As well as the suggestions for detecting AI-generated content elsewhere in this section, AI-written text has a distinctive style which appears over and over again. Obviously, use of the signs here - such as the words and phrases mentioned, does not conclusively prove that a text was written by AI. "Delve" is a favourite word of AI chatbots, but use of "delve" in a piece of text doesn't mean it was AI-written, use of "delve" and some of the stylistic features mentioned here might make you more suspicious, however.
There are no set rules for identifying AI-generated text, but here are some stylistic features typical of AI chatbots such as ChatGPT. To reiterate, the use of these features does not 100% guarantee the text is AI-generated, after all, these generative AI tools are trained on human writing. This is simply a list of features noted from many texts known to be AI-generated.
AI writing is formulaic, with repetition of words, phrases, sentence structure and length (see the rule of three below)
AI chatbots often emphasise the importance of the subject with reminders that it represents or contributes to a broader topic.
Chatbots often introduce value judgements in their writing, usually inflating the importance of a topic as above
Humans use transition words and phrases e.g. "furthermore", "however" "in addition, "to help the flow of their writing and connect ideas. AI chatbots tend to overuse them, in a stilted and formulaic way.
AI is software so it cannot provide personal anecdotes, emotions or unique perspectives that humans use to personalise their writing.
AI-generated writing often tends to overuse hedge words and qualifiers. Hedge words are terms that create ambiguity or uncertainty, such as "might," "could," "perhaps," and "generally." Qualifiers include phrases like "somewhat," "often," and "in many cases."
AI writing has a tendency to present information in a predictable problem-solution structure, often lacking depth, such as "Have you ever wondered..." or "What is the significance of...?"
AI chatbots will often end an answer by summarising and restating its core idea. Look for an end paragraph or last sentence beginning "In summary..." or similar.
As well as summarising what has been previously written, AI chatbots often include with a 'challenges' section that will begin "Despite" and often end on a positive note. This may appear instead of, or in addition to, a summary.
A parallelism is a grammatical device that using the same pattern of words to show that two or more ideas have the same level of importance. A negative parallelism does so with negative words such as "no" or "not", so a piece of AI-generated text might have the construction, "not only [X] but [Y]"
AI chatbots tend to insert superficial analysis of information, often in relation to its significance, recognition or impact. This is often done by attaching a present participle ("-ing") phrase at the end of sentences, sometimes with vague attributions to third parties (see below). These comments are generally unhelpful as they introduce unnecessary or fictional opinions.
AI chatbots tend to attribute opinions or claims to some vague or unsourced authority ("[X] has been described as..."), a practice called weasel wording, while citing only one or two sources that may or may not actually express such views. They also tend to overgeneralise a perspective of one or few sources into that of a wider group.
These are phrases that consist of a preposition followed by a noun or pronoun and additional modifiers, which can include adjectives, adverbs or other phrases. Their use can be clichéd and convoluted e.g., "taking into account", in the context of"
When giving examples of items within a set, AI chatbots will often mention these items within a phrase that reads "from ... to ...". e.g. "Somali cooking includes a range of meats from beef to chicken". "From beef to chicken" is not a range of meats.
AI chatbots overuse the 'rule of three'. This can take different forms e.g.,"adjective, adjective, adjective", "short phrase, short phrase and short phrase" or sentences with three clauses.
Unless specified otherwise, AI chatbots mostly use American English as their default language - most AI chatbots extant have been created by American companies after all. This means omitting the "u" in words such as colour, humour and behaviour and using "z" rather than "s" in -ise constructions e.g. "organize", "summarize"
AI chatbots may display various phrases in boldface for emphasis in an excessive, robotic manner.
AI chatbots use the Oxford Comma, also called the serial comma, is a comma that appears after the second-last item in a list. For example, without it, the colours of the Irish flag would be written as "green, white and gold". With it, it would be written as "green, white, and gold". ChatGPT uses the Oxford comma, which is a good to reason not to use it, unless really necessary (for example to avoid ambiguity).
Not a word or a phrase, but something that appears in ChatGPT-generated content than in human writing is the em dash - the long dash
as opposed to the en dash
The em dash is a punctuation mark that can replace commas, parentheses and colons and is usually used to link clauses, especially when the clause after the em dash provide explanatory or extra information relating to the preceding clause. The em dash is normally seen as standing out more than commas etc, and is often used as a stylistic device to highlight or draw attention to a particular bit of information.
When was the last time you used an em dash? When was the last time you saw an em dash? ChatGPT uses then very frequently.
In section headings, AI chatbots strongly tend to consistently capitalise all main words. As does the Digital Library. Oops.
AI chatbots often organise the contents of their responses into lists that are formatted in a particular way - numbered list or bullet points. The digital library uses lists, but it doesn't use AI, honest!
One thing to look out, as mentioned above, for is the use of certain words and phrases; ChatGPT in particular has a penchant for overusing words and phrases like:
The linked document below has been compiled from various Internet sources. Again, it has to be stressed that the use of these words and phrases in a text cannot definitively identify a text as having been written by ChatGPT; this is simply a list of words and phrases that appear in ChatGPT-generated output more often than in human writing
ChatGPT Overused Words and Phrases List
Here is a piece of text that incorporates many of the stylistic devices described above. How many can you find?
In today's fast-paced and rapidly evolving digital landscape, AI writing has emerged as a game changer— a powerful tool that is a gateway to enhanced productivity, efficiency, and creativity. At the forefront of this revolution, AI-generated content often captivates readers with its compelling journey through complex topics.
However, while it serves as a testament to innovative technological advancement, it often lacks the emotional depth and personal touch that human writers bring to the table. AI-generated content frequently employs clichéd phrases like "at the end of the day" and "in a nutshell," which can dilute the originality of the writing.
Additionally, it tends to rely on hedge words such as "arguably", "somewhat", and "generally," creating ambiguity that may leave readers wanting more clarity. While AI can delve into intricate subjects, it generally does so with a surface-level analysis, failing to explore the complexities of a topic.
Furthermore, the use of problem-solution language can lead to a formulaic structure, making the writing feel predictable and uninspired. It excels in generating content quickly, yet it struggles to evoke genuine emotion or connection. The reliance on overused expressions, such as "cutting-edge" "game changer", and "thinking outside the box" can make the writing feel clichéd, vapid, and unoriginal. Despite its ability to produce vast amounts of text, AI writing often lacks the nuance, heart, and creativity that characterize compelling narratives. In this rapidly changing digital era, it is crucial to navigate the complexities of AI-generated content with a discerning eye, recognizing both its potential and its limitations.
Ultimately, while AI writing represents a vibrant tapestry of innovation, it is vital to remember that the human touch remains irreplaceable in crafting truly engaging narratives. In conclusion, it can generate words, but it cannot generate warmth; it can produce information, but it cannot produce inspiration.
A few of the signs of AI writing have been taken from the Wikipedia page Signs of AI writing, available under the Creative Commons Attribution-ShareAlike 4.0 Licence
