The launch of Shopify’s Custom Data platform aimed to enable building and integrating content and custom data attributes with native Shopify tools, which previously required third-party partner apps or costly custom technology stacks.
However, Custom Data was initially born from explorations solely intent on building a Content Management System (CMS). Before the team had gone too far down that road, our user data showed that more advanced data and content modeling tools like Contentful or Sanity.io were swiftly becoming the go-to choice for large and enterprise merchants integrating with Shopify—effectively making Shopify a spoke in their e-commerce tech stack rather than a hub where merchant data and content could be centrally managed and leveraged throughout the rest of the platform with ease.
Unfortunately for merchants, these powerful tools come with a steep learning curve and additional expenses and complexities. So, rather than build a small, basic set of CMS features, we pivoted to build a foundation for managing bespoke data structures, which we knew would inevitably include customer-facing content.
With the launch of Custom Data, the next priority was streamlining the content creation process for merchants. I formed a team of UXers around this subset of user flows, and we got stuck into the Content Management roadmap. Shortly after, an opportune collaboration with OpenAI and the GPT-3 large language model (LLM) paved the way for Shopify Magic—a growing suite of features enabling merchants building storefront content to quickly move beyond the blank page.
Depending on their brand needs, customer segments, and the sheer amount of inventory they are listing, it can take merchants weeks or longer to prepare the content for their storefronts. The longer a merchant takes to launch their store, the larger their opportunity cost and the higher the likelihood that their endeavors will come to a screeching halt. Building and managing this content can quickly escalate into a full-time position for many small merchants. Merchants with staff for building content don’t want to waste time building or integrating the tools to manage it before they can start.
So, as a part of the Content Experiences group, the Custom Data product area was responsible for understanding, designing, and enabling many of the primary use cases of Shopify Magic. We commenced with a multi-pronged user research effort to identify and understand these use cases.
Our journey to define the core tasks where Shopify Magic could have the most impact began with an intensive user research phase. We conducted quantitative research, observational sessions, and segmented usability surveys to understand merchant behavior and their underlying needs better.
Quantitative Research: Working with our data partners, we identified a trend in a subset of merchants where they signaled relatively strong intent to build content within our CMS, but we hit a wall in creating content entries. Often, smaller merchants would opt-in to the onboarding for Custom Data, and some would even build their first content models. Many would create an initial entry only to delete it within a few days or weeks. Others would never create an entry to begin with. These trends gave us a great place to dig into.
Observational User Sessions: We observed and understood how merchants perform daily tasks. The process involved breaking down workflows into discrete tasks based on how users conceptualize their work. This exercise gave us an intimate view of the merchants’ writing process, modeling content, and categorizing products. We also spoke to a swath of merchants who had built models without entries to understand if barriers kept them from progressing with their content. We saw that while many merchants intended to round out their stores with content—some even with strategic ideas about the kind of content they wanted to build—they often didn’t understand tactically what a good content model looked like.
Segmented Usability Surveys: We wanted to understand the usability barriers merchants face in their everyday operations. We segmented our merchants based on their size, experience, and other factors to get a nuanced understanding of their specific challenges. Through these surveys, we identified the most time-consuming and complex tasks for each merchant segment that became the prime candidates for Shopify Magic.
As a result of our research, we identified the following user tasks as key areas where Shopify Magic could make a significant difference:
Shopify Magic offers an optional, streamlined set of features integrated within existing merchant workflows. The UX must be highly contextual, flexible, and lightweight.
Writing product descriptions for the merchant simplifies the product listing process while maintaining a consistent, high-quality content standard, fostering increased customer engagement and sales growth. Our Product team was keen to implement this feature, but we knew getting the details of the experience right would be crucial to our success.
It was clear that while many merchants would be excited to use this new feature, it would not be a good idea to treat it as an integral step in writing a product description. For one thing, merchants frequently revisit their product pages to make minor adjustments without looking to re-engineer the content entirely. For another, we know that a subset of merchants will prefer to ignore this feature, and it’s essential that it doesn’t get in their way. For this reason, the field flag UI component with the Shopify Magic icon gives enough visibility to this feature as a secondary action.
In our first explorations, we allowed Autowrite to work with only a product title and without any keywords in the description field. GPT was still able to generate content, but the specificity of that content was low and frequently inaccurate. For one thing, product names often reflect brand or market positioning more than they describe an item. Merchants usually have unique products that aren’t well represented by an “average” of all adjacent products in the market.
So it made sense to partially or fully decouple content generation from the Product Title field. With further testing, we decided that a minimum of 2 keywords within the description field would be enough prompt to generate a workable product description—with more prompts potentially leading to an even better outcome.
We also needed exploration done on where to place the generated content. When placed directly into the product description field, all manner of questions arose in my mind about how to handle any other content that might already exist within that field. After all, merchants might only be testing the feature or not wish to keep all or some of the generated content. I believed it was important that this feature be non-destructive to any writing already in the field. So, I pushed the team to find a lightweight way to showcase that content outside the description field until a merchant elected to keep it—allowing them to continue evaluating their keyword input and comparing any content they already had in the field to the generated description text.
In the end, the team’s explorations led us to contextually trigger expansions of the product description card so that the content and generative controls (including the ability to select a tone, rewrite, keep, and give simple feedback to our model) could live just below the description field itself.
Generating image alt text improves website accessibility and search engine performance, enhancing user experience and increasing organic traffic growth. Enabling this feature was such an obvious “quick win” that we could give merchants that we knew it made sense to offer it even if merchants weren’t yet asking us for it.
While this feature seems like a candidate for the same treatment as product descriptions, our research gave me enough signal that merchants don’t approach this task the same way. First, many merchants are not aware of the impact of image alt text on their SEO, accessibility, and discoverability strategies—meaning that many of them skip this task altogether. Secondly, image alt text is best when brief and descriptive rather than lengthy and evocative.
I determined that the best solution was to generate this content unprompted on image assets without any existing alt text and gently prompt users to either keep it or rewrite it. I challenged my team to use every subtlety of our text field component to accomplish this unobtrusively. The result is that many users who would typically skip this step without thinking can now seamlessly add it into their workflow with as little effort as a single click, and merchants intent on providing their alt text encounter virtually no static in the process.
Building custom-fit content models expedites content creation and ensures the consistent presence of crucial information, boosting merchant efficiency through standardized attributes that drive buyer conversion.
The most significant problems merchants have in building their content models is determining what constituent components are needed to effectively tell a story consistently across all the surfaces where merchants implement them and which of the available data types best fit each component. The best models are highly reusable and reflect the standards and nuances of the market each merchant is selling in. Building and then maintaining those standards for every possible market or industry requires a herculean effort, which is one reason no one in the CMS space has done it yet. With Shopify Magic, I saw an opportunity to skip ahead several years and construct these standards for each merchant—essentially in real time—and I tasked one of my product designers and one of my creative technologists to build a prototype of this quickly. The results were so gratifying that we had to get this out to merchants.
By including the available data types and their corresponding field validations in the prompt to GPT and the kind of content the merchant is trying to craft drawn from the model name field, we can build the model for them to a high standard in most cases. Even so, users must be able to pick and choose only those components they know will work for their storefront and discard the rest of the suggestions quickly.
Shopify Magic intelligently suggests product attributes based on the merchant’s storefront profile, inventory, and product description to classify items within a well-defined taxonomy automatically. This process enables customers to search and filter products easily, enhancing the overall shopping experience.
Building a universal, flexible product taxonomy that fits high-performing industry models and merchant mental models is an infinite game. Getting merchants to adopt this endlessly iterating model requires making the experience so simple and intuitive that a manualapproach can’t come close.
Working closely across product areas, my team and I led a cross-disciplinary consortium to determine what existing merchant input we could leverage to do as much of this work for the user as possible. It took us a couple of months to design and develop an MVP that we knew would serve merchant needs while remaining translatable to the taxonomies of external channels like Google and Facebook.
We can draw from our categories and attributes by analyzing the product name, description, and images while considering the storefront’s category and suggesting the most relevant categorization scheme with pre-filled attribute values. All the merchant needed to do was confirm them. Like everything else that’s part of Shopify Magic, merchants can still edit the values freely. After all, not every company uses the same word for “blue” or “large.”
The team nailed it, and I can’t wait to see it roll out globally soon. However, it was only possible by understanding and getting broad alignment on the merchant tasks and challenges. Auto-categorization and attribution were features that could never have coalesced without UX leading the way for the whole organization. For that, I’m very proud of the work my team and I did together.
Beta testing for the other features, such as auto-generating image alt text, content models, and product categorization, has also shown strong demand. Feedback from beta users and early adopters indicates that these features address critical pain points for merchants, simplify their workflow, and reduce the time and effort required to create and manage high-quality content. The team plans to launch these features globally within the next few months.
The rollout of Shopify Magic has generated significant interest and enthusiasm within the merchant community. The adoption rate has been remarkable in just a few short months since its launch. Currently, tens of thousands of product descriptions generated by the tool are being used on public storefronts, showcasing its widespread appeal and effectiveness. Many merchants are still testing the feature, highlighting its potential for continued growth.
Shopify Magic has demonstrated its capacity to significantly decrease the time from merchant sign-up to the first sale. Merchants have reported increased efficiency, improved content quality, and enhanced customer user experiences, leading to higher conversion rates and sales growth.
Moreover, the successful implementation of Shopify Magic has reinforced Shopify’s commitment to providing innovative and user-friendly tools for merchants and showcased the potential for further collaboration with AI and ML technologies. Moving forward, the team will continue to iterate on the existing features, gather valuable user feedback, and explore new possibilities to empower merchants with even more efficient and effective tools for their e-commerce businesses.
In summary, Shopify Magic’s strong initial adoption and positive results underscore its value in helping merchants launch their online storefronts quickly and effectively. As the platform continues to evolve and expand, it is poised to become an indispensable tool for e-commerce merchants looking to build high-conversion, quality storefronts with minimal time and effort.