AI sells for you – product descriptions and pricing in just a few clicks

This material has been created in both written and video versions. You can watch the video on our Beyond AI channel or continue reading. Thanks!

Watch this material on YouTube:

AI Sells for You – Descriptions and Pricing in a Few Clicks

Hi, my name is Ziemek from the Beyond AI channel. Today, I want to show you how to automate the process of selling things that might be cluttering up your home—items you want to sell but lack the energy and time to create proper listings for. I will show you how to prepare these listings automatically, using only a photo and artificial intelligence models.

First, let me show you how this program works. We assume you have a photo that you upload to Google Drive. Once uploaded, the program starts, downloads the photo, and calls the OpenAI model. This model analyzes what is in the image and prepares a title and a description for the ad. Then, it searches Google for similar items. Everything is finally saved into a Google Sheet.

Let’s see how it worked for a game controller and a pair of used snowboard boots.

The response is structured so that we have the product name and the assumptions made. Not every detail needed for a sale is visible in a photo, so I asked ChatGPT to think about what the important details are and to suggest them. It noted the assumptions it made and, based on those, created a title, wrote the content, and then searched the internet to see how much the item might actually cost.

In the case of the used snowboard boots, it got a bit carried away, but it still showed that used boots like these cost around 300 PLN, which is a reasonable price. I assume the price for the controller was for new equipment, but it still serves as a great starting point for determining a fair price.

What do you need to build this automation? A Google Drive account, an account on the make.com platform, and access to the OpenAI API. Building programs on make.com is very simple; it involves dragging "bubbles"—the steps you want your program to perform. The first step is connecting to Google Drive and showing it which folder to monitor. The downloaded file—the image—is then sent directly to ChatGPT with the Vision module.

Take a look at the prompt I use to tell the model what to do. First, I ask it to figure out what is in the photo. Then, I ask it to consider which parameters are essential for that specific item to prepare a proper description and title.

If it doesn't know certain things—for example, the size wasn't visible on the boots—I ask it to make it up and remember it, saying: "Save this as an assumption." In the next step, I ask it to prepare the name. It has to assemble the things it assumed with the things it recognized in the photo. This is the string I want to use later in Google to find the items.

Once we have everything, we can ask it to prepare the ad copy. Here, I again use the Chain of Thought method: before it writes the content, it’s good if it considers the benefits for people interested in that item. It then writes a message based on those benefits, which is stored in a variable I call "content." We want it to first think about the benefits for potential buyers so that the description matches what a person might gain from these used boots or whatever else we are selling.

One very important thing at the end: we need the response to be formatted and structured as a JSON. I have separate fields for the name, title, description, assumptions, and the search query. This query is the text that goes to Google to find prices for similar items. So: I grabbed the photo, sent it to the Vision model, and the answer goes into a "bubble" that has the right structure. These variables can then be used later—for example, in a block sent to Google, where I paste the query as a search parameter.

The response from Google is HTML code, so it goes into a "text parser" block that converts it to plain text, stripping out HTML tags and unnecessary Javascript. This, in turn, goes to a second OpenAI step where we ask for a price estimate. What does that prompt look like?

It looks like this: I inform the model that I will provide a text fragment received from Google and ask it to look for prices within it and then suggest the most likely price.

The step where we send the request to Google will return a page containing many links to stores. Some will be auction platforms for used items, but most will be for new ones. Therefore, it is important to find the information in that flood of data that helps us figure out how much our item should actually cost. That’s why the prompt is designed specifically to find the most likely price for a used item.

Once all the data is collected, we save it all in Google Sheets by adding a row with the following structure: name in the first column, then assumptions, title, and description. All these things are available because we ensured earlier that the response was structured so we could fill the individual columns.

Preparing such a program is simple and automates quite a bit... regarding the price, it found that it costs about 25 PLN. Is that true? It would need to be verified. This type of automation will definitely work at home when you have a dozen or several dozen items; it saves time, especially for people who don't like writing descriptions themselves. Once you have the sheet, it’s easy to copy and paste to OLX, Allegro, Lokalnie, or Facebook Marketplace. It doesn't matter where you list it—you have the full set of info ready to go.

Would it work if there were thousands of items? No, that would require a different automation, likely involving a specialized agency dealing with large-scale model implementations. However, if you have ideas for other automations, let us know in the comments. We are happy to look into them. In fact, this episode was created thanks to a comment on one of our previous videos.

Visit our YouTube channel!

If you want to learn more about the fascinating world of artificial intelligence and its everyday applications, visit our YouTube channel – Beyond AI. It is your guide to the dynamic world of AI!