How to Detect AI Content: A Guide for Skeptics

According to Google Trends, the search volume for AI detectors is soaring and is at its highest.

Good, right? Well, not really.

As this search volume grows, more and more questionable AI detectors are popping up like mushrooms after a rain shower.

However, most of them are just quick money grabs trying to lure you into buying some other service, giving AI detectors a bad name overall.

This poor reputation and general dissatisfaction with AI detectors is very visible on social media, where people often claim that some questionable tool flagged their original content, which they wrote in high school, as AI-generated.

We as humans like to make quick conclusions, and thus there is a large population of people who simply claim that AI detectors don’t work.

Well, life is not as black and white, and so we’ve decided to create a guide on how to detect AI-generated content, specifically aimed at skeptics.

Disclaimer: This guide is intended for detecting content written in English.

Step 1: Choosing the Right Tool

So, as with everything, you need to have the right tool for the job at hand.

Let’s assume you are a surgeon. A rusty drill bit that you found in your driveway is a great tool to put up pictures at home, but you won’t use it on your patients for obvious reasons – this is common sense.

The same goes with AI detectors – a tool that you find online that calls itself an AI detector can be just that – neglected, badly developed, and completely unfit for the task at hand.

Then why do these tools exist, you might ask?

Just to lure you into using some other service or product they are offering. On the internet, traffic is king. If you use your high-authority domain to offer free tools, they can rank well on Google, and it is free marketing for the website.

Where can you find tools you can trust then?

In our directory, of course. Just filter the tools as you like, and look at the highest-rated ones. While our directory isn’t perfect, it will get you on the right path, and it will minimize the risk that you will land upon a bad tool.

Unfortunately, in our experience, paid tools tend to work better – as with anything in life, you get what you pay for. Luckily, most of the tools have a freemium or free trial option, so you can still try them.

Step 2: Analyzing Text Using an AI Detector

You’ve got yourself a good tool from our directory? You’ve read the description and understand where it could potentially fall short? Great. You’re ahead of 99% of people tackling this task.

Now, let’s assume that you have to evaluate text submissions from 4 different people. You run the analysis with your awesome new tool and the results are as follows:

• Text 1: 100% human
• Text 2: 100% human
• Text 3: 90% AI
• Text 4: 40% AI

As we can see, the hypothetical detector thinks with high confidence that texts 1 and 2 are free of AI. Nice – good job by these people. Nothing more to do here.

For texts 3 and 4, we want to move onto the next step of checking – we want to use our own logic and judgment to find out what’s going on with these submissions.

You might ask – wait, what? I just paid for an AI detector, and now I have to manually check?

Well, yes. Think of the detector as a tire pressure monitor on your car – it indicates if there is an error, but you have to confirm if it’s really an issue or not.

Step 3: Manually Checking Text for AI

You start by checking Text 3 (the one which was flagged as 90% AI). By just looking at it, you can tell it is completely off:

• It’s missing deeper analysis or insights – A classic AI move, it sounds like a politician who likes to generalize everything.
• The text repeats itself – You notice the same phrases or words popping up way too much; no way anyone taking pride in their work writes like this.
• The facts are off – For example, the text says that an event is happening in March, but you know for a fact that it’s actually in June. It’s odd to get such a basic fact so wrong.
• The overall tone is very cautious – Do you find many words like “may,” “should,” “appears”? AI – where is your confidence?

You confront the person who submitted Text 3, and after a while, they admit that they just copied it from Claude 3.5 Sonnet. Busted.

But what about Text 4? 40% AI? Not that much? Agreed. This is the complex one to analyze.

However, there are scenarios where this can happen – the writer may be more tech-savvy and have used a humanizer tool to disguise the AI output and make it appear more engaging and unique.

So what tends to happen? Good AI detectors will still flag that something sketchy is going on. However, they aren’t as certain.

This is where you come into play – in order to fool a great AI detector, the humanizer option is more likely than not set to “aggressive,” making the text awkward to read – you ask yourself, “Why would anyone write like this?”

The person who submitted it does not have any good explanations for why they wrote this way. If you look at any texts they’ve previously written, the style is completely different. They are simply busted.

Final Thoughts

Can somebody be falsely flagged for using AI, even if they didn’t? With professional tools, this is pretty uncommon (but we won’t say it can’t happen).

However, this is where your own judgment and manual checking comes into play. AI detectors help you when you have bulk content and make it easy to flag suspicious cases, but when concerns arise, the final decision for critical texts should always be human.

Still don’t believe that an AI detector can handle most of the checking?

For those individuals we suggest giving the AI content detector by Pangram Labs a go – experiment with it and see for yourself how effective a professional tool can be.

Good luck!