How To Check for AI-Generated Content (AI Content Detectors and Manual Hacks)

AI content has divided publishers into two camps. One group is pro-AI content (as long as it’s edited and fact-checked), and the other is anti-AI content.

Hold on; I missed something.

Oh, right. There’s a third group.

The one-click publishers. The ones who use AI-generated content with zero edits. These pandemonium publishers belong in the eighth circle of hell.

Okay, that got a bit dark. Anywaaayysss…

Whatever your stance is on AI content, I think we can all agree that no one should be forced to do something they don’t want to. Despite explicitly asking, many writers don’t disclose that they use artificial intelligence to write content. This sucks for publishers.

My argument for writers revealing AI use is simple: it allows publishers to set up additional steps (if they deem it necessary). Plus, it’s the right thing to do.

So in that spirit, let’s look at ways to level up your bot-spotting skills.

How do AI content detectors work

AI content detectors use a combination of machine-learning algorithms and natural language processing to distinguish between human-written and machine-generated text. The specific mechanism behind each AI content detector can vary depending on the technology and approach used, but in general, they involve:

1. Data collection: A large dataset of human-written and AI-generated text is used to train the system.
2. Feature extraction: In this step, the system extracts important features from the data, such as word frequency, n-grams, the presence of certain words, phrases, grammar, and the complexity and structure of the text. This process is done to understand the text and identify patterns in the data.
3. Model training: In this step, supervised learning techniques are used to train the system to use labeled data and extracted features to identify AI-generated text.
4. Inference: Presented with a new text, the trained model analyzes it for linguistic and syntactic patterns observed in the training data to establish whether it was written by humans or machines.
5. Continuous learning: The system continues to learn from new data and feedback, allowing it to improve over time.

Best AI content detection tools

With AI-generated content taking over the web, tools to detect it are also popping up equally fast. Here are some of the well know AI text detectors.

Originality.ai

This AI super sleuth was built for mass publishers and content agencies. It comes with AI content detection tools and plagiarism tools to ensure your content is original. They claim a 94% accuracy rate, but I found some false positives during my testing. It also has some nifty features such as team management, full site scanning, and user-specific scan history in the works.

Huggingface OpenAI detector

Hugging face Open AI detector was one of the first AI content detectors. Although a pioneer, the tool isn’t updated to detect text generated by newer AI models like GPT3 and GPT3.5. The GPTrue or False Chrome extension makes use of the Huggingface OpenAI detector and brings it right to your browser. Since the extension runs on the same technology as the Hugging face detector, it won’t be able to detect the newer AI models.

KazanSEO AI GPT3 detector

AI content detector tool is just one of the many free tools Kazan SEO offers. It works with individual URLs as well as multiple URLs. All you have to do is add all the URLs you want to inspect and run the tool.

Writer.com AI content detector

Writer is an AI content platform designed for editorial teams. Writer released a free AI content detection tool to help publishers spot AI-generated content. With customizable style guides, a bespoke AI writer trained to write in your brand voice, and robust reporting, Writer is an excellent choice for businesses looking to boost their productivity without losing brand consistency. The AI detection tool is just an extra perk.

Grover AI detector

Grover was developed to combat those pesky AI-generated fake news. It can spot AI fingerprints in machine-written text. Grover claims over 92% accuracy in distinguishing human-written stories from computer-generated junk.

Unfluff

With Unfluff, you don’t just get AI content detection as with other tools on this list. Staying true to its name, Unfluff highlights all the superfluous text in your content. I love this tool because you’re forced to cut out those extra sentences you desperately want to keep to sound clever.

The good ol’ gut instinct

When the bots and bytes can’t help, a good old-fashioned gut check might be what you need to spot AI-generated text. You can usually spot AI content with a quick read because it leaves clues. Here are some hints to look out for:

Dry text overloaded with facts

AI content is often dry and factual. It usually sounds like trigonometry married history and had a baby, and the baby sold its soul to a robot who converted it into binary code and blogged about it.

It won’t have interjections or personable analogies (or hyperbolic metaphors about dull writing). Keep in mind that formal and academic writing doesn’t use interjections, so the lack of interjections doesn’t indicate AI content. But lifeless prose does.

So next time your writer hands in a piece of content overloaded with facts but lacks any soul or opinion, you might want to dig a little deeper. It’s probably AI content, or you just hired a writer you can replace with an AI generator.

Repetitive structure

Most AI writers tend to repeat the same structure throughout the article, especially when no human is in the loop to break up this pattern. This formulaic writing style makes for boring, predictable content. Here’s an illustration:

It’s easy to spot AI-generated content when it follows a repetitive pattern like this.

Check for Inconsistencies

Most AI writers are inconsistent with facts, timelines, and logic. Here’s a text I got from a leading AI writing tool (who shall remain nameless):

“Alice was born in 1980. She graduated college in 1985 and got her first job in 1995.”

This clearly doesn’t add up, as it would mean Alice graduated college at five and got her first job at 15. Even Sheldon from The Big Bang Theory would know that’s impossible. These small indications can help you identify AI-generated content.

Another dead giveaway is inconsistent pronouns. Most AI writers struggle to keep track of pronoun references. The text is probably auto-generated if you see a person referred to as a “he” in one sentence and a “she” in another.

Likewise, if there are sudden changes in tense or point of view, that’s another clue that a human didn’t write the piece. Irresponsible publishers don’t check AI-generated content for these details. Of course, these aren’t foolproof indicators, but they’re red flags worth investigating further. 

Look for Poor Grammar

Grammatical mistakes were surprising because grammar is rule-based, and AI algorithms are usually great at following logic — even better than humans, in most cases. But, for whatever reason, most AI text I’ve come across had typos and grammatical mistakes. It’s not like humans are perfect at grammar, but it’s still a good indicator that the content was machine-generated.

Stiff syntax and awkward word choices

Word choices are nuanced, and AI algorithms aren’t always the best at picking up subtle differences in language. So, if you see odd phrasing or words that feel out of place, there’s a good chance a certain cyborgy friend was involved in creating the text.

Look for new angles or original ideas

AI tools can’t think logically or critically, so they can’t reason or create original concepts. Any machine-generated text — without a human in the loop — will lack depth and perspective. It will most likely be a rehashed version of existing material on the web.

Unnatural transitions between paragraphs

Many AI writers that are commercially available today rely on templates to generate their content and have limited contextual understanding. The lack of context often leads to abrupt transitions between paragraphs. Do you notice a sudden shift in the tone or flow of the text? If you do, then you might be looking at AI-generated content.

Lack of personal experiences

Since our bot friends don’t traverse the world of humans (for now), they cannot draw on personal experiences and analysis. This often creates flat, uninspiring content without any personal touch.

Do you have any other tricks up your sleeve to sniff out AI content? Connect with me on LinkedIn and share your tips.