I try not to judge people for using AI for writing. People may have different requirements, and different scenarios may have different rules. Yet, everyone would agree that it is definitely a disappointing situation when something you wrote is flagged as AI-generated. Sure, AI writing detectors aren’t always reliable, but there’s another possibility: your writing sounds like ChatGPT.
As you probably know, AI content detectors work by analyzing the structure and writing patterns. So, if you use the same writing conventions as ChatGPT, your writing is likely to be flagged as AI-generated. You may have learned to write like that from the beginning, or the increasing amount of AI-generated content on the internet may have gotten to you — let’s not judge.
Here, you will find five common writing habits that make you sound like ChatGPT even when you are writing on your own.
4 Reasons Why AI Checkers Might Flag Your Writing
Many schools use AI checkers to flag students suspected of writing with AI. However, these are ineffective and often lead to false-positive flags.
Some words have become AI tells
Researchers are already tracking them by model era
If you regularly read content generated by platforms like ChatGPT, this might not be news to you. AI-generated content is known for (over)using a specific set of words. Sure, different models from different developers have had different sets of words.
For instance, during the GPT-4 period, it was difficult to find a ChatGPT response without words like tapestry, delve, pivotal, underscore, or vibrant. In later models, we saw other sets, such as fostering, highlighting, and enhancing. While these specific words change over time, AI detectors and people looking for AI content often associate them with ChatGPT.
Therefore, even if these words were your favorites before ChatGPT, you may want to search for alternatives. Even better, you can try not to overuse them.
Everything gets listed in threes
Triplets feel intentional until you notice they never stop
The rule of three is another writing habit that can lead to your content being tagged as AI-generated. ChatGPT and other chatbots tend to present information in a three-component structure. It could be a combination of three adjectives, three nouns, or three short phrases, depending on the context. Sometimes ChatGPT may use “other elements” or “other options” as the third entry in this rule of threes.
Therefore, following this structure is another common reason why your writing is labeled AI-generated. Changing this style wouldn’t be difficult, though. You can do it by modifying the number of items in the list.
“Not just X, but Y” adds length, not meaning
Both halves usually say the same thing at different volumes
Content from ChatGPT also tends to use another common structure for emphasis, which can often label your content as AI-generated.
These are called negative parallelisms and have been part of writing culture for quite some time. They are also mostly used for emphasis. LLMs that power platforms like ChatGPT tend to use these constructions more often than not to make things clearer to readers.
For instance, instead of saying that something is x and y, LLMs tend to say that something is not only x but also y. At times, it will also use an em dash to separate the clauses. Therefore, if you constantly use constructions like “it’s not this — it’s something else,” there’s a high chance AI detectors would label your writing as AI-generated.
“Is” and “are” quietly disappear
“Serves as” and “features” replaced them — and someone noticed
The noticeable disappearance of “is” and “are” is another common sign of AI-generated content. Instead, chatbots like ChatGPT tend to use constructions like “serves as” and “features”. For instance, instead of saying that “the new smartphone from Google has a 50MP camera,” ChatGPT tends to say that “the new smartphone from Google features a 50MP camera.”
Once again, there are times when your writing demands the use of this structure. However, overuse of the construction risks being labeled ChatGPT-like. Therefore, when you can simply use ‘is’ or ‘are,’ do so.
Factual sentences get interpreted before the reader can
The trailing -ing clause is doing analytical work
This is more of a writing habit that has probably been picked up because you have been reading quite a lot of ChatGPT or AI-made content. LLMs like ChatGPT tend to interpret a factual sentence by inserting a comma between the words. Here’s an example:
“As of the April 2008 census, the population of Douera stood at approximately 56,998 inhabitants, creating a lively community within its borders.”
Here, the phrase following the comma is essentially analyzing what the earlier part of the sentence says. The problem here is that the sentence does the analysis/synthesis before the reader gets to it. It may not be a great idea from a readability standpoint, either.
ChatGPT’s New Creative Model Is “Good at Writing,” but I’m Not Convinced
Sam Altman certainly seems impressed, but I still think humans are better writers.
AI didn’t invent these habits — it just made them impossible to ignore
None of the writing styles mentioned here was invented by AI, per se. Instead, we have the AI training processes and internal biases to thank/curse for the overuse of such constructions in AI-generated content. Nevertheless, given how widely used AI detectors are, we cannot continue using these writing habits. Therefore, if you are in a position where getting your work labeled AI-generated is risky, you may want to avoid these writing habits.
- OS
-
Android, iOS, Web
- Developer
-
OpenAI
- Price model
-
Free with optional subscription
ChatGPT is the flagship AI chatbot from OpenAI, and it’s loaded with features.
