How to Detect AI Writing | 9 Methods

AI is getting better and better at writing texts that mimic humans. This means it’s becoming increasingly important to understand how to detect AI writing and know when content was written by a human.

There are strategies that can help detect AI writing, including:

• AI detectors
• Plagiarism checkers
• Manual language analysis
• Manual style analysis
• Fact-checking and source-checking
• Intent analysis
• Alternatives to writing
• Metadata and timestamps
• Comparison to other AI samples

How to detect AI-generated content

No method for detecting AI writing is completely foolproof. To have the best idea of whether a piece of writing was created by a human or AI, we recommend using various methods of detection.

• Use AI detectors
• Use plagiarism checkers
• Analyze the language
• Analyze style
• Check facts and sources
• Think about if it answers the question
• Consider alternatives to writing
• Check metadata and timestamps
• Generate an AI example and compare

Use AI detectors

AI detectors are a type of AI. They scan texts and evaluate if they look like they were written by a human or a generative AI (an AI that generates content, like ChatGPT).

So, how do AI detectors work? AI detectors use machine learning and natural language processing—the same principles behind generative AI—to analyze different features of a text. Two features AI detectors look at are:

1. Perplexity, or how likely a text is to surprise the average reader
2. Burstiness, or how varied sentence structure and length is across a text

When perplexity and burstiness levels are low, it’s more likely the text was written by AI. AI’s goal is to write content that’s easy to understand for the average reader (which leads to low burstiness), and its algorithms rely on predictability in language (which leads to low perplexity).

Bear in mind these tools are never 100% accurate. They can make mistakes, especially as generative AIs are becoming more sophisticated. Moreover, humans are getting better at editing AI-created content to help it pass these detectors.

If you want to try an AI detector as a way to detect AI writing, you can use QuillBot’s AI Detector for free.

Use plagiarism checkers

Plagiarism checkers are another type of digital tool to evaluate writing, but their goal is different from that of AI detectors. Plagiarism checkers look to see if a text was copied from another source.

AI detectors analyze text based on what they’ve “learned” during training. On the other hand, plagiarism checkers scan a text and then compare it to a large database of sources.

That said, plagiarism checkers can also provide insight into how to detect AI writing. AIs often rely on published information without citing the sources. AIs mostly write unique sentences, but they may occasionally copy something from a source or paraphrase it so slightly that the checker marks it as plagiarized.

Plagiarism checkers are mostly used for academic writing (theses, dissertations, research papers, etc.) but could be used in other contexts, too.

Analyze the language

Even if we work more slowly, the human brain is still leagues beyond AI when it comes to analyzing language. Therefore, another way to detect AI writing is to manually analyze a text yourself.

When analyzing writing you think may have been created by AI, look for:

• Repetition or redundancy in words and concepts
• Formulaic or uncreative sentences
• Very predictable word choices and sentence structures
• Little variation in sentence structure and length
• Generic explanations and a lack of personal examples
• Unnatural word choices for the desired meaning (e.g., “tapestry” for “mixture” or “embark” for “start”)
• Grammar mistakes that inhibit flow or meaning

Analyze style

Generative AI is quite good at writing text that makes sense, but it still sometimes struggles with applying style. This makes analyzing style another way to detect AI writing.

AIs are generally programmed to act as helpful assistants, so their writing tends to be cheerfully formal. Unless the user prompts them to adopt a particular style, they’ll default to this one.

If you know the tone of voice of the person whose work you’re reviewing, you can compare their style to what you find in the text. If there’s no overlap, the text may have been written by AI. Likewise, if you find a flat, monotonous tone of voice in the text, it could be another indicator that AI wrote it (unless the human author also writes monotonously).

Another thing generative AI struggles with is themes, particularly thematic closure. Human writing tends to have themes worked in throughout the writing, and a good conclusion brings those themes to a close. AI often doesn’t do this, so look for a lack of themes to help detect AI writing.

Finally, look at style when it comes to format. AIs sometimes have “default styles” they use to present information. AIs use these styles because they’re clear formats for presenting information, and if you pick up on what they are, you can easily spot them.

Check facts and sources

If you’re learning how to detect AI writing, another method is to check facts and sources. AIs can be wrong, or they may use irrelevant or outdated sources.

“AI hallucination” is the term used for when AIs perceive nonexistent patterns or objects, which leads to nonsensical or incorrect outputs when it creates content. Factual inaccuracies in a text could indicate it was written by AI.

AI can also commit errors when it comes to the consistency of facts. For example, it may say one thing at the beginning but then contradict itself later. Checking that facts align throughout the text can help you detect AI writing.

Check sources, too. If a user asks AI to cite sources, it will sometimes—especially when dealing with niche material—cite outdated, irrelevant, or disreputable sources. It also may not follow citation manuals (e.g., the APA citation guide) correctly.

Think about if it answers the question

When trying to detect AI content, another thing you can examine is if the content answers the question it was supposed to. AI sometimes gets too far away from the question at hand.

This is especially true for web content. AI copywriting and content writing is growing, as AI is cheaper and faster than human writers. However, the quality of content almost always suffers.

For example, in SEO content writing, search intent directly impacts how a writer prioritizes and structures content. AI doesn’t always pick up on the nuance of search intent and can therefore fail to answer the question people really want answered.

Consider alternatives to writing

Whenever possible, use oral alternatives to help detect AI writing. Speaking to someone about their writing can reveal if they know it inside out (and probably wrote it) or if they’re not familiar with it (and may have used AI).

This method is most relevant for education. Teachers can assign oral presentations, ask students to defend an essay, or set up class discussions to complement written tasks.

Just keep in mind that this method isn’t foolproof either. Some students may study their AI-created content and be able to defend it flawlessly. And others may complete their own writing but struggle with oral exercises.

Check metadata and timestamps

Checking metadata and timestamps in documents can also help you detect AI writing. Timestamps show how a document evolves over time and are one type of metadata. Metadata is the data “in the background” of a document (file names, authors, software info, and so on).

For example, using the version tracker in word processing software could help you see how a document was written and edited. If timestamps show that large portions of text were pasted in very quickly, it could be a sign of AI-created content. Very few to no revisions could also point to a lack of human editing.

Some AIs claim that they plan to incorporate watermarks into their content. AI watermarks are invisible markers or patterns embedded within digital content that can be used to identify when it was generated by AI. These don’t exist yet, but as the field of AI ethics grows, we may see them develop.

Generate an AI example and compare

Finally, if you’re experimenting with how to detect AI writing, go to the source. Put the question or prompt of the writing into a generative AI and see what result it gives you. Then, compare that result to the text you’re analyzing.

AIs often provide unique responses to the same prompt, but this may not be the case 100% of the time. And even when responses are unique, there will most likely be glaring similarities between them. Spotting these can help you match texts you think were written using AI to texts you know were written with AI.

Frequently asked questions about how to detect AI writing

How can I use AI for writing?

There are many opinions about how to use AI for writing, and these may vary based on the industry or subject matter at hand.

For example, AI copywriting may be acceptable, but in academic writing, you should avoid relying completely on AI to generate output. Educators know how to detect AI writing.

You can use AI to help you brainstorm research questions or organize an outline. When you use AI in these capacities, check if your institution requires you to include it as a citation.

You can also use AI to check your grammar and spelling after writing. A QuillBot Grammar Check will do just this.

What are some common AI words?

Some common AI words are:

• Tapestry
• Crucial
• Intricate
• Interplay
• Elevate
• Resonate
• Enhance
• Offerings
• Leverage
• Embark

Just because you see one of these words in a text doesn’t mean it was written by AI. But these words are commonly used when AI generates content.

If you’re learning how to detect AI writing, brushing up on common AI words is one way to do that.

You could use QuillBot’s free AI Detector to analyze writing and see if it looks like it was created by AI.