ChatGPT has fundamentally changed the conversation around Content Creation. People view ChatGPT simultaneously as a saviour and a threat, with authors writing web content and students writing essays having different views on the benefits (or challenges) that ChatGPT can bring.
As a writer, developer, content marketer, and business owner, where ChatGPT fits into the future of written content is top of mind these days. While we’re used to the concept of manual labour being replaced by automated mechanical solutions, the concept of creativity being replaced by an automated digital solution is confronting.
It’s important to understand what ChatGPT is — and, more importantly, what it isn’t — to answer whether the role of content writer will disappear.
ChatGPT in a Nutshell
ChatGPT is simply a model of human-generated text. By ingesting massive amounts of text from websites and books, the model contains:
- A statistical representation of how words and sentences relate to each other.
- The importance of words within a sentence.
- The expected form of a response from a given input.
It was trained on a massive amount of human-generated text and then refined by having humans create thousands of questions and responses to fine-tune the algorithms.
Simply put, ChatGPT pulls together words to create statistically plausible responses by looking at what everyone else has written and composing responses that mimic the forms in which it was trained. ChatGPT combines the text it has consumed and provides a statistically relevant sample, organised according to the written form it’s been trained on.
It generates “new” content in the same way an artist mixes paint to generate a new colour. It can create new phrasing, but it can’t provide new information. At least not new factual. It’s very adept at stating facts that aren’t facts, or indeed even correct. It can’t develop new ideas. It can’t “think” and specifically can’t generate ideas outside what it’s learned. It can’t perform a leap of logic, a flash of inspiration, and has no empathy or subtlety. It has no opinion, no personal style, and any bias it has is deeply rooted in the data it has been trained on.
In other words, tread carefully.
Can ChatGPT Replace Content Creation?
Working in content marketing, it didn’t take long to wonder: Will tools like ChatGPT replace creative writing and content creation?
The short answer is not yet.
I asked ChatGPT to write a piece of content based on a single task, for a specific product and platform, as a guide — classic content marketing to a technical audience.
The task was to “Write a 500-word article with a short code snippet that explains how to install python3-pycoral onto a new MacBook Air.” ChatGPT completed the task with aplomb.
Here’s an excerpt of one part of the article:
- Now that we have Python installed, we can move on to installing pycoral. Pycoral is a Python library for Google’s Coral Edge TPU. It allows you to run machine learning models on the Edge TPU, which is a specialized hardware accelerator for machine learning.
- To install pycoral, we first need to install the Edge TPU runtime. You can download the runtime from Google’s website: https://coral.ai/software/#edgetpu-runtime. Once you have downloaded the runtime, open the Terminal, and navigate to the directory where you downloaded the file. Then, run the following command:
bash ./edgetpu_runtime_20210819.zip
A little further down, the article reads:
- Finally, we can install pycoral using pip, which is the Python package manager. Type the following command in the Terminal:
pip install pycoral
ChatGPT’s text is wonderfully clear, to the point, action-oriented, and practical.
And completely wrong.
PyCoral, Coral.AI and macOS
The pycoral package is, in fact, a Python library that acts as a command-line interface for the Allen Coral Atlas. The library you actually need to install is python3-pycoral, but you need to install that using a package from Google. The issue here is that the packages required to install PyCoral for Coral.AI only support macOS version 11 on an Intel-based Mac or macOS 12 on an Apple Silicon Mac. macOS is currently version 13, so you can’t, in fact, install PyCoral for Coral.AI on macOS.
Code Samples
Let’s dig a little more.
Consider:
bash ./edgetpu_runtime_20210819.zip
This, unfortunately, doesn’t make sense. Bash is a program to run scripts, typically files ending in .sh. Bash can’t run a zip file. It needs to be expanded first, and then any scripts inside may be able to be run.
Moreover, there’s no such file as edgetpu_runtime_20210819.zip in Google’s GitHub repo. There’s a 20210119.zip and a 20201204.zip, but no 20210819.zip.
Subtleties
In the text produced by ChatGPT the code samples containing the script are all marked as Bash. This is fine and works on a Mac, but macOS uses zsh instead of Bash as the default scripting engine.
While that may seem like splitting hairs, creating content that feels like it’s written by someone who lives and breathes that topic is important. Use zsh or bash on a Mac or PowerShell versus CMD on a Windows Server.
Are variables camel-case, kebab-case or snake-case in code samples? The details matter and your readers will know.
The Core of the Issue
One could argue that I used a trick question. To this, I’d say that it’s not the nine pieces of content you get right that build a reputation. It’s the tenth piece, where you crash and burn.
Content Marketing strategies can be presented to a marketing team by anyone: the CTO, the marketing head, the developer in the hallway who flipped off a quick idea, the summer intern, an Internet search, or a specialist in content marketing. Choose carefully.
An article’s core idea might sound, but its scope may be expanded without checking. “We have an article on Coral.AI for Windows and Raspberry Pi. Can you do one for macOS too?” So much is cross-platform these days that it’s a very innocent request. You need to ensure each piece of content is realistic and makes sense.
You need a subject matter expert to review topics and content and review it carefully. An expert would, hopefully, tell you that on macOS, you could just use TensorFlow-Lite and load the Coral.AI library as a delegate to the interpreter. Not exactly something your average content marketer would, or indeed should, know. It’s too technical, and that technical depth means your marketing team is spending more time keeping up with tech than they are marketing trends, tools, and threats.
Another key issue is that ChatGPT knows very little of what’s happened in the last 18 months. Training models is time-consuming and incredibly expensive, and at this point, ChatGPT simply isn’t aware of anything relatively new. Content Marketing in tech is specifically around new. Be very, very aware of this constraint.
Use ChatGPT for What It Is
ChatGPT is essentially a chatbot tool equipped with all the knowledge you’d find in a search engine. The results are similar to you typing a question into a search engine, scanning through 5 pages of search results, visiting each link, and summarising what you’ve found in a form that matches a typical and reasonable human-like response. It’s still based on the masses of information – both good and bad – that we’re used to on the internet. It’s still mostly unfiltered and still filled with all the mistakes everyone else has made.
Ask any developer how good online documentation is, and they will snort at you. Ask them how much misinformation is presented by content writers repeatedly cutting and pasting the same information (clearly untested) in their blogs. Just don’t ask the developer to write it for you: They’re way too busy, expensive, and sometimes too direct and unnuanced to provide context around the strengths and weaknesses of your product.
Use ChatGPT to:
- Generate a framework for an article that the author can fill in. This template will be written in a consistent, professional style.
- Surface ideas you may not have thought about or indeed known about. Remember that ChatGPT is like a search engine with an amazing UX.
- Point you to resources or references you may not be aware of
- Get past writer’s block. Still trying to figure out where to go next? Get ChatGPT to throw a random paragraph that will break the block.
- Reword an awkward or mediocre paragraph into something better.
What ChatGPT Can’t Do for You
ChatGPT can’t come up with a strategy for you that will answer your marketing needs unless you are specific to the point of knowing the strategy you need. It can’t guess what you need and, at best, will only produce a generic approach based on the most likely method contained in the ingested content.
The output from a ChatGPT query depends on your input. Suppose you’re unclear on what you need or have an idea that may be unsound due to not having a deep enough understanding of the market or of the customers using your product. In this case, ChatGPT will happily produce something that’s exactly what you ask for and absolutely not what you need.
When creating great pieces of content, ChatGPT can’t self-edit. It can’t understand if it got the tone correct. It can’t always avoid mentioning a competitor, avoid stating a claim that isn’t true, or, as we’ve seen, create blatantly wrong content.
ChatGPT can’t test code samples by loading them in a test rig, signing up to a service, running through the UI and testing the code end to end. It can’t objectively rate the content for suitability for a particular intent. It can’t think through an issue’s different angles and ensure it has covered them all. It can’t put itself in the customers’ shoes and intuit what they might need. It can’t understand a business’s values and ethics, and it has absolutely no yardstick to judge whether its content is true or false, offensive or biased, culturally improper or simply non-sensical.
By far, the biggest problem with generative AI such as ChatGPT are hallucinations: confident responses from the chatbot that aren’t grounded in the data on which it was trained. Content created by ChatGPT may include completely fabricated “facts” that the AI will stick to doggedly. This issue can be controlled by adjusting the ‘temperature’ (the degree of randomness) for a given query, but the trade-off between generating creative content and accurate content needs to be understood and managed.
Will ChatGPT Take Over Content Creation?
This is the big existential question: will content creators lose their jobs to ChatGPT?
For now, no. ChatGPT will, however, augment content creation. It’s like a search engine that doesn’t know what to search for. It’s a printing press without a story.
An important point in a discussion about Content Marketing is that the actual act of writing content is only one piece in a much, much bigger process of Content Marketing.
Content marketing requires:
- An understanding of a company’s goals and the promoted product. Where does the product fit? What makes it unique? What solution does it provide? To whom is it targeted? In a nutshell, you need to understand why the product was created.
- A strategy for content creation based on what is needed. Is there a need to educate a market on a new product and provide a differentiating story for a new solution to an old problem? Is the content for a technical audience familiar with the product (which allows you to make some assumptions in the article), or will your audience be totally unfamiliar? Is the content to address concerns around the product, the company, or general issues such as privacy or reliability? There needs to be a specific plan for the pieces of content being created.
- The tactic that a given piece will satisfy. The strategy may be sound but within the strategy it may be clear that the marketing funnel shows a need to move a prospective client from awareness to engagement. There needs to be specific purpose for an article.
- An outline that everyone agrees on. The head of marketing needs to be sure this outline fits into an overall plan. The subject matter expert needs to sign off that the outline makes sense, and the marketing team must ensure the content will fit in the slot they have allocated. A 3,000-word whitepaper makes a terrible blog.
- The content to be written. Intelligently, efficiently, correctly, and with voice and tone matching needs. The author must be up to date with the latest information on a product. ChatGPT is, as mentioned, 18 months out of date. This is a critical issue.
- The content must be reviewed for tone, voice, correctness, length, grammar, and spelling. It needs to pass the scratch-and-sniff test for general writing.
- The content needs to be reviewed for technical correctness. All code samples need to be tested. Any systems such as hosted services need accounts created and steps outlined in the content walked through to ensure correctness. It needs to be reviewed by someone who works in the area discussed in the article. It needs that fine-tuning that will remove jarring missteps.
- All the suggestions and issues need to be corrected by an editor. All the rough edges need to be sanded down and paint chips repaired. It then needs to go back for a second look.
- Someone needs to review the final piece from the eye of the marketing team for the company involved to ensure that this piece of work 100 percent checks all the boxes identified in the first step. We’re not throwing paint against a wall. We’re very carefully spinning a tale with a storyline and conclusion for the reader.
- A presentation to the person who will authorise the use of the content. Does it fit their needs? Was anything missed? A subtlety, like an accidental use of the wrong word (think issues like “We no longer use that brand name, we use the new brand name”) can have wide-reaching impact. Content marketing is a service, and if there are any issues, there needs to be a remedy for the client.
ChatGPT can help with, but not replace, the fifth step. There are nine other steps, none of which can be ignored.
All AI-based content creation systems allow you to dig a very big hole very quickly, but it’s up to you to decide where the hole is to be dug, how deep, what shape and what you will fill it with to create something useful.
Use AI to generate ideas, get over writer’s block, discover content you may have missed, and move (to continue the metaphor) large chunks of earth quickly. Generative AI systems such as ChatGPT will generate content quickly that is generally well-written, comprehensive, and with few if any spelling and grammar mistakes. And with absolutely no interest in the article itself.
Use it intelligently as one tool in a much larger set of tools for your work.
