Did AI Write This? 4 Ways to Check if Text Was Generated

Almost all of us have recently read something that felt so robotic, soulless, or fake, that it left us with this question: did AI write this? As AI tools like ChatGPT become more ubiquitous, the line between human and machine writing will continue to blur.

Even if you have a “gut feeling” that something was AI-generated, academic integrity and/or hiring decisions may be on the line. Your gut feeling that a piece of writing was AI-generated could be wrong, and if that is the case, someone might be harmed as a result of that gut feeling

We will cover four proven methods to determine if AI wrote a particular piece of text. These four methods range from manual linguistic analysis to using advanced AI detection software that has been tailored for today’s AI technology.

1. Analyze Sentence Structure

Looking for sentences with a uniform length and structure can reveal if a piece of text was AI-generated. Many AI models generate sentences that feature a uniform length and structure. But, most human writing features burstiness: a mixture of short, punchy sentences and long, complex sentences. If a piece is too uniform, it might be AI.

Unlike human writers, AI rarely makes grammatical errors or typos. Because of this, one of the best ways of spotting AI writing is to see if a text features grammatically flawless sentences, but feels monotonous or overly formal. This can suggest that a piece of content was AI-generated.

Many AI writing patterns feature sentences filled with far more words than is necessary to convey a piece of information. Such as using three words when one word would be just fine. If you read a piece that features plenty of extraneous words that fill out a sentence but fail to expand on a particular point, that piece might be AI-generated.

2. Spot the "AI Phrases" (Vocabulary Tells)

You can spot the “AI phrases” to tell if a text was AI-generated. Large Language Models are trained to be helpful and polite. As a result of this training, it is not uncommon for written content, produced by these models, to overuse specific transition words and empty adjectives that often appear in their training data.

To check for AI text, you may want to look for these AI phrases:

• Underscore.
• Delve.
• Tapestry.
• Pivotal.
• Harness.
• Illuminate
• Testament.
• Landscape.
• In conclusion.

If a piece of writing heavily features any of these phrases, there is a chance it may have been AI-generated.

On top of being helpful and polite, AI models often default to a neutral, overly diplomatic tone. Phrases like “It is important to note” and “On the other hand” are featured frequently in order to avoid taking a hard stance. Keep an eye out for these phrases, too. Writing that heavily relies on them may have been AI-generated.

3. Use Tools to Detect AI with Accuracy

The most reliable method for how to tell if text is AI-generated is to use a dedicated AI detector like Pangram. Even though your intuition is useful, it is also subjective. Tools like Pangram are objective.

Pangram identifies statistical patterns that are invisible to the human eye. By identifying these patterns, Pangram can determine whether or not a piece of content was produced by AI.

There are many different AI detection tools available to you. Many of these tools are less-than-effective, though, because they look for perplexity - how predictable a piece of text is - to determine if a piece of writing was produced by AI. This has led to works such as the Declaration of Independence being flagged as AI-generated.

Instead of searching for perplexity or other signs of ChatGPT, AI detectors like Pangram use deep learning to find specific syntax patterns. This lets them accurately determine whether or not a piece of writing was AI-generated.

The best AI writing detectors now have a 99.98% accuracy rate and near-zero false positives. Pangram is one of these tools. In contrast, free AI detection tools are often unreliable and prone to false positives.

4. Distinguish the Nuance of "Assisted" vs "Generated"

You can distinguish between “AI-assisted content” and “AI-generated” content to tell if a text was generated with AI. This is because the question “Did AI write this?” isn’t always binary. For example, someone can write the core ideas that comprise a piece of content and then use AI to edit this content.

You can use advanced analysis tools, such as Pangram, to reveal if a human wrote the core ideas within a piece and used AI to edit, or if the text was fully machine-generated.

AI writing exists on a spectrum:

• Fully AI-generated writing is writing that was completely generated with AI; this includes the core ideas, the structure of the piece, and the content, along with anything else the piece encompasses.
• AI-assisted writing is writing that was initially written by a person and then polished using AI.

Try an AI segment analysis:
AI Detector Tool