How to Check AI Written Text: Tips, Examples, Insights

5min.

Comments:0

How to Check AI Written Text: Tips, Examples, Insightsd-tags
02 April 2024
Using AI to generate content can lead to problems if the texts mislead readers or are riddled with linguistic and syntactic errors. Even though various tools are said to detect AI-generated content, they aren’t still foolproof. This means that if you’re looking for some ideas on how to check if the text was written by AI, try to recognize distinctive traits: repetitions, redundancies, peculiar word structures, and inaccurate information.

5min.

Comments:0

First things first, using generative artificial intelligence to create practical content isn’t considered negatively, which is stated clearly in Google’s guidelines on AI-generated content. From this article, you can also learn that when creating articles for your users, the primary goal should be to provide reliable and useful information that adheres to the E-E-A-T standard. This doesn’t exclude helping yourself out by using AI in your content marketing strategies.

However, issues arise when content is misleading or qualifies as spam. This includes content that is unclear, contains false information resulting from AI hallucinations, or fails to add value for the target audience.

Similarly, it’s quite problematic when the pieces of content you receive from an agency are written by AI, despite an agreement specifying it would be human-written. Just imagine ordering a couple of blog posts, and being told a real person would write it. Yet, what you get looks weird and resembles some robo-talk. It would be wise to know how to check if an article was written by AI, wouldn’t it?

But how can you determine if a given text was written by AI? Read on to learn how to identify AI-generated content.

Tools to Help You Check if a Text Was Written by AI

The sudden advancement of generative AI has spurred the creation of tools specifically designed to help us identify and/or check if a text was written by AI. Below, you’ll find a short list of tools that can help you find out whether a text you look at was generated by AI. But first, let’s see how those tools operate.

Was the Text Written by AI? This Is How AI Detection Tools Work

Tools for detecting AI-generated content focus on textual differences. They analyze the structure and style of a copy to pinpoint characteristics that might suggest the text was written by AI.

Generally speaking, AI detection tools use advanced algorithms and language models to evaluate how natural a copy sounds. This involves examining a couple of elements such as:

  • the complexity of sentence structures,
  • repetition of phrases, and
  • the use of vocabulary that diverges from typical human writing.

Additionally, these tools can identify patterns that are commonly used by AI, effectively revealing the so-called “digital fingerprint” of popular language models.

3 Examples of AI Detection Tools

Below you will find 3 examples of the most popular tools to detect AI-generated texts:

  1. Copyleaks is a tool that comprises modules for detecting content created by AI, particularly noting ChatGPT and Gemini. It’s also equipped to identify plagiarism, including paraphrasing and source code plagiarism. As said by the designers of Copyleaks, this tool fights AI with its own weapons, using advanced artificial intelligence algorithms to analyze text patterns characteristic of language models.
  2. GPTZero evaluates text syntax and identifies potential logical flaws. It bases its analysis on the so-called perplexity score, which focuses on checking how random are the structures used in a given text (a lower level of randomness suggests a higher probability that the content was AI-generated), alongside measuring deviations from this randomness parameter.
  3. Smodin AI Content Detector is a free, machine learning-based tool that scans text for stylistic elements and keywords typical of AI-generated content.

Do AI Detectors Work as Intended?

Although tools designed to determine whether text was generated by AI are usually good at their job – sometimes with an impressive level of accuracy – they are still far from being infallible. Their precision of assessment varies significantly between different models and is influenced by numerous factors, such as the type and style of the text. That’s why, there is always a fat chance of getting both false positive and false negative results.

Let’s say that again: The above-mentioned tools, despite their huge popularity, are not always correct. For example, in tests, Copyleaks has incorrectly classified paraphrases created by AI (specifically by QuillBot) as human-generated content. Despite relying on extensive databases, no AI-detection tool is completely error-free. Hence, these detectors should be considered as just one mean of a comprehensive verification process, rather than being relied upon solely.

Moreover, it’s worth keeping in mind that the underlying technology of language models and generative AI is continuously evolving. This rapid development means that even with regular updates, not all tools may keep pace with the swift advancements in this field.

How to Check If the Text Was Written by ChatGPT: Look for Characteristic Phrases and Errors

Tools aside, AI models, particularly ChatGPT, generate texts based on identifiable patterns. By understanding these patterns, we can tell if a text was written by AI or a human.

A notable example is the use of repetitive phrases or sentences. When tasked with writing about a specific topic, ChatGPT tends to treat it as a keyword, frequently inserting the exact phrase throughout the text without varying the language with synonyms.

Just to show you how this looks in action – take a look at the below screenshot. ChatGPT was asked to generate a text about alarm systems for commercial buildings. One phrase – alarm systems – appears 6 times in the first paragraphs.

screenshot of chatgpt repeating one phrase

source: ChatGPT

Another thing is that ChatGPT often overuses the same set of phrases for introductions and conclusions. Some of those include: “in the dynamic world of,” “let’s delve in,” “vibrant landscape”, “navigate”, or “ever-evolving world” – one of the examples is easily visible in search results:

chatGPT overusing phrases in the introduction Google search results

source: Google search results

Also, overuse of phrases such as ‘landscape,’ ‘supercharge,’ ‘unlock,’ ‘cutting edge,’ ‘unleash,’ and ‘leverage’ often indicate that AI has generated a text.

Grammatical, stylistic, and syntactic errors are also common, especially when generating texts in other languages than English. Why is that? AI tends to translate English sentence structures into a target language, creating loan translation and other linguistic mishaps. These are immediately apparent at first glance – if you use ChatGPT, you’ll certainly notice that some texts or their sections simply sound unnatural.

Moreover, AI can also write either in an excessively flowery and strikingly serious manner, resulting in a tone that doesn’t go well with the context. In most cases, you will be able to sense something is off with a text.

Finally, the biggest issue with content created by artificial intelligence is what’s known as hallucinations – completely untrue information made up by the AI. This can seriously mislead users. You may come across some instances online showing how creative AI can be.

chatgpt hallucination example

source: sify.com

Tips for Checking If a Text Was Written by AI

Even if a text seems to be written by a human, it’s always good to be a bit cautious and skeptical. Actually, there are four things you can take a closer look at to verify if the content you’re engaging with is man-made.

TIP #1: First, review the sources of information if they’re provided. Are they credible, coming from an authoritative and reputable author, or are they rather unknown or of questionable origin?

TIP #2: Next, compare the text with other articles on the same topic. Are the details consistent with what other sources say, or are there discrepancies?

TIP #3: Then, check the publication date and relevance of the information. Is it outdated, or does it still reflect the current state of knowledge in the field?

TIP #4: Lastly, examine the style and language of the text. Is it clear and logically sound? Does the vocabulary and tone fit the topic and the style of the content?

Summing up: How to Check AI Written Text

While generative AI based on language models is extremely useful in various fields – take, for example, its application in Google Ads or using ChatGPT for keyword research – it is still fairly limited and happens to be unreliable.

So, if you feel like verifying whether a text was written by AI, here are some steps you can take:

  1. Use a few AI detectors so that it will be easier to determine if the text is generated by a language model.
  2. Look for weird-sounding repetitions that focus excessively on the main theme of the content.
  3. Examine the beginning and end of the text for structures that are typical of AI-generated content.
  4. Assess whether the style and vocabulary of the text are appropriate for the context.
  5. Verify the information so that you can be sure it comes from credible sources and isn’t the result of AI’s hallucinations.

Concerned about your website being filled with low-quality AI-generated content? Strive to publish only high-quality and valuable marketing content? Reach out to us! We will map out a fully customized content plan and later produce engaging articles based on proven SEO copywriting standards. This way you can be sure your website content is human-friendly and high-ranking.

Author
Maciej Jakubiec
Author
Maciej Jakubiec

Junior SEO Specialist

A marketing graduate with a specialization in e-commerce from the Cracow University of Economics, hailing from the picturesque Podkarpacie region. He joined Delante in 2022. In SEO, he particularly appreciates the relationship of this field with attention to positive user experience and high-quality content. Privately, he spends almost all of his free time on music production, which he has been doing for years – or discovering and testing new recipes.

Author
Maciej Jakubiec
Author
Maciej Jakubiec

Junior SEO Specialist

A marketing graduate with a specialization in e-commerce from the Cracow University of Economics, hailing from the picturesque Podkarpacie region. He joined Delante in 2022. In SEO, he particularly appreciates the relationship of this field with attention to positive user experience and high-quality content. Privately, he spends almost all of his free time on music production, which he has been doing for years – or discovering and testing new recipes.

Leave a comment

Get a
Free Quote

Awards

Award - Deloitte 2021 Award - European eCommerce Awards 2022 Award - European Search Awards 2022 Award - Global Agency Awards 2022 Award - IPMA Award - US Search Awards 2021

SEO SEM Agency based in Europe