Skip to Main Content
     

Guide to: AI in Scholarship

Strategies for Identifying AI Generated Content

Identifying AI Generated Content

AI generated content like images, videos, or text, is not always easy to identify. Here are some resources to help you determine what you are interacting with and whether or not it is genuine or AI produced. 

You can also determine AI generated text pretty easily by careful reading. Keep these things in mind when looking to see if a text is AI generated: 

  • Inconsistencies and repetition: Occasionally, AI produces nonsensical or odd sentences which can be a clear indicator of AI-generated text. Abrupt shifts in tone, style, or topic can point to AI that is struggling to maintain coherent ideas. Whereas humans often vary structure to create a better flow, repeated phrases or sentence structures can point to AI relying more on memorized patterns. Occasionally, AI produces nonsensical or odd sentences which can be a clear indicator of AI-generated text. 

  • Context and content: If the text seems to be unable to grasp the larger context of the writing, is missing the point entirely, or references specific details without appropriate context, it could be AI because these models lack real-world understanding and nuance. Uniqueness and personalization are hallmarks of human writing, so if the text feels basic and formulaic it might be AI-generated. And obviously, if there are blatant falsehoods or unverifiable facts, it’s either an unreliable source or AI. 

  • Buzzwords and grammar: Excessive use of buzzwords and jargon can be a sign of AI filling the gaps of its knowledge with generic vocabulary. This is particularly common in marketing text and instances where the larger context is more complex. And watch for perfect grammar, as even the best writers may make mistakes or intentionally break grammatical rules to emphasize an idea. 

  • Quick responses: As people often take time to think and respond, rapid replies could be a sign of AI. This is common in customer service platforms like chatbots. 

  • Placeholder text: Some AI text generation models use placeholders like "Insert name here" if it lacks sufficient information to fill in the gaps itself. The appearance of these placeholders in text used outside the model can be an indicator of automated content.  

  • Lack of citations: Platforms like ChatGPT can insert citations but they have been found to be unreliable, incorrectly linked, and improperly formatted. Bodies of text with these types of citations may indicate AI was used. 

Source for this text linked below.

Identifying AI Generated Content

AI generated content like images, videos, or text, is not always easy to identify. Here are some resources to help you determine what you are interacting with and whether or not it is genuine or AI produced. 

Reverse Image Search

If you suspect an image to be AI-generated, try a reverse image search to see if you can find the original source of the image. Reverse image searches can be conducted using tools like...

In Microsoft Edge and Google Chrome, you can also search an image found on the web by right-clicking and then selecting "Search image on the web" or "Search image with Google," respectively. 

Note that many of these tolls will not show you image results in chronological order. You may need to scroll through several pages of results to find the earliest appearance of the image online. 

Detecting Altered Images

You can use free analysis tools such as the ones below to detect digital altering or otherwise 'faked' images (though note that these are imperfect tools!).

Your own Observation!

But don't underestimate the value of your own observation skills! Silly as it sounds, checking for details that don't make sense (including: missing fingers, body proportions that aren't quite right, blurring that is visible only upon zooming in, nonsense text in signs etc.) is another way that you can catch artificially-generated images. 

This can also be done in combination with the reverse image searching strategies above -- for example, try conducting a reverse image search of only parts of the image, perhaps images showing distinct natural or architectural features, to see if they exist and if they match the location the image is 'supposed' to represent. 

The suggestions in this box are credited to Twitter/X user @Shayan86 and his useful threads: 

And to: 

Detecting AI Generated Content