How to Detect an AI-Written Essay

They say that AI is a force for good in the world, but anyone who works in academia will tell you otherwise.

The standard plagiarism checkers used in schools and colleges are being outwitted by the likes of ChatGPT. Worse still, all those tools that claim to specialize in detecting AI content have a habit of returning false positives.

So it looks like you are going to have to identify AI generated essays the old fashioned way: using a checklist.

Drawing from a dataset of 50,000 responses from 7 of the most popular chatbots, we have found several indicators that suggest an essay wasn’t fully written by a human.

Unnecessary adverbs in the introduction

School-age students love to add color and life to their writing through the use of adverbs anyway, so you can’t swing down the hammer and declare their work AI-generated on the basis of one too many adverbs.

John Steinbeck’s novella “Of Mice and Men” intricately weaves a narrative that delves into the lives of marginalized individuals amidst the backdrop of 1930s America.

However, if they appear in the very first sentence, this should arouse suspicion, as we’ve found that AI chatbots have a tendency to use superfluous language in essay introductions.

Using dashes to expand on points

A well-written essay flows nicely from start to finish. A ChatGPT-written essay likes to bring the reader to an abrupt pause from time to time by inserting dashes and going on a bit of a detour.

Through the characters of Crooks, Candy, Lennie, and Curley’s wife, Steinbeck offers a poignant commentary on the societal outcasts—those relegated to the fringes of society due to factors beyond their control.

I think this is down to the fact that these chatbots are informational tools at heart – they want to provide you with additional information at every opportunity, which isn’t always welcome in the realm of essay writing.

Presence of common AI words

For a long time now, users of tools like ChatGPT have observed a phenomenon of favouritism displayed by AI for certain words, to the point where it will shoehorn them into any text (not just essays).

This essay will delve into the fates of these characters, elucidating the broader statement the novel makes about the treatment of those deemed as outsiders in society.

Words like “elevate” and “delve” are two of the most common AI words and you’ll find that a lot of users have blacklisted them. But there will be students out there who haven’t instructed their chatbot to blacklist these words, in which case their presence could be indicative of an AI-written essay.

Use of first-person plural pronouns

Whether an essay is written in first person or third person depends on the topic. First person is ideal for sharing a personal experience and third person is more applicable to argumentative essays.

By examining their experiences, we uncover the pervasive injustices and prejudices that shape their lives, inviting readers to contemplate the consequences of societal marginalization and the urgent need for compassion and equity.

Generative AI tools have an overwhelming preference for writing using first-person plural pronouns like “we” and “us”, even if it’s not the optimal option for the topic at hand. This is because these tools are designed to be friendly chatbots – inclusivity is deeply rooted in their programming.

Final thoughts

AI detection tools are not yet able to determine with accuracy what was written by a human and what was spat out by a large language model.

It’s possible to do a much better job using manual methods, especially when it comes to spotting AI-generated essays. Look out for grammatical oddities like dashes, first-person plural pronouns and excessive use of adverbs.