunds—a process that eats into billable hours and delays time‑to‑market.
A vivid illustration of the frustration came from photographer and marketer Abigail Hogue, who launched a Valentine’s Day campaign for a boutique bookstore. After approving creatives that featured carefully styled chocolates, macarons, candles, and books, she awoke to find the live ads displaying garbled text and product images that resembled cheap knockoffs. Friends and acquaintances began teasing her about “AI slop,” prompting Hogue to panic, dive into Ads Manager, disable every AI enhancement, and republish the corrected version. The ensuing back‑and‑forth with Meta support yielded only vague assurances of a “sporadic glitch” and a promise to look into a refund—one that, as of the latest update, had not materialized. This episode underscores how a single AI misfire can trigger reputational ripple effects that far exceed the original media spend.
Beyond Hogue’s experience, other advertisers have shared similarly unsettling outcomes. Reports include an elderly model dressed in loungewear who appeared with exaggerated proportions, and a fashion shoot where a model’s leg was rendered bent at an impossible angle. Such distortions are more than comic fodder; they can convey unintended messages about a brand’s attention to detail or respect for its audience. Luke Jonas of Nest Commerce emphasized that keeping a human in the loop remains essential, quipping that an AI optimizing for six million advertisers will occasionally serve up something as absurd as a bike with two handlebars—a direct reference to the infamous REI Instagram ad that drew consumer ridicule.
The persistence of bugs that automatically reactivate AI features compounds the problem. Karissa Tuccio of Mediassociates said a toggle‑flipping bug had affected most of her fifteen Meta‑advertising clients, and she had flagged it to her Meta representative as recently as a week prior to the interview. While two other advertisers claimed the issue had been patched for their accounts, Tuccio noted that a similar problem remained active for her portfolio as of last week. This discrepancy suggests that any fixes may be unevenly applied, leaving many advertisers still vulnerable to unwanted AI interventions.
In response to growing concerns, Meta revealed that it has built an internal quality‑control dashboard designed to help large advertisers verify that AI enhancements are fully disabled before launch. According to Tuccio, a Meta representative told her that clients could submit a list of ad IDs for the system to scan and confirm that no unwanted AI tweaks remain. The very existence of such a dashboard implies that the platform acknowledges a systemic risk of inadvertent AI activation, yet it also places the onus on advertisers to proactively seek verification rather than fixing the root cause.
Meta’s efforts to increase transparency include the recent rollout of an “AI info label” that appears when an ad has been created or significantly edited using its proprietary tools or third‑party generators like Midjourney or DALL‑E. Users must click the three‑dot menu on an ad, choose “About this ad,” and then select “AI info” to see the disclosure. This mirrors a similar initiative by Google, which began labeling AI‑altered ads just weeks earlier. While the label offers a degree of visibility, it relies on users taking extra steps—a behavior that is unlikely to become widespread among casual scrollers—meaning its impact on mitigating consumer confusion may be limited.
On the model‑development front, Meta launched Muse Image, a new foundation model from its Superintelligence Labs aimed at producing higher‑quality creative assets. Initial rollout promised advertisers a powerful tool for generating variations based on a seed image. However, the feature quickly drew criticism when it allowed users to create AI images derived from other people’s public Instagram posts, a capability Meta described as having “missed the mark” and subsequently removed. The episode highlights the tightrope Meta walks between pushing cutting‑edge generative capabilities and respecting privacy, intellectual property, and community norms.
Meta is not the only platform experimenting with automated creative adjustments. Google’s Performance Max and AI Max products similarly scrape brand‑owned text, trim and video‑content to produce auto‑cropped shorts, rewritten copy, and resized images for placements such as YouTube Shorts. Some of these Google features are enabled by default, yet the company has largely avoided the high‑profile visual scandals that have plagued Meta. Industry observers attribute this difference to Google’s more conservative model tuning, stricter internal review processes, or perhaps simply to the fact that its AI tools have been in the wild longer, allowing more time for bug‑fixing and user feedback loops.
Despite the headaches, Meta’s advertising ecosystem remains indispensable for most brands. The platform generated roughly $196 billion in ad revenue last year and offers access to 3.5 billion daily active users alongside unparalleled targeting granularity. As Robert Webster of TAU Marketing pointed out, this scale gives Meta considerable leeway to make decisions that may frustrate advertisers but still drive profit, because few businesses can realistically abandon a channel that delivers such reach and conversion efficiency. Jessica Gleim, a consultant who works with female‑founded brands, echoed this sentiment, calling Meta “still the best platform” due to its depth of options and data richness—even as she navigates its AI‑induced quirks.
For advertisers navigating this turbulent landscape, a pragmatic, layered approach is essential. First, always treat AI‑generated suggestions as drafts rather than final assets; subject them to rigorous human review for brand safety, factual accuracy, and visual coherence. Second, leverage the new AI info label as an internal audit tool—export reports of ads flagged as AI‑altered and cross‑check them against your master creative library. Third, consider maintaining a separate, non‑AI‑enabled campaign template for high‑risk launches or brand‑sensitive verticals, toggling off all enhancements at the account level and enabling them only on a case‑by‑case basis after testing. Fourth, document any irregularities and share them with your Meta representative; aggregated feedback can accelerate platform‑wide fixes. Finally, diversify your media mix: allocate a test budget to rival platforms such as Google, TikTok, or emerging retail networks to reduce reliance on any single ecosystem and gain bargaining power when negotiating with Meta’s support teams.