to generate product descriptions using ChatGPT at scale

It’s hard to create enough quality content. How can we scale up the process of creating content, especially on ecommerce websites with lots of products?

You’d be quickly out of money if you paid a copywriter thousands of dollars to create product snippets.

What happens if you spend $1,000 on new product descriptions but only half are live a month later? You need to find a more efficient, cost-effective way of doing things. ChatGPT is a great tool for this.

The native web interface of ChatGPT is a time-saver and very helpful.

If we need to create hundreds or even thousands of product descriptions, then there is a way that’s more efficient than copying and pasting the prompts. Here’s how.

The output of mass production content snippets

You may want to create product snippets for your ecommerce site using data from the product information management system (PIM).

Imagine you have all the data in a spreadsheet.

Excel formulas can be used to combine (or join using the “&” operator), data into rich prompts ready for ChatGPT. You can use the following example:

You may need to use multiple “IF” statements in your formula. This is because some of your data might be missing.

Some products, for example, may not specify certain parameters (data in certain columns). Your formula must be flexible. You can ask ChatGPT for help in writing the formula.

You can then copy and paste the generated prompts in a word processor or notepad.

Even if some data points are missing, it’s a good idea to check a few of them to make sure the text is logical.

You can test the results of your formula by sending a few to ChatGPT manually (using the web interface).

You want the AI to be able to handle as much work as possible. We engaged in a “prompt crafting” process that was so deep.

Are you satisfied with the initial prompts and your responses? It’s now time to move forward.

OpenAI: Fetching new content snippets for your product

You now have a long list of products or other types of websites for which you would like to create content.

We’ll use a fictional sample of 100 items in this example. Now you have a complete list of your products, separated either by URL, SKU (or some other unique identifier).

You can also assign rich prompts to these products. ChatGPT is a limited web interface. How can you send them all at once?

You’ll need to be familiar with API and basic scripting. OpenAI API accounts can be created to access the ChatGPT interface.

I created a simple Python script for my company. Although I cannot share the script with you, I am happy to review the documentation and processes.

If I want to syndicate the script, it would be best if I built it using marketing-accessible technologies and endpoints. To do this, I created an Excel spreadsheet:

The sheet is simply a place to dump the items that need processing (identified in “Item Name”, in this case by product name). The prompts can be added here.

The second tab is for parameter settings. OpenAI’s documentation contains all the .

These settings can be used to fine-tune the content creativeness allowance, wordings that are unusual, maximum token spend per request, and even redundancy of content. The OpenAI API key can also be saved here.

The Python script will launch automatically once a button is clicked in the spreadsheet.

The script first defines the URL of the endpoint / request. The script then sends request headers as well as request data.

The spreadsheet shown previously allows you to change most parameters of the header and data for the request.

The response text from OpenAI is then received and entered into the “data dump”, another separate spreadsheet.

Only one script needs to be run. Two separate spreadsheets are also needed.

The script will resolve all the queries and save all the text snippets here.

You may be concerned about the uniqueness of content if you look at the output above.

The generated content is more varied. While the first snippets all begin with “Introducing the [product ]”),”, the paragraphs are increasingly diverse. It’s actually not as bad looking as it seems.

You can also ask the AI to create unique content. But you will need to be very firm in your request.

The temperature and frequency parameters can be adjusted to enhance creativity and reduce redundant language.

By combining these technologies (OpenAI API, Excel and Python), we are able to quickly generate text for any input prompt.

It’s now up to you to decide what to do with the newly processed data.

It is highly recommended that you convert it to a format that your editorial team will understand.

By creating rich prompts, we have managed to mitigate this problem. You can never be sure until you see the output.

ChatGPT output Notes

If you are happy to use ChatGPT there are a few things that you should keep in mind.

• Let’s talk cost. It is difficult to provide a breakdown of the costs associated with using OpenAI’s GPT-4 ChatGPT model via their API. The input and output word counts are not the only factors. The AI’s thinking time is what determines pricing. More complex requests use more tokens, and therefore cost more.
• We only paid $1.74 for each 100-prompt test set we generated from our sample data. Overall, we generated 22,482 content words. 22482 words for $1.74 is a good deal, but you should consider more.
• A human-led editorial process (in our opinion) is fundamentally necessary due to AI’s tendency to infer.
• This technology can transform the expensive task of creating content from scratch into a more cost-effective task of content editing.
• You must also factor in the time it takes for the data specialist / AI specialist to create and run scripts.
• AI can “creatively” infer things, in addition to determining where there is a lack of data. In our data sample, the AI inferred the existence of a clothing sizing chart (sizing guide) from the content produced. It would be pretty ridiculous if there was no sizing chart.
• Send AI content to a human editor for a fact-check, accuracy and (most important) creative flair.
• ChatGPT can be further automated by integrating projects such as Auto-GPT. These AI ‘agents,’ give ChatGPT more processing power and tasks. OpenAI API keys are still required for projects such as this. Due to their infancy they can eat up credits before learning to perform tasks according to standard.

AI can help you scale your content creation processes

AI can produce diverse content snippets that are suitable for the purpose, with minimal intervention.

It’s still probably better to use the AI interface for long-form content and then iterate on the AI’s response.

The article How can I use ChatGPT for product descriptions on a large scale first appeared on Search Engine land.