AI-generated writing has a certain pattern due to repetitive expressions and grammatical structures. Recently, a writing guide that leverages this pattern, known as the “Humanizing Prompt,” has been unveiled.
The prompt designed by Jordan Gibbs, co-founder of AI hiring company Elevora, specifies sentence rules to make AI writing appear human-like. It has been proven effective in experiments where it passed AI detectors.
The prompt is essential and should be inserted at the beginning of AI chat sessions or as a system command. The initial sentence states, “You are a human writer.” This is followed by consistent writing rules: sentences must average 10–20 words in length and remain in active voice. Avoid complex sentences and include only one idea per sentence.
Complicated punctuation like semicolons and em dashes is prohibited. Conjunctive adverbs like “however,” “nevertheless,” and “also” are also restricted. Formal or exaggerated language, such as “cutting-edge,” “innovative,” and “efficient,” are classified as forbidden words.
The key principle of the prompt is to “ban abstract expressions.” Jordan Gibbs noted, “AI often uses vague expressions like ‘good,’ ‘fast,’ ‘efficient,’ but humans speak with specificity.”
For instance, saying “user experience has improved” is ambiguous. Instead, you should guide the AI to use specific figures like “the number of button clicks decreased from an average of 3 to 1.” A statement like “this system is responsive” should be written as “response time was shortened from 3 seconds to 1 second.”
In an experiment, Jordan Gibbs applied this prompt in hypothetical email scenarios. The scenario was to “explain in a friendly and natural manner why you can’t attend the July conference in Chicago, and share some tips based on last year’s participation.”
The AI initially generated a response in the usual manner, which was overly formal and abstract, causing it to be categorized as “AI-generated” by detectors. However, after applying the humanizing prompt, the second response was deemed human-written due to its realistic reason for absence and concrete advice.
Through this experiment, Gibbs demonstrated that word choice, sentence structure, and fact-based descriptions can change detector judgments.
Generative AI can be effectively used in marketing, customer service, content creation, etc., due to the demand for naturally human-like writing. However, this also has the potential to compromise transparency.
If AI-written content is disguised as human, it becomes unclear who the real author is, leading to issues of content reliability. Currently, some countries are preparing legislation to mandate labeling AI-generated content. Korea also needs to discuss this matter.
As AI writing technology rapidly evolves, ethical standards for “how” it should be used must also be established. Technology is merely a tool, and responsibility ultimately lies with the user.