This August, Newegg Commerce Inc. unleashed generative artificial intelligence onto its ecommerce site.  

The web-only consumer electronics brand built a generative AI tool that summarizes a product’s reviews into one succinct comment. Newegg displays this review, called SummaryAI, above the individual reviews.  

The generative AI tool also displays “Pros” and “Cons” for each product at the top of the reviews section, again taken from the published site reviews. These “Pros” are also displayed directly underneath the product image at the top of page and labeled as “Review Bytes.”   

“We feel that this would be helpful for many customers that are looking for a quicker analysis,” says Andrew Choi, director of brand and website experience for Newegg.  

Reviews are important for Newegg. 20% of Newegg.com shoppers read reviews and they spend 40% more than shoppers who don’t read reviews, so any way to enhance this section of the website is important to Newegg.  

“We ultimately want the customer to get the right product for their needs,” Choi says.  

Newegg is one of the early-adopting retailers of generative AI, the rapidly developing form of artificial intelligence that can create text, images and other kinds of content that often seems like it was produced by a human. These tools draw on large amounts of data they are exposed to, such as, in the case of online retailers, all the attributes and images of a product catalog.    

Major brands including marketplace eBay Inc. and chocolate brand Ghirardelli Chocolate Co. also are using generative AI to enhance elements of their product detail pages, such as optimizing the pages for conversion and creating pages faster. Many other retailers and brands are experimenting with how this type of AI can improve product detail pages, though most are still testing it and have not yet debuted their results to the public.  

AI is a hot topic in retail. 26% of retailers said they planned to increase their investment in artificial intelligence in 2023 compared to 2022, according to a Digital Commerce 360 survey of 135 retailer executives in October 2022. Plus, 11% said investments in AI will be critical to improving online conversion in 2023, and an additional 44% said it would be important, according to the survey.   

What’s more, that survey was conducted before the OpenAI consortium released generative AI-based chatbot ChatGPT in November 2022, sparking massive interest in the new form of AI. The bot uses generative artificial intelligence and natural language processing to answer questions and create content.   

Generative AI helps retailers with large product catalogs  

Retailers can use generative AI to create new content on a product detail page, such as a summary of the product reviews, or the actual product details. Each element on the detail page is “selling” the product with information with the goal of getting shoppers to add the product to their carts and ultimately buy it. But if these pages are not optimized, they are not doing the best job possible at converting shoppers.   

“Generative AI can help identify the opportunity among the product catalog and product pages and potentially suggest copy and ways to optimize pages,” says Kassi Socha, director analyst, retail, at research firm Gartner.   

Many product detail pages are not fully optimized for various reasons including:  

• Large consumer brand manufacturers have thousands of SKUs.  
• Brands sell their goods online through dozens of retailers and marketplaces.  
• Global brands sell across dozens of sites in many countries.  
• Marketplace operators host thousands of sellers, each responsible for their own product pages.   

Generative AI can help, particularly by automating the creation of product page content.   

“So many retailers have extensive product catalogs that they simply cannot manually update them on a regular or semiregular basis,” Socha says.   

“A more complete product page with more extensive product information allows the consumer to make a more informed decision and quicker decisions,” Socha adds.  

EBay Inc. saw the potential generative AI could have on its marketplace of more than 1 billion SKUs to help address two key challenges: helping shoppers find items faster and more easily, and helping sellers list items faster with fewer obstacles, says Xiaodi Zhang, vice president of seller experience at eBay.   

Xiaodi Zhang, vice president of seller experience at eBay.

“On the buyer side and the seller side, and this is a new tool to help solve these problems,” Zhang says. “It may not work for every type of customer, but there are a lot of different types of customer problems that generative AI can help solve in a way that other types of technology haven’t been able to address before.”  

Early in 2023, the marketplace built a tool based on Open AI’s ChatGPT that creates a product description based on data sellers provide about a product’s category, condition, color, brand and more. It took about a month to develop the feature, Zhang says.   

“What’s amazing with generative AI is how quickly it works. We had a proof of concept in a matter of weeks,” Zhang says.   

EBay piloted the tool internally before slowly rolling it out to a randomized group, representing 5% of sellers in May 2023. In August, eBay offered it to all sellers listing an item via its app. 

EBay has been “pleasantly surprised” by the overwhelmingly positive seller feedback, Zhang says. As of late July, the customer satisfaction score was 80%, which is among eBay’s highest CSAT scores for recent, new feature launches, according to the marketplace. 

The tool is meant to address “the cold start problem” that consumers face when trying to sell items, Zhang says.   

“When you find something to sell at home, and you are going through the listing, and you are asked to describe the item, a lot of people face writer’s block of, ‘How do I describe this dress I bought last year that I never wore?’” Zhang says.  

Zhang says roughly 30% of sellers shown the generative AI tool use it, and of those, 95% accept at least part of the description. Sellers can tweak the description, edit and delete as they like, Zhang says. The AI bot tends to add adjectives and be “a little flowery in the way it describes things,” Zhang says, and it’s up to the seller to edit the wording and ensure it’s accurate.   

Because the marketplace seller edits and approves the listing, eBay does not label which reviews have been created by artificial intelligence.   

Newegg, on the other hand, labels any content produced by its generative AI tool. And, unlike eBay, Newegg does not have a human review the content before making it live on the website.   

Brands exercise caution before launching generative AI  

Gartner’s Socha cautions brands that at this early stage of generative artificial intelligence, it would be wise to have a human oversee content before displaying it to shoppers.   

“Generative AI can suggest copy and suggest opportunities, but there still needs to be a team in place to validate some of the outputs,” Socha says. “With any machine learning or artificial intelligence, it’s only as good as the inputs, and it take time to optimize and learn.”  

Socha says consumer distrust of generative AI is another reason retailers need to tread lightly with the technology.   

In a July 2023 Gartner survey of 303 consumers, 34% of consumers rated their comfort level with generative AI in retail as “very or somewhat troubled,” 53% said they think it will strongly or somewhat negatively impact society, and 66% said they are concerned about discrimination or bias in generative AI.  

“I applaud retailers that are adopting the technology with confidence and testing the use of generative AI, but the continued guidance at Gartner would be to approach with caution and be prepared to adapt quickly if a weakness is exposed,” Socha says.   

This lack of maturity is what makes both Ghirardelli Chocolate Co. and tool manufacturer The Stanley Black & Decker Co. hesitant to go live with generative AI tools now.   

“We’re excited. It’s such an interesting time with AI right now, but it needs to be tempered with a bit of caution and guardrails,” says Pam Perino, ecommerce content operations and development manager at Ghirardelli.  

That’s particularly true for a food manufacturer like Ghirardelli, given strict government regulations about food and beverages. For example, Ghirardelli can’t call its products “white chocolate,” because they do not contain cocoa and are not technically chocolate. Instead, products are labeled as “white baking chips” or “vanilla flavored.” Perino isn’t confident that generative AI would understand this distinction.   

Similarly, The Stanley Black & Decker company ran into issues when piloting generative AI with what it would call “claims,” says Dean McElwee, director, global ecommerce collaboration.  

For example, the tool it piloted might say, “These are the best ear pods and they will last longer.” That type of content would be considered a claim, and Stanley’s legal department would reject it, McElwee says.  

Stanley pilots generative AI  

Still, like many large brands, Stanley is looking to generative AI to speed up creation and optimization of product detail page content.  

Of Stanley Black & Decker’s 200,000 SKUs, only 5%-10% of them are optimized, McElwee says, although those items account for roughly 80% of its revenue, he says. There is potential to unlock more revenue if more product pages were optimized.  

With that in mind, Stanley tested technology from two generative AI vendors, Jasper.AI and Copy.AI, for three months. The brand compared the product descriptions and detail page content the two vendors produced for 10 different categories across a mix of its brands in English, Spanish, French and Portuguese.  

It would take a human about one hour per page to do this work, McElwee says, and the goal would be to reduce this to about 10 minutes for a person to validate the content and check with the brand’s legal department for any issues.  

Stanley measured the AI tools’ performance across such factors as accuracy, readability, tone of voice, style guidelines, legal parameters and usability. Neither tool scored perfectly, and each had a few errors across all the measured categories, with many errors in tone of voice and style guidelines, McElwee says.  

“These tools are great — but, caveat — we found they don’t quite meet our needs for a number of reasons,” McElwee says.  

For example, Stanley Black & Decker has several brands, each with its own audience, including its do-it-yourself-focused brand Black & Decker, and its professional power tools brand Dewalt. The Dewalt buyer wants information quickly about how the tool will work and a quick demonstration, while the Black & Decker shopper needs more information, education and inspiration. The AI tool created content that did not match the tone of voice of each audience, such as using exclamation points for Dewalt shoppers, McElwee says.  

“These tools don’t really understand nuance and tone of voice,” McElwee says. “Even if we were to use them, we need to go in and check tone of voice and in some cases rewriting a large portion, which negates the use case of speeding everything up.”   

Translation is also important to the global company, as language varies, even among English-speaking countries. For example, in the U.S., the term for the concrete where pedestrians walk is “sidewalk,” whereas in the U.K. it’s “pavement.”   

Using the correct term also greatly impacts SEO. For example, a U.S. shopper looking for a tool to trim grass would search for a weedwacker, while shoppers in the U.K. and Ireland would enter terms like “string trimmer” or “strimmer.”   

If the term “weedwacker” is not consistently used on the product detail page, the product will not show up high in search results in Google in the U.S. or on a retailer’s website.  

“When people search for things on the internet, it’s unbranded normally with a category-based descriptor,” McElwee says.  

Because of the issues encountered in its test, Stanley Black & Decker is still looking for a generative AI vendor that could help it optimize product detail pages, and McElwee estimates it will be 18 months before the company is actually using generative AI for this purpose.  

“It’s still immature,” he says.  

Stanley Black & Decker is among the 5% of retailers using AI who say it is not meeting their expectations, according to a Digital Commerce 360 retailer survey of 97 retail marketers in May 2023.  

Colgate uses generative AI within software   

Toothpaste CPG brand Colgate-Palmolive is also in the process of testing how generative AI might improve its product detail pages.   

Unlike Stanley’s test that focused on creating content, Colgate is piloting generative AI as a way to more efficiently extract information from analytics data in the Profitero Inc. platform it uses.   

Colgate has close to 1,000 online product detail pages when factoring in dozens of SKUs across the 10-plus retailers where it sells its products, including Amazon.com, Walmart.com, Thrivemarket.com, Instacart.com, Albertsons’ brands and more, says Todd Hassenfelt, global digital commerce director, strategy and execution at Colgate-Palmolive. It takes a team of people to keep those product pages up to date and resonating with the target audience on each site.   

Profitero’s tool allows brands to ask the generative AI bot questions about specific details on any of those many product detail pages. For example, if Colgate wanted to improve the conversion rate of its mouthwash on Walmart.com, an employee could ask the bot, “What is the optimal title length in this category on Walmart.com?” The bot could answer with the ideal number of characters for that category on that website.  

Another example of how Colgate plans to use the tool is to produce automated summaries of the sentiment of ratings and reviews, replacing a manual analysis. Colgate could ask “what is the sentiment of reviews for this product?” If many of the highest-ranking reviews highlight a certain attribute, such as good flavor, Colgate could update the product detail page to highlight flavor.  

Effective product detail pages impact in-store as well as online sales, Hassenfelt says. Shoppers frequently check their phones while shopping in stores and might look up additional details on product detail pages. Plus, retailers may pull information from a product detail page for in-store signage or product promotions. Ensuring its online information is current with the information a retailer’s audience cares most about could help boost sales in stores, Hassenfelt says.  

Hassenfelt expects this tool to save its teams weeks or months of manual analysis. At present, it takes a long time to create content and suitable variations for one SKU in multiple places, and it could take months to analyze conversion rates and make changes, he says. Generative AI potentially could do all that faster.  

In the short term, the goal is to, “make it simpler and faster for our teams to take the analysis from our products to then actionability, going to act on those insights, specifically for conversion rate and improving sales,” Hassenfelt says. Longer term, he says, it could make it easier to consolidate data and identify trends.  

While many retailers are focusing on how generative AI can produce text for product pages, it also could play a role in creating more effective images faster than humans can.   

Ghirardelli Chocolate Co. is testing generative artificial intelligence both for updating product copy with relevant search engine optimization words and for creating images, Perino says.  

For example, Ghirardelli wants its product detail pages regularly updated with season-appropriate SEO words, such as chocolate for Halloween, Christmas, Mother’s Day or graduation. Today, this copy is updated manually to ensure product pages show up high in search results.   

Like Stanley Black & Decker, Ghirardelli finds that generative AI is not yet refined enough to have mastered brand voice and tone, and it is waiting until it improves before deploying AI to write copy for its site.    

It also is waiting for more refined generative AI tools to help with image creation. Today, adding new images to a product detail page takes about a month from idea, photo shoot and editing to going live. With generative AI, it could be “10 to 100 times faster than a manual process for image creation,” Perino says.   

“That’s a great opportunity to do something quickly versus having to wait for a photoshoot or having to use stock images,” Perino says.  

For now, Ghirardelli is using generative AI to help edit images. On a recent photo shoot for its new no-sugar-added baking chips, the brand used generative AI to remove part of a napkin in the image and fill in the background with a part of the product’s bag.   

Ghirardelli often makes small tweaks like this to improve images. Since Q2 2023, the brand has used image-scoring software Vizit to evaluate how well its images engage shoppers. Vizit uses artificial intelligence to analyze how impactful an image is and to catalog the attributes of the image. Based on publicly available metrics, such as likes or shares on a social media website, Vizit can analyze how impactful an image is and list its attributes.  

The technology is helpful when deciding which images to use on product detail pages and the changes Ghirardelli might make so the images perform better, Perino says. For example, Vizit’s technology scored many of its closely cropped baking images higher than those that were zoomed out, and so those are the ones Ghirardelli will use on its page, Perino says.   

But when it comes to actually creating images to use on its site, Perino says generative AI is not ready yet. For example, if the brand were to say, “create a Ghirardelli logo,” Perino is not confident the AI would get the logo exactly right.  

But that immaturity could disappear fast. The more brands and consumers test, use and implement generative AI tools, the smarter they become. And while generative AI tools may not be creating entire product detail pages today, it may not be long before they do.