Improve Amazon’s AI-assisted content generation experiences inside A+ Content Manager (ACM).
Context
A+ Content Manager (ACM) is Amazon’s B2B content creation platform, used by brands and sellers to build enhanced product-detail experiences across Amazon’s marketplace.
As gen-AI capabilities emerged, we began exploring how AI could help sellers create content more efficiently without sacrificing quality or control. My work needed to sit at the intersection of UX writing, systems thinking, and AI product education.
Goals
Creating great A+ content had historically required a lot of work: gathering product information, writing copy, sourcing images, and organizing everything into structured modules that could be published at scale.
For this project to succeed, we needed to design updated workflows that made AI generation understandable, usable, and trustworthy for sellers working inside our complex publishing system, where getting up complete information can be a huge barrier for sellers creating A+ product content.
Our goals included:
- Reducing the effort required to create high-quality A+ content
- Seamlessly integrating AI into the current A+ creation flow
- Helping sellers understand how generated outputs and insights related to their product and category data
- Creating guidance that reduced confusion around emerging AI capabilities
- Incentivizing prompt updates and iterations that would help sellers get the final results they wanted
- Maintaining transparency and seller control throughout the generation process
Integrating AI into existing systems
Given the newness of the technology and Amazon’s business focus around gen-AI adoption, we needed to help sellers understand how AI-generated outputs related to their own product data, prompts, and existing content. At the same time, the technology was still being developed on the backend so we were both designing around what didn’t yet exist as well as extreme tech limitations.
Our main challenge was updating AI generation flows without disrupting existing publishing workflows. I helped shape experiences that integrated AI directly into existing modules and publishing patterns rather than treating it as a disconnected feature.
This included module-level generation, inline editing, contextual help content, generation status messaging, editable AI outputs, and prompt guidance.
Explorations of content-generation workflows, selection states, and regeneration behavior
Our first mocks showed how AI-assisted content generation could be integrated directly into the A+ publishing workflow.
Another consideration was how to populate the A+ content module with newly generated content and images—in the AI-gen side panel or directly inside the A+ module, where revision-related CTAs would open the side panel back up.
Our designs also had to seamlessly incorporate insight generation based on proprietary data and solve for potential edge cases. A big challenge was balancing generated insights with prompting options, and within that, balancing text- and image-related prompting happening within the same panel (but neither being required to generate product-related content).
As we were iterating, another Amazon team launched a seller “creative studio,” a dedicated flow for generating higher quality images than our ACM was capable of generating. Within our designs, we had to account for this by displaying an option to visit the creative studio. This was a stop-gap until the creative studio could be fully incorporated into the ACM content generation experience.
Designing AI guidance for sellers
The new workflow needed to combine current product data, new customer insights, prompt input possibilities, and yet-unknown capabilities for generated text and imagery. I worked on instructional UX and guidance systems that needed to explain:
- What “AI Ready” meant
- How generation worked
- How prompts influenced outputs
- What sellers were responsible for reviewing
- Where generated assets were saved
- How to regenerate or refine outputs
The challenge was not only about understanding and incorporating this new technology into the ACM, it was also about helping users confidently navigate the generation process itself.
My work throughout focused on reducing ambiguity and helping sellers understand how AI suggestions connected to their inputs through additional tooltips, updated inline messaging, and clarifying help copy.
Key Takeaways
AI products require explanation as much as generation. One of the biggest challenges wasn’t generating content—it was helping people understand what had been generated, why it appeared, and what responsibility remained with them. Instructional UX, status messaging, and contextual guidance became just as important as the generation experience itself.
Trust is a design problem. Sellers needed confidence that AI-generated outputs reflected their products accurately. Clear feedback loops, transparent workflows, and opportunities for review were essential to adoption.
Emerging technology rarely arrives all at once. Throughout the project, underlying capabilities, business priorities, and adjacent AI initiatives continued to evolve. Designing for AI meant designing for change by building experiences that could accommodate shifting technology while remaining understandable to users.
Human judgment remains critical. While AI could accelerate content creation, sellers still needed to evaluate accuracy, brand appropriateness, and business relevance. Much of the work focused on helping users understand where automation ended and human decision-making began.
I’ve spent much of my career helping people navigate new systems while they’re still being defined. As AI-assisted creation becomes more common, content strategy plays an increasingly important role in defining expectations, establishing trust, and helping users exercise good judgment.