In e-commerce, there’s one particularly daunting task: writing product descriptions. It’s part of what’s known as PIM (Product Information Management), and yet it’s indispensable. Firstly, for customers, who need to know what they’re buying, and secondly, for SEO, because in e-commerce, product descriptions are the key to SEO. ChatGPT (and LLM or Large Language Models) promise to revolutionize Product Information Management. We demonstrate this in this article, where we generate product descriptions in combination with ChatGPT’s OCR plugin.
The benefits of generative AI for Product Information Management
- ChatGPT makes it easy to generate product descriptions that are completer and more attractive than those normally used on e-commerce websites.
- Product descriptions can be completed by asking ChatGPT to summarize customer reviews on the web (using a plugin such as WebPilot).
- Some retailers are already integrating back-end functionalities to simplify the creation of product sheets based on generative AI.
- The use of generative AI is particularly interesting for second-hand websites. Product descriptions, written by private individuals, are very brief. To find out more about these platforms, read our analysis here.
- Creating a product description can even be conducted from a photo. An automated description function based on uploaded photos should soon be proposed on second-hand websites.
In the remainder of this article, we look at 3 concrete cases where generative AI can simplify Product Information Management (PIM):
- First, we look at the case of food product descriptions in the retail sector.
- We then show how second-hand platforms can also benefit from generative AI to improve the descriptions left by salespeople.
- We conclude with an example of a product description based on a photo.
Product Information Management (PIM): how to improve product description with ChatGPT
We’re exaggerating when we say it’s all in the description, but that’s not far from the truth. Product sheets include legal information, but what sells them is the textual description of the product, its qualities, and its applications. Incidentally, it’s also the product description that enables us to be referenced in search engines.
The example below (a bottle of Italian wine) is symptomatic of what you’ll find on supermarket websites. The information provided is limited to legal notices, and the product description is sketchy.
It needs to be more than just describing your product correctly. It’s also important to do so in a varied way so as not to be penalized for “duplicate content”. And that, of course, is where things get tricky. When your online store is mono category, which is frequent, the exercise becomes even more perilous. Generative AI can be an invaluable ally in simplifying Product Information Management. Here’s an example of what can be achieved with ChatGPT and the WebPilot plugin (to access the product web page in question). To further highlight product features, we’ve even included the collection of customer reviews in the prompt.
How to integrate generative AI into second-hand websites
Generative AI also represents a huge opportunity for users of second-hand websites. Product information is also generally extremely poor, as in the example below, which I found on the classified ads website Leboncoin (Schibsted Group).
Individuals are even less aware than e-retailers of the importance of an excellent product description. This is seen in the ad above, where minimal effort is made despite the interest in the product. With 11 words, there’s little chance of this ad being properly referenced on Google. The seller can only rely on the Leboncoin.com search engine to find his ad… and sell his product more easily. Incidentally, a better product description also positively impacts customer satisfaction, as the product becomes more attractive and easier to “discover.”
However, doing much better than this would have been quite easy. ChatGPT would have produced a much richer result. For example, here’s what a simplistic prompt like “Can you create me a description of the Chanel bag mini boule dorée for sale on a second-hand website” would look like.
It’s still a little “lyrical,” but it’s already more complete than the descriptions that private individuals are used to writing. Since these descriptions are of interest from an SEO point of view, it’s reasonable to assume that second-hand platforms will soon integrate a writing assistance function. You must enter the item’s name, and the product data sheet will fill in automatically.
Creating a product description from an image in ChatGPT
ChatGPT can produce a product description directly from an image for even more automation. Activate a plugin such as SceneXplain. The result is still not as precise as that obtained from a web page, but it’s far more complete than what you’ll find in most product descriptions.
For this test, I randomly chose a product from a second-hand website. It’s a Hermes bag (Birkin model) accompanied by several photos.
The written prompt is minimal, as you can see below.
And yet, the SceneXplain plugin manages to identify the product correctly. This enables ChatGPT to write a complete product description, including information on price trends.
We’ve carried out several other tests with images, and the process remains random. The object must be well isolated and visible, and the photo must not contain other objects that could “parasitize” the product description.
Nevertheless, the potential is enormous. We can easily imagine e-commerce platforms integrating this functionality very shortly, either as a back-end to simplify the work of their teams or as a front-end to help users produce better descriptions.
Posted in Innovation.
