13 Ways to Spot AI Writing
/Tips for determining if an article is likely written by AI.
OVERUSED WORDS. AI-written articles tend to come back to the same terms multiple times. Examples would be comprehensive, delve, meticulous, versatile and pivotal. Before 2024, overused AI words in scientific research papers were typically nouns. More recently, researchers say AI excessively uses "style" words—mostly verbs and some adjectives. The phrases AI picks up can often make the text sound more like marketing material than academic scholarship or quality news writing.
TORTURED ACRONYMS. Generative AI will sometimes pick up the wrong words for an acronym. For instance, a data science paper might use "CNN" to refer to "convolutional brain organization" instead of "convolutional neural network.”
NONSENSICAL PARAPHRASES. An academic paper written by AI might have “glucose bigotry” instead of “glucose intolerance,” where it changed a single word and did not recognize the context.
ACADEMIC CITATIONS. AI-written articles with academic citations have been known to include incorrect or incomplete references. AI writing has been also known to take quotations out of quotation marks, paraphrase them, and delete the citation.
STYLE CHANGES. A sudden change in writing style within an article or essay may indicate that the author’s work was rewritten using AI.
PERFECT GRAMMAR. A typo, particularly in student writing, could indicate the article or essay is not wholly the work of a bot. Mistake-free writing is, ironically, a red flag. However, savvy writing prompts may ask the AI to include some errors in order to mislead inspectors.
MECHANICAL STYLING. AI tends to mechanically repeat expressions that appear often in the internet material that it was trained on. The result is often uninspired and generic prose that often lacks any specific point.
ARTICLES. AI will make errors in the use of definite and indefinite articles, often because it does not recognize the context to determine whether an article is required and which one. For example, AI editors will often fail to use the definite article before common nouns such as “participants” and “results” when referring to a study. “Results show that…” is a general reference while “The results show that…” are those of the present study. Generative AI will miss this distinction.
SUBJECT-VERB AGREEMENT. AI often fumbles subject-verb agreement when the verb does not immediately proceed the verb.
VERB TENSE. Generative AI will confuse when to use past tense and present tense.
LINKING WORDS. AI editors tend to delete words linking sentences and paragraphs, such as “however,” “therefore,” “in contrast,” and “moreover.”
ARCHAIC LANGUAGE. Since older texts from the early twentieth have been more available to use as training data sets for LLMs than current writing samples, some researchers have found overuse of words that were popular then but have since fallen out of common usage as evidence of generative AI.
PREDICTABILITY. Text is more likely to be human than AI when it includes sarcasm, current pop-references or insults the reader. Writing that doesn't match predictable patterns is more likely to be human generated.