How to check if you're reading AI content

Like it or not, AI adoption is growing rapidly, shaping our digital landscape.

From articles and blog posts to product descriptions and beyond, artificial intelligence (AI) has revolutionised producing and consuming content. 

For content marketing, this opens as many doors as it does ethical conundrums. The consensus seems to be that we should be using AI as writers to speed up our processes, but perhaps not taking outputs word-for-word. 

Many people think that Google has been an AI crackdown since the helpful content update at the start of 2023, but this isn’t technically the case. Before the widespread adoption of ChatGPT, Google generally considered all AI content spammy and low-quality. However, their guidance was updated at the start of February 2023 to reflect that AI tools can be used to create helpful content. They now say: 

“not all use of automation, including AI generation, is spam …. AI has the ability to power new levels of expression and creativity, and to serve as a critical tool to help people create great content for the web.” - SocialMediaToday

With this in mind, it’s increasingly important that content writers don’t just prompt tools like ChatGPT to write for them and stop there. At best, you might be able to rank first before other more quality content is available, but your readers will have a slightly uncanny experience. At worst, you’ll have to revisit and rewrite the article anyway because it’s not quality enough to rank. 

Whether you’re a content producer looking to improve your AI content, a team lead or a client trying to discern why that content piece doesn’t sound quite right to you, or just trying to snoop on your competitor's production methods, there’s plenty of markers to look for to detect how likely it is you’re reading AI content. 

Telltale Signs of AI-Generated Content

Due to how it’s produced, there are several telltale signs that AI has written content rather than a human. The AI surge has come from huge advancements in natural language processing (NLP) models. These enable machines to analyse and understand the patterns of human language. These models can then generate text that mimics the style and structure of human-authored content. But this is often where AI content misses the mark: machines can mimic but not understand the logic behind patterns, so they may miss elements or put language together in a way that a human might not. While AI offers efficiency and scalability, it lacks the nuanced creativity, empathy, and cultural understanding inherent in human communication. 

From my experience, some tell-tale signs of AI content include:

• Lacking human nuances and emotion: this is especially evident if you’re creating or reading content where the brand has a more informal, fun or even tongue-in-cheek tone. AI can’t replicate uniquely human turns of phrase or add personal anecdotes and stories like many expert authors do to liven up their content. If the content is fairly to the point, with little deviation from standard fact, then it could be AI content.
• Repetition: Remember, any AI tool is only as good as the data used to train it, which means common or freely available tools have decent but not in-depth knowledge of most topics. So they can create good articles, but beyond a certain word length, they can start to repeat the same information or touch on the same sub-topics repeatedly.
• Errors: Equally, AI tools don’t have their own ‘brains’; being as good as the data they’re trained off, they can make frequent small errors or produce information without a source. As AI becomes more sophisticated, expect this one to crop up less and less, but if you can spot small errors throughout a piece it’s still obvious that the content hasn’t been created or reviewed much beyond a machine
• American English: All those tools coming through? Since most of them are from Silicon Valley, they won’t write in UK (or even Australian) English unless you tell them to. This is more of a telltale sign that someone may not be creating AI content very carefully, but it’s definitely something to look out for.
• Coherence and flow: As AI outputs mimic human patterns of writing, they can lack the kind of flow between sentences and paragraphs that human writing has. Each sentence serves a specific purpose, which doesn’t mean each sentence follows the previous subject. This can be harder to pick up on, but if you feel that the content you’re reading is a bit choppy and fast-paced with quick transitions between topics, this could be AI content. 
• Listicles: My personal favourite sign is one that most often comes from content created by ChatGPT. When prompted to write any kind of blog article, generally, it will break down the subtopic into either bullet points or numbered points, even when this means these are repeated or very similar to one another. Personally, I think it even does this for sections that could be covered by paragraphs, but that’s my personal opinion. This alone doesn’t mean the content you’re reading is generated by AI but if that’s a consistent change in style from previous blog articles, then it could be a strong tell. 

How to analyse AI content

Analysing AI content can be the same as analyzing normal content, depending on available resources. 

If all you’ve got is time and the will, you can get your best editor/proofreader hat on and look for the above signs to explain any strange uncanniness or feeling you get while reading the content. You’ll likely find common patterns and turns of phrases that whatever tool was used to generate the content understands and draws on by going through several articles. For competitor analysis, you could compare articles written recently with those they’ve published in previous years. However, if there’s no credited author, any change in tone or style could also come from staff or brand changes. 

Equally, if you suspect content was generated using ChatGPT or a tool you can easily access, give it a go trying to generate similar content with the kind of prompt that could have been used, e.g. write me a blog article on X for Y audience, in Z words. It’s then easy to compare your output to the finished piece and detect similarities. 

Finally, if you’ve got a bit more budget to back up your AI content detective work, plenty of AI tools exist purely to detect AI content - it’s the circle of NLP life. Usually, these can be accessed for a fee, so only consider getting into these if you need to check content regularly. 

AI’s capabilities grow rapidly every day, and there is no doubt that there will be some point in the near future where even the tips in this article will become obsolete! However, we all need to remember that however we’re using AI, it’s still a tool in human hands. It’s not magic fairy dust that makes everything better; as with any other tool, if you put crap in, you get crap out. It might work to make a quick buck or rank, but ultimately, the generation of AI content comes back to the same central idea that content marketing always comes back to - it’s quality over quantity.