Key takeaways
- Lack of specificity and depth: AI-generated content tends to be less detailed and lacks actionable insights, offering broad statements instead. This makes it easier to tell if something was written by ChatGPT
- Detecting if something was written by ChatGPT isn’t straightforward: Studies have found people can be duped into mistaking ChatGPT content for human content
- Common phrases and generic language: ChatGPT often uses repetitive and overly formal phrases like “in summary” and “ever-evolving landscape.”
- Artificial hallucinations and misinformation: ChatGPT can produce errors and false information, such as incorrect citations and far-fetched claims.
- Predictable patterns and formulaic structure: ChatGPT content often follows a predictable structure with consistent sentence lengths and heavy use of bullet points.
ChatGPT content detection is not always easy for people
Imagine telling someone at the stroke of midnight, as they welcomed the year 2000, that within 25 years, we would struggle to distinguish between humans and machines. Most likely, you would have received a few eyerolls and laughs and been labelled one of those millennium-bug weirdos. Yet here we are. Advancements in artificial intelligence have blurred these lines significantly, and it has become hard to tell if something was written by ChatGPT. As AI becomes more sophisticated, it becomes increasingly challenging to detect ChatGPT content.
However, if you are struggling to detect content written by ChatGPT you are not alone. In a study led by Northwestern researchers, they found a team of scientists could be duped by ChatGPT. Yet, thankfully, you don’t need several PhDs to identify content written by ChatGPT.
In this article, we outline how you can detect ChatGPT content without any tools. There are techniques you can use to humanise AI content but they require a lot of effort and human input. Moreover, AI content is getting better but it’s still unable to mimic humans.
How can someone tell if I used ChatGPT?
ChatGPT is like Ross from Friends. It can appear smart but will constantly make the same errors. It’s one reason why trust in ChatGPT has dropped resulting in declining traffic.
However, this doesn’t mean you can look for one clue and, voila, be able to tell with certainty if something was written by ChatGPT. It’s more intelligent than that.
OpenAI, the creators of ChatGPT have said there is no guaranteed way of unearthing content written by the platform. This means you need to look for several clues, which we have outlined in the following steps.
However, even if you managed to avoid detection with your ChatGPT content, it doesn’t mean there are no consequences. For instance, Google is cracking down on mass-produced AI-generated content, meaning your search rankings can be impacted.
Moreover, the legality of AI-generated content and images is debatable. Therefore, even if you can fool a few people or AI detection tools, it doesn’t mean it’s without consequence.
As we move into the future of search, the need to add your insights to content will only grow. The search ecosystem will become more complex, and therefore increase competition.
As a result, the tools people use to search will depend on what they are searching for – increasing the need for a multichannel approach.
Consequently. despite a 317% increase in interest in AI content and a 4850% increase in AI SEO tools, automating your SEO and content strategy will be almost impossible. Instead, you will likely need to do more work as you look to build an AI search strategy underpinned by unique insights for Topical Authority, Entity SEO and AEO.
1. ChatGPT uses common words and phrases: The easiest way to detect ChatGPT content
ChatGPT content often sounds like it’s trying too hard. It’s like the person who needs everyone to know they are intelligent. Yet, rather than sounding like Charles Dicken’s they sound pretentious.
So, look out for some common phrases. These include:
- In summary
- In conclusion
- A rich tapestry
- Delve into
- Ever-evolving landscape
- In the digital realm
- It’s important to note
- A testament to
Perhaps the biggest clue is how these phrases fit a specific text and context. For instance, an article about the National Health Service might say “The digital revolution is coming to the ever-evolving landscape of the National Health Service.” Now, putting all political opinions aside I think we can agree, that the NHS rarely ever changes and the same issues reappear year after year, decade after decade. Nobody with any insight would use such a phrase.
2. If you know the subject, you can spot hallucinations and misinformation
This is not the easiest step, because you need to know the subject you are reading about. For instance, if you are reading about zone 2 running but don’t know the difference between cadence and Vo2 Max, it will be challenging to effectively detect ChatGPT content.
However, AI constantly makes errors or false claims, making it easier to tell if something was written by ChatGPT. Known as artificial hallucinations, it occurs because of the probabilistic nature of AI. This is because platforms like ChatGPT try to predict and produce coherent text based on training data and public information. This means people’s assumptions might be presented as facts when you ask ChatGPT. This can lead to disturbing inaccuracies, as proven in our experiment.
This is because ChatGPT is powered by narrow AI, designed to perform a specific task. This means it can rapidly process large volumes of data, often in a matter of seconds. For instance, swiftly detect patterns or correlations in datasets that may be overlooked by humans. Narrow AI systems depend on high-quality, relevant datasets to learn and make accurate predictions. Biases or inaccuracies in these datasets can lead to flawed decisions or reinforce existing societal biases.
Moreover, a study published in Nature included researchers using ChatGPT to investigate brain regions for antipsychotic treatment. They wanted not only insights but also academic sources. However, they discovered, that ChatGPT provided sources that did not exist or were wrong. More concerningly, ChatGPT also misquoted many of the sources.
So, how do you tell if something was written by ChatGPT? Consider the following:
- Do the citations or links accurately reflect what it’s referencing? For instance, an article about the benefits of weight training on heart health might link to a journal about sleep.
- Are there passages of text that seem far-fetched?
- Does the article provide one-size-fits-all solutions? For instance, eating a healthy diet is the answer to running faster for longer.
3. Even when people use AI, they still make errors
You would think, that if people use AI, they would at least do the basics and make sure their conversation with ChatGPT is not included. Surprisingly not.
It’s not uncommon to see text such as “Here is an updated version.” Moreover, if someone has used ChatGPT to summarise another article you might have strange sentences which would indicate the text is a review. These are potential clues that the text was generated by ChatGPT.
4. You can tell if something was written by ChatGPT when the content is not very specific.
Imagine you are reading an article about a new supplement called NMN. It appears super interesting because the article headline suggests NMN has gone viral online and has risen 480% in interest.
However, the article you read outlines the generic health benefits of NMN. It fails to outline how social media has played a part and what has caused it. This is one way to tell if something was written by ChatGPT.
Moreover, a ChatGPT article can often fail to provide actionable insights. For instance, an article might say to get more search engine traffic, you need to have great content. But what exactly is great content? It’s like saying to become a better runner you need to move your feet quicker. Thanks, only if I had thought of that!
In contrast, human-written content, by an expert will go into specifics. For instance, our article on AI overviews details how to prepare for the phase in search engine development.
5. A lack of personal experiences, insights, anecdotes or metaphors
Have you ever realised people like to talk about themselves? We are all guilty of it, and when writing it is common to inject elements of our personality. So, factor this in when trying to detect ChatGPT content.
Moreover, tools like ChatGPT do not understand human emotion. More importantly, they are not trained on your life experiences or company insights. Therefore, they will often appear devoid of any real-world experiences.
So, if the text you are reading seems like it’s trying too hard to be funny or adding a metaphor that nobody would use, it’s a red flag.
Therefore, consider the example of ORCHA, the leading trusted source of digital healthcare compliance in the world. One reason for the quick growth has been providing a human touch to their content strategy.
6. The use of formulaic text can help you detect if something was written by ChatGPT
ChatGPT content has a predictable rhythm. It will use similar sentence and paragraph lengths. This is because it is designed to follow patterns. However, as humans, we are much more unpredictable. This is another thing to look for to tell if something was written by ChatGPT.
Our writing naturally varies in sentence length and structure, reflecting the nuances of thought and the flow of ideas. For example, if you read an article that reads like a monotonous list, it’s an indication that ChatGPT was used.
One of the key indicators is the use of lots of bullet points. While they offer a great way of simplifying information, it’s unlikely a human will use them without much commentary.
7. It’s easier to detect ChatGPT content the longer the text is
The longer and more complex a piece of content is, the harder it will be for ChatGPT to avoid detection. This is because shorter texts allow ChatGPT to rely on its pre-trained patterns more effectively.
It can mimic human writing style for a limited scope. However, as the text length increases, the repetitive nature of these patterns becomes more evident. You might see the same sentence structures reused, similar vocabulary choices throughout, or a reliance on generic phrases instead of specific details.
Therefore, make sure you read a few paragraphs to detect ChatGPT content
How do I tell if something was written by ChatGPT? Can I use an AI detection tool?
While it is possible to use an AI detection tool, they are not always effective. In our research, we found AI content detectors are only correct 66% of the time.
Moreover, people are getting better at tricking AI content detectors, by manipulating ChatGPT’s content output. Therefore, you shouldn’t rely on them completely.
Will detecting ChatGPT content become harder?
As AI becomes more advanced, it will inevitably be harder to detect ChatGPT content. OpenAI has already started building more partnerships with leading media companies to help enhance and train its data sets. Consequently, it’s likely in the next few years it will become almost impossible for humans to differentiate between human and AI content.
However, that doesn’t mean AI content will be effective. Companies like Google recognise the need for original content to continue enhancing their AI capabilities. Without original content, we run the risk of an age of information loss rather than human enhancement.
Therefore, expect AI content detectors to become more advanced making content created by ChatGPT easier to detect.
0 Comments