The Takeaway
- Amazon has used generative AI to refine search for past couple of years
- Merchants report 15-20% sales increases with AI-friendly listings.
- Amazon is testing a new metric to measure accuracy of product listings
When ecommerce consultant David Khandrius had to help a client who sold baby strollers write a product listing for Amazon’s website in the past, he typically tried to jam as many words into the listing as possible to ensure Amazon’s search algorithm would capture it. That might result in a convoluted title for a listing, like “travel stroller for beach park super compact airplane overhead.”
Nowadays, though, he said, the title might be “foldable compact travel stroller,” while the description would use conversational phrases like “great for moms on the go” or “take it to the beach, the park, put it in the airplane overhead.” The changes are an example of how product listings on Amazon are evolving in response to artificial intelligence tweaks to Amazon’s search algorithm.
In addition to introducing more conversational language to their product listings, merchants are adding more information about how products are used by consumers and changing the images. The changes can help the listings show up more prominently both in Amazon search and its new AI chatbot, Rufus.
“Everyone used to say, ‘Fit as many as many keywords as you can into your bullets without sounding like gibberish,’” said Khandrius, founder of e-commerce consultancy Peachy.
But that often means listings get packed with a string of similar words that don’t actually provide valuable information to shoppers.
Just as companies focused on how their businesses appear in Google search results are rethinking their approach to search engine optimization, adapting it to Google’s AI-enhanced search results, so are brands adapting to Amazon’s use of AI in its search engine—and its introduction of Rufus.
Amazon says it has been using generative AI to refine its search engine for the past couple of years. It’s been working to make its search engine better understand the shopper’s intentions and focus less on keywords.
Early results of the product listing rewrites are positive. Merchants who have changed their approach have seen increases in sales and traffic, people who advise sellers told The Information. Khandrius said his clients’ sales had risen 15% to 20% since they introduced the new style of listings in the second quarter.
Amazon says using generative AI in search has led to better results that give shoppers more context on the product than previous search methods did. Amazon also says its Rufus AI chatbot can provide more information and context about products than a traditional listing can.
“If you think about how many things in a static product page that you’d be able to discover, maybe there were 20, 30 different attributes that might show up on that page,” said Dharmesh Mehta, vice president of worldwide selling partner services at Amazon.
“I can have a conversation with Rufus, and Rufus knows not just about that product—it knows about frequently asked questions that the manufacturer has buried in their product manual somewhere,” he said. “It knows about 17 similar products and can tell you how to compare and contrast them, and that’s just such a different kind of intelligence someone can get shopping than what they could do before.”
Capturing Context
Amazon also offers AI-powered tools to sellers to help them create more detailed product listing pages, and these AI-generated listing pages on average are higher quality than the product listing pages that had been previously created manually, Mehta said.
The company is also testing changes to how it assesses the accuracy of its product listings, which is important for its business. In the latest issue of Harvard Business Review, an Amazon economist, an Amazon engineer and a Harvard Business School professor co-authored a paper that noted customers were more likely to buy a product when the written description seemed to match the item. Accurate descriptions help build trust in the platform, they said.
Amazon uses a metric called item data quality to help measure the accuracy of product listings. But the metric has limitations—it checks that a listing had at least one keyword and a brand name in it, for instance, but it can’t assess the product description more broadly.
For example, it wouldn’t be able to judge if the product listing “super fast blender for fruits” accurately described the product, said a person who talked to Amazon about the changes.
Amazon is testing a new metric, composite data quality, said the same person. CDQ scores listings on attributes like description quality and title accuracy, Amazon has told sellers. It can go beyond yes-or-no answers and form a more nuanced assessment of the listing’s accuracy, the person said.
Amazon is also trying to tailor search results to more specific user queries in other ways. The company has recently been testing a new feature known as ImageSmith on some listings for its U.S. site, according to messages from Amazon representatives to sellers reviewed by The Information. ImageSmith selects the most relevant image from a product’s listing to appear in search results based on the specific query a customer uses, according to the messages.
Better Sales
Max Sinclair, a former Amazon employee, started the marketing firm Azoma three years ago to help brands perform better in AI search, particularly on Amazon. His clients include David Protein and tech firm HP.
An example of the old keyword-focused product listings for David’s protein bars included one like this: “David, Protein Bar, Blueberry Pie, Pack of 12 Protein Bars | 28g of Protein | 150 Calories | 0g of Sugar - High Protein, Low Carb, Gluten Free Protein Bar Flavor: Jammy blueberry pie filling with baked pie crust. Sweet with white chocolate chunks and vanilla cookie crisps. Each bite is filled with flavor and different textures.”
The listing designed to show up better in AI searches is much longer and written much more conversationally. It starts: “Fuel your fitness journey with David Protein Bars, engineered for health-conscious professionals who demand results without compromise. Each bar delivers an impressive 28g of high-quality protein packed into just 150 calories, making it one of the most protein-dense snacks available.”
Some shoppers might prefer the older listing, which presents more details in a more straightforward way. But Sinclair says the newer listing helped the snack brand surge to the 33rd position in the sports nutrition bar category from the number 400 rank over several months.
Still, it’s possible the merchants seeing the improvements could have gotten similar results from other tactics. Some of them had little experience with sophisticated ways to improve their search rankings.
These better rankings could have a downside for Amazon, if merchants decide sales improvements from product listings make buying ads on Amazon’s marketplace page less important. Amazon’s $56 billion ad business mostly comes from merchants advertising their products on the e-commerce site.
That’s already happening with merchants working with Katya Constantine, CEO of ad agency DigishopGirl Media. Constantine’s clients enjoyed 20% to 30% growth in traffic after their listings began to incorporate conversational English, which is one reason she’s spending slightly less on Amazon ads this year.
But other sellers still find ads useful. Khandrius said they ensure his products get seen in the first place, while improving the listings to be more AI-friendly makes it more likely shoppers will actually buy his offerings once they find them.
Meanwhile, shoppers’ growing use of Rufus has forced merchants to rethink their approach to search. Amazon marketing agency Lunge Marketing recently helped its clients, which include brands like Disney and New Balance, make up for a drop in the number of shoppers typing queries into Amazon’s search bar, according to Jonathan Wilner, Lunge’s vice president of brand and advertising.
Rufus doesn’t give advertisers as much data as typical keyword searches, Wilner said. But its queries help his clients better understand the types of shoppers that are browsing for certain products, which can help the clients fine-tune their pages and product descriptions. (Amazon says Rufus is just one of many spots on its site where shoppers can learn about their products, including Amazon Lens and the Amazon home page.)
