It shouldn’t surprise anyone that we’re seeing more and more AI-generated written content across the internet as more and more genAI platforms launch and older models upgrade. But it may be shocking to learn how quickly the rate of AI content is being added.
According to research reported by AI detection company Copyleaks earlier this year, the rate of AI content growth since ChatGPT 3.5 launched in November 2022 through March 2024 was a stunning 8,362%. Granted, the starting base was very low but during that time frame the numbers are telling. The findings show that:
- The amount of AI-produced content online increased 187% from November 2022 through January 2023.
- During the next 12 months, the amount of AI-written copy rocketed 2,848% on the interwebs.
- While a McKinsey study from March found that the use of genAI jumped to 65% of respondents for 2024 up from 33% in 2023.
And there are no signs of genAI use slowing down. Now more than ever it’s important to be able to distinguish between AI content and human-produced content for the sake of digital integrity.
Here are five ways to tell if written content came from an AI bot or not.
Language Patterns
You can frequently see in AI-generated text that it lacks emotional subtlety and tends to use overly formal or complex words. So the sentences sound stiff or stodgy without the pip and punch of quality human writing.
AI sentences also don’t vary in length. Perhaps because of their training or programming, their sentences are eerily similar in word count. They can read very monotone and drone on.
Consistency Issues
AI also seems to struggle a bit with consistency when it comes to narrative details, especially when it tries to write fiction. You may notice it has abrupt changes in the description of a setting where a scene is occurring without explanation or details of a character’s appearance without warning.
Or there might inconsistencies that don’t align with previously stated facts along the story arc showcasing AI’s limited contextual understanding at times. Regardless, it’s one to watch out for.
AI Tool Detection Recommendations
Copyleaks was mentioned at the beginning of this write-up, and it claims on its website to be the most accurate AI detector based on multiple independent studies. One feature that makes it stand out from other solutions is that it works in 30 different languages.
Another widely used AI detector for writing is GPTZero. It’s very effective at analyzing text patterns, overuse of phrases and common word combos that GPTs tend to cycle through. It offers a free version up to a certain character limit, with more features for the paid options.
Below are two basic scans of an uploaded submission today written by a human from back in 2021 before genAI was around–the screen correctly labeled it as human produced.
The second image was a ChatGPT-written poem uploaded today that was also accurately detected. Both Copyleaks and GPTZero are useful and it’s a best practice to use more than one.
Depth and Context
Another key giveaway in AI writing can be lack of detail, vagueness or over generalizations. For some reason, when AI models write they tend to avoid deep contextual issues or robust topical understanding.
The current generation of these generative pre-trained transformers skirts around complex subjects that require personal experience or expert knowledge. Provided explanations might be delivered confidently, but without supporting details to validly back up the point.
Unusual Language Errors
While infrequent, AI-generated texts sometimes include odd errors or turns of phrase that are awkwardly out of place. One example I personally saw was an AI sample article about climate change where trees were once referenced as “Arbor Day notables.”
Those kinds of unusual combinations of common phrases, malapropisms or basic grammar errors – such as the recent dustup over ChatGPT maintaining that there are only two letter “r”s in strawberry – are useful tip offs that the text was AI derived.
It’s clear we’re in the early pages of the chronological book about genAI. Since Artificial Intelligence isn’t going anywhere — except to new levels of accelerated adoption — it’s our responsibility to keep up with the latest developments in both AI generation and detection so that AI isn’t left to write its history on its own.
Read the full article here
