REI ditches Meta tool after AI adds extra handlebars to bike ads

This incident exposes the friction between platform-driven automation and high-stakes brand safety. While Meta’s Advantage+ tools promise efficiency through automated variants, the removal of human-in-the-loop verification creates immediate reputational risk, especially for premium retailers where product accuracy is non-negotiable. For the broader ecosystem, this highlights a growing trend of 'stealth' AI enrollment that bypasses traditional creative approval workflows. Strategists should monitor whether platforms implement more granular controls or mandatory approval gates for generative alterations to prevent further advertiser churn toward manual creative controls.

The REI incident is part of a broader pattern of advertiser frustration with Meta’s Advantage+ Creative suite. In April 2026, Marketing Brew reported that apparel brand Snag Tights sought to reduce its ad spend after Meta's AI tools modified product imagery without authorization, leading the CEO to characterize the inaccurate results as 'scamming the customer.' Industry reports from HyperFX in May 2026 suggest these errors are partially linked to the rollout of Meta’s 'Andromeda' ad retrieval system, which has been associated with aggressive default settings that re-enable AI enhancements even after users attempt to toggle them off.

Simultaneously, Meta is attempting to bridge the trust gap by introducing more restrictive brand controls. Per Social Media Today in September 2025, Meta launched a 'Restricted Words' feature allowing marketers to exclude specific terms from AI-generated copy. Despite these additions, many operators report that 'optimization' features, such as image expansion or background generation, continue to produce physically impossible visuals, such as contorted limbs or gravity-defying objects. Per Digiday, July 2025, while 53% of U.S. marketers identify social media as a top threat to brand reputation, platforms continue to prioritize speed and scale, forcing ad-ops teams to decide between the performance lift of AI automation and the oversight burden required to verify every variant.