Opinions expressed by Entrepreneur contributors are their own.
Whichever products you sell, they at least need proper descriptions so that your customers can find them and understand what they are looking at. Writing hundreds of descriptions is a chore, though, and takes hours, days and weeks.
A big company like Amazon with its 1.5 million employees may not even care about content production costs. For a small business where human resources are always limited, it becomes a real burden. Here’s how you can write content fast.
Facing the problem
Let me introduce myself: my name is Andrei and I’m a senior editor at A1 SolarStore, an American online solar equipment store. There are two types of content that our website requires.
The first one is articles about solar energy and how to use it. This is high-quality content that we keep engaging and helpful for readers: we pick a topic, we do the research, and we spend a lot of time editing and polishing the text. Our efforts bear fruits, but it still would be nice not to spend over 10 man-hours on a single article.
The second is all sorts of service texts. They are SEO-optimized and frankly dull and exhausting to write. In particular, our business needs descriptions for every solar panel that it sells. To make a description, a copywriter needs about 1-2 hours. There are hundreds of solar panels in stock, with new ones being added constantly, so you can imagine the sheer volume of work.
We always seek solutions, allowing us to do more and work less. That’s why we embarked on a quest to minimize the time we spend writing the same things over and over again.
AI can be limiting
Of course, we tried to make ChatGPT write high-quality content for us. While this AI is a wonder, the results were controversial.
As you get familiar with ChatGPT, its flaws shine through. Here are the ones that we found most problematic:
- ChatGPT 3.5 knowledge cuts off in September 2021. You can’t use it if you need the latest information.
- You have to do fact-checking. ChatGPT makes mistakes: some come from outdated databases and some appear just because the bot was taught incorrectly. In-depth articles are what it struggles with, especially if your subject is complex and uncommon to begin with.
- It doesn’t write well. The sentences are wordy, dull, and complex. Editing them sometimes takes more time than writing a piece from scratch.
When your blog has an established style, the quality of content is important and you need up-to-date info, ChatGPT capabilities are limited. We still use it a lot though. Rather than making it write articles for me, I ask it to explain difficult concepts for me, do the math, brainstorm ideas for headlines and many more.
Finding the right prompts is important. A query like “Explain x in simple terms”, “Explain x to me as if I’m a beginner” sometimes returns answers that you can use in your articles as is. Be specific about your task. Set up limitations: “Write two paragraphs, three sentences each”.
If you don’t like what you get, try breaking up the task into smaller pieces. Instead of asking to write an entire article, try asking ChatGPT questions that different parts of your text should answer. If you have a part that gets repeated often in your articles, I recommend trying QuillBot for paraphrasing.
Some articles just can’t be written with ChatGPT. With some, AI does most of the job and the result needs just a little polishing. Overall, AIs have proven to be immensely useful and have sped up our work by 20-60% on average.
Unfortunately, we couldn’t make ChatGPT write product descriptions for us. However, our copywriter Max got so fed up with depicting different solar panels that he devised his own solution. This is where we’ve got our own breakthrough with how to write content fast.
Product descriptions follow or can be made to follow a similar structure. In the case of solar panels, the content is also very similar. There are a few technologies and features that may be different across individual solar panels, but there are not many overall. To give an example, solar panel descriptions often mention power output, efficiency, weather resistance, design features, etc.
When a product goes out of stock, the efforts that went into talking about it seem wasted. But what you wrote there may be valuable in and of itself. If you like how you conveyed a concept like warranties of a solar panel in a particular description, you can use it in the next one. Rather than just using Ctrl+C Ctrl+V, Max decided to automate the whole process.
Devising the algorithm
A little bit of creativity and basic coding skills can go a long way. This is the process that Max followed in order to automate our descriptions:
- Max identified the common structure and content of solar panel descriptions. After writing descriptions manually for so long, it wasn’t difficult to come up with a universal structure.
- Using the structure, he created a template for solar panel descriptions. This template was divided into multiple logical parts, with some parts being optional depending on the specific panel.
- Max separated existing descriptions into building blocks. He used our stockpile of written descriptions, separating each into the building blocks for our template. This involved identifying the specific information that each description contained and fitting it into the appropriate sections of the template.
- Our copywriter replaced individual recurring data with variables. This would allow us to easily change the descriptions to reflect the specific characteristics of each panel.
- Max wrote a Python script for a program that would create a description based on 25 individual characteristics that would be filled manually for each panel in a Google sheet. Based on these traits, our script would automatically fill in the appropriate details and pick the template sections required for each particular description.
- Finally, we were able to Implement the automated process. We used Google Colab to create a simple interface. Whenever a new panel came, we could just fill the Google sheet with its characteristics, run all cells in the notebook and get a link to the Google doc with the description in just a few seconds.
How to write content Faster: the Results
The program that Max developed for this automation process wasn’t all that complex, and we didn’t even have to send a request to our IT department. The Python script uses Numpy, Gsread and Docx libraries to handle and manipulate data. Manually written and proofread texts meant we didn’t need advanced natural language processing techniques to ensure that the generated descriptions are grammatically correct and semantically meaningful.
Still, the script includes a number of validation checks to ensure that the input parameters are accurate and complete. For example, the script checks to see if the input values fall within a certain range or if they match a specific format. This helps to prevent errors and ensure that the generated descriptions are as accurate as possible.
The solution took us under 30 hours of work, including preprocessing of the manually written texts. It works: it now takes around 10 minutes to write a description for a solar panel. Aside from filling the Google sheet with the panel’s characteristics, which takes about 10 minutes, all that’s required is a couple of clicks. When you build a text like a Frankenstein monster, it isn’t always pretty. But it doesn’t have to be: these descriptions do the job, and the tool saves our copywriters countless hours each week.
All the magic is in making a universal template and preprocessing the available texts. Just 5–7 alternatives for each logical block in your description template can give decent results. Make sure the alternatives are fully interchangeable and really think about the characteristics you’ll use and how they will affect the composition of the resulting description. This will guarantee linguistic diversity and good structure. Shoutout to Max who taught us how to write content faster.