The best way to use ChatGPT to generate product descriptions at scale

We wish extra high quality content material for our web sites, but it surely’s tough to supply sufficient. So how can we scale the content material creation course of, particularly for ecommerce websites with loads of merchandise?

If you happen to have been to pay for a copywriter to supply 1000’s of product snippets from scratch, you’d probably be out of pocket fairly rapidly.

What in the event you pay for 1,000 new product descriptions, however solely half of these merchandise stay one month later? Clearly, you want a sooner, more cost effective strategy. That is the place ChatGPT will help.

ChatGPT’s native internet interface is admittedly useful and an awesome time-saver. 

But when we’ve got a whole lot or 1000’s of product descriptions to create, there’s a extra environment friendly means of utilizing ChatGPT with out copying and pasting prompts. Right here’s how.

Mass manufacturing of content material snippets: Scaling the output

When you’ve got an ecommerce web site, you would possibly want to produce product snippets utilizing knowledge from a product info administration (PIM) system. 

Let’s say you will have the information on a spreadsheet.

We are able to use Excel formulation to concatenate (or be a part of, utilizing the “&” operator) knowledge into wealthy prompts, prepared for ChatGPT. For instance:

Notice that your formulation might require one or a number of “IF” statements. That’s as a result of your knowledge might have holes in some areas. 

For instance, some merchandise might not have sure parameters (knowledge inside sure columns) specified. You want your formulation to be versatile, and you may at all times ask ChatGPT that will help you write the formulation.

As soon as your formulation is returning a immediate for every row (on this case, for every product), you’ll be able to copy and paste a number of of the generated prompts right into a phrase processor, even notepad. 

It’s good to spot-check a number of to make sure the textual content is smart, even when some knowledge gadgets have been lacking.

Upon getting verified that your Excel (or Google Sheets) formulation is producing the forms of prompts that you really want, you’ll be able to ship a number of of them to ChatGPT (manually, utilizing the online interface) to see in the event you just like the outcomes.

The generated snippet(s) will probably require human editorial oversight, although you need to get the AI to do as a lot of the work as potential. That’s why we engaged in such a deep “prompt-crafting” course of.

Comfortable together with your preliminary prompts and responses? Good, then it’s time to maneuver on.

Fetching your new product content material snippets from OpenAI

So, you now have a listing of merchandise (or different forms of webpages) which you’d prefer to generate content material for. 

On this instance, we’re going with a fictitious pattern of 100 merchandise. You now have a listing of all of your merchandise (both separated by URL, SKU or another distinctive identifier). 

These merchandise even have assigned wealthy prompts which you will have generated. However ChatGPT’s internet UI is restricted. So how are you going to ship these throughout abruptly?

For this, you’re going to must get snug with fundamental scripting and with dealing with API requests. You’ll be able to create an OpenAI API account to entry the ChatGPT internet interface. 

I put collectively a fundamental Python script for my company. Whereas I can’t share the script, I can overview a number of the processes and documentation wanted.

Ought to I want to syndicate this script later, constructing it on marketing-accessible endpoints and applied sciences could be greatest. As such, I first produced an Excel sheet:

The sheet merely offers an space to dump gadgets for processing (recognized by some distinctive identifier within the “Merchandise Title” column, on this case, product title). As well as, the prompts to be processed will also be positioned right here. 

One other tab accommodates parameter settings for the request. (You’ll be able to find out about all these through OpenAI’s documentation.) 

A few of these settings fine-tune content material creativity allowance, uncommon wording deployment, max token spend per request and even content material redundancy. That is additionally the place the OpenAI API secret’s saved.

As soon as a sure button inside the spreadsheet is clicked, the Python script launches routinely and handles the remaining:

First, the script defines the request / endpoint URL. After this, the script sends the request headers and the request knowledge. 

Most parameters for the request header / knowledge, could be tweaked inside the spreadsheet pictured beforehand. 

Lastly, the response textual content is obtained from OpenAI and logged inside the “knowledge dump,” one other separate spreadsheet. 

I’ve three scripts for this deployment, although just one must run. I even have two separate spreadsheets, each of that are wanted.

As soon as the script resolves the entire queries, the entire snippets of textual content are saved right here:

Trying on the above output, you will have some content material uniqueness issues. 

Whereas the entire snippets start with the precise phrase (“Introducing the [product name]”), the produced content material will get extra various throughout the generated paragraphs. So, it’s not as dangerous because it seems. 

Additionally, there are issues you are able to do to try to make every generated snippet much more distinctive, equivalent to categorically asking the AI to generate distinctive content material (although it’s important to be fairly agency and repetitive on this regard to get anyplace). 

You too can tweak the temperature and frequency parameters to regulate content material creativity and keep away from redundant language.

Weaving these applied sciences collectively (OpenAI’s API, Excel, Python), we are able to rapidly confirm generated textual content snippets for all enter prompts. 

From right here, it’s as much as you what you need to do with that newly processed knowledge. 

I extremely advocate shifting it right into a format your editorial crew can perceive. 

We’ve got considerably mitigated a lot of this by crafting very wealthy prompts. Nonetheless, you’ll be able to by no means make sure till you test the output.

ChatGPT output notes

Assuming that you simply’re pleased to work with ChatGPT, there are a few issues to remember:

• Let’s discuss the fee. It is powerful to provide a price breakdown for utilizing OpenAI’s GPT-4 mannequin of ChatGPT through their API. It is not simply the enter phrase rely of the immediate or the output phrase rely. Pricing revolves across the AI’s “considering time.” Extra advanced requests will use extra tokens and value extra (even when the enter / output phrase rely is diminished).
• Our check batch of 100 prompts from pattern knowledge value us solely $1.74 to run and return. We generated 22,482 phrases of content material total. 22,482 phrases of content material for $1.74 appears good, however there’s far more to contemplate.
• Because of AI’s propensity to deduce, a human editorial course of continues to be essentially required (in our opinion).
• Nonetheless, utilizing this know-how does rework a expensive from-scratch content material creation process into a way more cost-effective content material enhancing process.
• The information / AI specialist’s time for immediate crafting and operating scripts should even be factored in.
• On prime of inferring the place knowledge is missing, AI also can “creatively infer” issues. In our pattern knowledge set, the AI determined to deduce the existence of a sizing information (clothes) inside the produced product content material. If no sizing information existed, that may look fairly foolish.
• At all times ship AI content material by means of a human editorial overview course of for fact-checking, accuracy and (most significantly) further artistic aptitude.
• You’ll be able to additional automate ChatGPT by plugging in tasks like Auto-GPT. These AI ‘brokers’ add extra lively processing and tasking energy to ChatGPT. Nonetheless, tasks like this nonetheless want your OpenAI API key. And as a result of their infancy, they will chew up plenty of credit earlier than they study to carry out duties to straightforward.

Scaling your content material creation course of with AI

AI can scalably produce various snippets of content material which might be match for goal with minimal intervention.

For long-form content material, it’s most likely nonetheless higher to make use of the interface and iterate the AI’s responses.