How to tell if an article has been written by AI

The world of online content is changing, with a growing number of organisations and publications using AI to create or assist in the creation of articles, blogs and website copy.

With media conglomerates like News Corp admitting to using AI to publish over 3000 articles a week, It’s understandable that many are questioning whether the news sites they frequent are also dabbling in the world of AI. 

In response to the discourse, several tools have been developed to detect whether a piece of content has been written by AI, including CopyLeaks, Winston AI, and OpenAI’s classifier. 

Turnitin, a similarity detection system widely used by universities, has also launched an AI writing indicator to help educators identify assignments that have been written or created using artificial intelligence.

Even without the use of these programs, there are signs that can help readers spot when something has been written by AI.

Though the absence of these won’t guarantee that an article was written by a human, this list can be used as a basic guide to help identify AI generated content. 

1. Repetitive language and structure

If there is one thing AI is known for, it’s creating repetitive content. AI language models like ChatGPT have a tendency to follow the same sentence structure over and over, which can make a piece of writing sound robotic. There are also specific words AI language models just can’t get enough of, including ‘enigma’, ‘labyrinth’ and ‘tapestry’.

This is because, while articles written by journalists reflect human thought and creativity, AI uses a formula to write its content. Though human writers may be guilty of repetition, they have the ability to revise and edit their work to ensure it sounds professional, engaging and unique.

2. ‘Monotone’ writing

While each journalist has their own, unique writing style, AI language models have been trained on data. This means that, while human writers will often vary their tone throughout a piece, for example, switching between factual and emotive language, articles generated by AI will have one concrete tone, and may feel repetitive or robotic. 

3. Lack of ‘human touch’

AI will focus on the technical aspect of writing, filling an article with facts and information rather than adding personality or emotion to the content it is generating. 

On the other hand, Journalists bring their own voice to their work, and may incorporate emotive language, slang, or references to other events into their articles. 

4. Mistakes and inaccuracies

The quality of an AI is determined by the data it is trained on. Though some are more accurate than others, many are still prone to mistakes, such as providing incorrect definitions, false information or making typos and grammatical errors.

Though it’s certainly possible for a journalist to make mistakes in their writing, this risk can be managed by performing thorough background research and proofreading articles before publishing. 

5. Bizarre errors

While AI is becoming increasingly able to replicate human writing, there are times when it will generate ‘facts’ that are obviously false, insert unrelated information into a piece of writing or become stuck in a bizarre loop of repetition. 

Though a lack of incorrect facts or seemingly random comments doesn’t indicate an absence of AI involvement, the presence of these things could be a sign that something other than human writing is at play.

6. Lack of relevant sources

A lack of both primary and secondary sources is another indicator that an article may have been written by AI. 

An AI language model cannot source interview subjects, nor can it attend a press conference or watch a politician speak on the news. Instead, an article written by AI will only focus on the facts that are available to it.

In contrast, journalists can use all the sources available to them, perform their own research, interview sources and use their knowledge and experience to ensure their articles are as informative as possible.