How to Tell if Content Is AI-Generated

It’s no secret at this point that artificial intelligence has changed the way content is created. A few lines of instructions can lead to anything from a blog post and an essay to a news report or a social media caption within mere seconds. The global trend initiated by ChatGPT has grown so much that AI-generated text can sound extremely polished and very natural.

While all that is good on the one hand, it begs the question of how anyone can tell if the content was written by an actual person or by using artificial intelligence models. The problem gets even more complicated when you consider how AI-generated writing can look so different from one tool to another. Many people resort to AI detection tools to check text for AI-generated patterns, but even the best tools can make mistakes and misidentify manually written content as artificial.

For this reason, you must know the ways you can identify content generated by AI tools like ChatGPT.

Start With the Simplest Step: Reading Carefully

Oftentimes, all you need to do is read the content carefully to notice whether something feels off or too robotic. Generative AI usually leads to content that is too polished and smooth. Some of the points you might notice include:

• Grammatically clean sentences
• Perfectly connecting paragraphs
• Balanced arguments, with no leaning to one side
• Transitions where you’d expect them

This is not to say that a piece of writing by a human cannot have these qualities. It’s more about the frequency and combination of certain elements that make a text read like it’s artificial. Looking for these clues may not be enough to confirm if a text is artificially generated, but it gives you a basis on which you can then lean on more advanced testing methods. Learning how AI detectors identify these patterns is helpful, as they often look for a clue known as sentence rhythm.

Sentence Rhythm Helps With Detection

Human writing normally comes with varying sentence lengths. One short sentence may be followed by a longer explanation, or the writer may decide to use an unusual expression or an awkward construction. On the flip side, an AI writer’s large language model may lead to uniform sentences that all seem to have similar grammatical patterns. A good way to check this is to read the text aloud. Does it sound monotonous or repetitive? Does each new paragraph seem to reach its conclusion in the same predictable way?

You may also do a little experiment to check whether AI-generated content is actually constructed this way. Write a short essay and try to make every sentence a similar length. Run it through AI checkers and see if it gets flagged. This will also let you see how GPTZero handles AI detection, or any other tool, for that matter, and help you spot AI writing yourself.

Is It Too Structured?

All major AI systems, including free AI text generators, are very good at organizing information into neat little packages. Ask AI models like ChatGPT or Claude to explain any subject, and the result may be categorized into a clear introduction, several labelled sections, bullet points, and a conclusion. While this is helpful, it’s also quite repetitive, becoming an indicator of artificially generated content.

Check Facts and Sources with AI Checkers

Another crucial step in determining if a piece of text is written by AI is to double-check facts. There are many tools out there that will verify claims in AI-generated content, so you don’t have to do it manually. A lot of AI-generated texts end up having factually incorrect information, despite sounding extremely confident. It is, therefore, very important to check for unverifiable facts and unreliable citations to identify AI-generated work.

You don’t have to do this manually either. Simply upload the content into the relevant tool and ask it to check the sources or give you the sources for the statements presented in the text. Then, double-check manually whether the provided links contain data that’s included in the original content. If you do decide to check AI content for unsupported claims and find multiple ones right away, that is a very good sign that the text might be AI-generated.

Notice Contextual Mistakes

To spot AI-generated content, you should also look for contextual mistakes. It’s true that AI-written content has gotten very impressive, with fewer mistakes and more natural-looking results. But even the best tools sometimes struggle with context. While the description of the main subject may be accurate, the surrounding circumstances might have issues worth noting.

Some examples of this include:

Some of these mistakes may be hard to notice because of how natural they sound. If you wish to test the limits of AI detector accuracy, give it a piece of text like this and ask it to detect AI. Many AI content detection tools can fail here, which is why human intervention and understanding of artificial writing patterns are crucial to determine if a text was generated or manually written.

Look for Personality and Tone

Another sign that a text was written with AI is a lack of personality. This is one of the biggest differences between AI and human writing, as AI struggles to achieve the kind of messy writing humans naturally gravitate towards. People have preferences, so they are enthusiastic about one subject while bored with another. They often make jokes or show their dislike for things. They rely on local expressions that AI can’t reproduce quite the same way, although some models are becoming more believable.

Writing patterns of AI are often neutral. This results in emotionally flat writing, or something that is exceedingly diplomatic. For news stories, AI software is found to showcase stronger positivity than humans. On the contrary, soft news written by AI conveys sadness more often. Some major noticeable differences in human and AI-generated pieces may look like this:

Personality is not always present as much in human writing, either, so you must be careful in your assessment. Some writers maintain a neutral style deliberately, especially in academic writing, corporate communications, etc. There’s a chance your free AI checker for ChatGPT will fail to notice the difference here, too. Just see how Grammarly’s AI detector performs to get an idea of what it actually flags and what goes amiss. With the very different writing styles from person to person, there’s a good chance of getting a false positive now and then.

Investigate Editing History

Perhaps the simplest way to detect AI outputs is to check how the document was produced, instead of focusing only on the final product. An essay written by a human develops slowly, with notes and unfinished thoughts that are deleted and rewritten.

If you check a document’s history and notice that some 1,000+ words suddenly appeared out of nowhere, there is a chance that artificial intelligence was used to write the document. It’s worth noting that this doesn’t necessarily mean that the text is AI-generated, since the document could have been written somewhere else and then pasted into the final file. However, a combination of version history, notes, and evidence gathered from the writing process itself can be used to detect AI-generated content.

Ask the Writer

Of course, asking the writer if their text uses artificial intelligence isn’t very helpful, but asking them to explain their argument or thesis can be very revealing. This works for academic work more than anything else. A professor, for instance, can ask their student to explain why they selected a particular source, or how they reached their conclusion.

Work with an AI Content Detector

The phrase, “If you can’t beat them, join them,” fits very well here, because you can use AI to identify AI. There are plenty of reliable AI detector tools out there that can give you a good starting point to find AI-written content. AI detection software options may look for unusual readability patterns, false information, fake sources, grammatical signs, and much more to tell if content is created using AI or not. This does vary, however, based on which detector you’re using, as some focus more on linguistic or statistical patterns.

These tools must be used with care, though, as an AI detector alone isn’t enough to truly judge someone’s work. An AI detector is designed based on certain limitations and biases, which means if it encounters a human-written text that demonstrates patterns associated with different tools, it’ll give you a false positive result. Here are some scenarios in which a human-written text may be flagged for AI:

• If the writing is very formal, using a rigid structure, or very polished and devoid of personality.
• If the text is written by a non-native writer whose language patterns differ from what the detector is built to identify.
• If the writer has used short, repetitive sentence lengths.

This is why you should be responsible with AI yourself, as identifying AI assistance in a text goes far beyond AI detection using software alone.

What About Images and Videos Generated With AI?

While it was pretty easy to spot images generated with AI only a little while ago, thanks to unrealistic hands and facial features, incoherent text, or famous logos that looked strangely distorted, the reality now is very different. Generative AI has improved to the point where it’s nearly impossible for anyone to tell whether an image or a video is real or fake. There’s no singular way anymore to identify artificially created images, but the following few ways may help.

Look for Perfection

Some images just look too perfect the moment you see them. This could mean a person looking too polished for their given situation, or someone’s skin that looks airbrushed when it shouldn’t, or the subject just looking out of place in their surroundings.

Check for Physical Anomalies

Images generated by artificial intelligence may have all the right visual elements in a photograph, but inaccurate shadows, weird reflections, or a physically impossible or inconsistent perspective. These are all tells that the software created the image as opposed to a camera.

Look at the Metadata

The metadata of any image shows various properties, including what gear was used to create it as well as the date and time it was produced. If you see a time stamp that doesn’t match the photograph’s context or story, or if the camera information is missing, the image could be AI-generated.

Listen to the Audio

In the case of videos, a good way to check for authenticity is to listen carefully to the audio. AI has become very strong at imitating anyone’s voice, but gaps in emotional authenticity and natural pacing can still indicate artificial generation. Listen for the lack of breathing, missing background noise, and a robotic accent or inflection for more clarity on a video’s origin.

Analyze the Context

Does the image show plants that shouldn’t really exist in the area that’s depicted? Are there any vehicles in the scene that don’t belong in the time period or the location? Is the architecture accurate to the place? These are some things you can check within the context of an image, since AI sometimes misses out on perfecting contextual details even if it creates a picture that looks real in isolation.

Trust Your Gut

The bottom line is that there is no definitive solution to tell whether an image is real or not. This is where your gut feeling comes in. If something looks too perfect or doesn’t quite make sense in terms of the physics or the context, you could run some further tests to develop a better understanding of how a visual was created. Without any perfect solutions to help you, it’s crucial to combine your gut feeling with tools that are designed for deeper visual analysis.

Conclusion

So, how do we tell if someone’s work is made with the help of AI? It’s not a straightforward answer, unfortunately. With rapidly evolving, advanced AI models doing more than one could have imagined, it’s getting increasingly difficult to tell what’s real and what’s not. But using a combination of manual verification and AI technology itself, you can get a good idea of the level of AI influence in a piece of content that’s been sent to you. It’s important to remember that the current methods available to us cannot say for sure whether or how much AI is used to create a piece of content. You can only make an informed guess based on your own research and testing.