Do you have suspicions that a lot of the content you are consuming is being written by a machine rather than a human?
Or maybe you are a content creator yourself and you want to make it harder for your readers to spot the use of AI in your work.
The obvious thing to do is paste a piece of content into any of the countless AI detection tools on the web and see what the verdict is. But the problem with this method is they all say different things, and users often report false positives.
By the way, our AI humanizer is pretty damn good if you want your content to avoid the tells listed below and bypass AI detection.
Whether you read articles or publish them, the following red flags will help you gauge the likelihood that an article was written by ChatGPT.
1. Tricolons
Once you notice them, you can’t not notice them.
Tricolons, or serial lists, are lists within sentences which have three items. They are actually a really powerful rhetorical device that sound rather pleasant to the human ear, but ChatGPT abuses them.
Here are some three-item serial lists that came from one article I generated in ChatGPT (there was far too many to list them all):
- It refers to the increasing interconnectedness and interdependence of economies, cultures, and societies across the globe.
- This process has been driven by advancements in technology, transportation, and communication.
- It has also led to concerns about job displacement, wage stagnation, and the exploitation of labor in developing countries.
- The exchange of ideas, values, and traditions has led to a blending of cultures.
- Digital communication has made it easier for people to connect, conduct business, and share knowledge instantly.
If that’s not bad enough, sometimes you are treated to two sets of tricolons in the same sentence!
2. Titles
Using colons in titles is a great idea, especially for clarity and engagement. They help break up the title into a catchy hook and an informative subtitle. For example:
- “Small Spaces, Big Style: How to Maximize Your Home’s Potential”
- “The Psychology of Color: How Your Home Affects Your Mood”
I don’t have a personal gripe with this approach — I find myself often relying on colons to squeeze in as much info as possible into my article titles.
I’m merely pointing out that it’s predictable for the chatbot do this, especially for light-hearted topics and magazine-type content.
Could the use of colons in titles be the ultimate ChatGPT ‘watermark’?
3. Conclusions
If you’re reading an article and it signs off with a conclusion, there is a good chance it was produced entirely by ChatGPT.
Conclusions aren’t really a thing in article writing, or at least it’s not normal to explicitly write one using classic academic phrases such as “In conclusion” and giving it its own heading called “Conclusion”, both of which are key hallmarks of a ChatGPT article.
4. Em dashes
Just like humans, ChatGPT is an avid user of dashes to expand on a point or to make a tangential point. But unlike humans, it sticks to one type of dash religiously.
The em dash isn’t even a standard key on most keyboards which is why most folks, and therefore most human-written articles, employ hyphens to do the job. This would normally be either a single hyphen (-) or two hyphens together.
ChatGPT on the other hand won’t settle for anything other than an em dash (—), which you can see is considerably longer than a hyphen.
5. Common AI phrases
This one is a pretty big tell-tale sign of AI-generated articles and it also has wider ramifications for indexing and ranking in Google, so pay close attention if you’re somebody who publishes AI articles.
ChatGPT has a really bad habit of using the same phrases in the same locations when writing an article. This includes phrases like “In this article, we’ll explore…” and “In this comprehensive guide…”
If you put out an article with either of the above phrases in the introduction, it will instantly be flagged by Google because your content is identical to millions of other AI-written articles.
We’ve been tracking ChatGPT’s favorite phrases and you may find our ‘blacklist’ useful for keeping your content as original and human-sounding as possible.
6. Common AI words
It’s pretty much a given that you’re reading a ChatGPT-written article if it contains the chatbot’s favorite words.
“Elevate” and “delve” are perhaps the worst culprits, with the former often appearing in titles, headings and subheadings. If you use AI for your own content production, then make sure to blacklist these words along with the common phrases.
7. Formatting
The final giveaway that a piece of content is a copy & paste job from a generative AI tool is in the formatting.
If you ask ChatGPT to write a guide to SEO, it will inevitably include a list that looks something like this:
Each item is often highlighted in bold, and then ChatGPT likes to throw in a colon to expand upon each point.
There’s nothing inherently wrong with presenting lists in this fashion, but it has become ChatGPT’s signature style and therefore easily identified as AI content.
Sometimes ChatGPT does too much expanding and the list becomes an ugly wall of text — If you come across such poor formatting then I would suggest the person publishing it isn’t even trying to hide the fact they are using AI.
Final thoughts
The fact that all these AI detection tools claim to be highly sophisticated and trained on the patterns of human writing means nothing if they can’t actually determine with pinpoint accuracy whether an article was written by ChatGPT or not.
But that doesn’t matter, because it’s possible to spot AI content manually, so long as you know what to look for in terms of the content and formatting.