Free Methods to Reduce the AI Detection Rate of Papers: From 74% to 2.6%

Recently, many universities have adopted AIGC detection tools to identify the proportion of AI-generated content in graduation theses—commonly referred to as the "AI detection rate." Writing a graduation thesis is already stressful enough, with plagiarism checks and similarity reduction driving people to frustration; now we have to deal with the "AI detection rate" too? Does this mean we can no longer use AI to assist with thesis writing? Absolutely not! The key lies in understanding the principles of AIGC detection and how to mitigate associated risks.

1 How Exactly Is the AI Detection Rate Calculated?

When it comes to checking the AI detection rate, everyone probably wonders: What exactly is being analyzed? How do platforms distinguish between human-written and AI-generated content? AIGC detection works by comparing text against characteristics such as linguistic style, sentence structure, and content coherence. Simply put, it assesses whether a passage "sounds like it was written by a human."

Verbose Content

AI-generated sentences are often overly long and complex, making them difficult to comprehend. Text generated by tools like Deepseek, in particular, sometimes resembles the exaggerated style of "cheesy marketing copy."

Example: Against the backdrop of accelerating globalization and its complex impacts on cultural diversity, this study employs an interdisciplinary perspective to clarify the dynamic balance between cultural integration and conflict, and explores the significant implications of this balance for the formation of social identity.

Incomplete Content

AI-generated text is frequently overly concise, lacking adverbs and specific details, which makes it hard to follow.

Example: This study analyzes data, draws conclusions, and puts forward suggestions.

Rigid Content Structure

AI text tends to use formulaic organizational patterns, such as "firstly, secondly, finally" or "first, second, third." While this may seem structured, it results in severe templating.

Example: There are three solutions to this problem: resource optimization, technological innovation, and personnel training. First, the budget can be replanned; second, new technologies can be adopted; third, training courses can be held.

2 Practical Prompts to Reduce AI Detection Rate

How can we easily and effectively lower the AI detection rate? My answer is to use AI tools themselves to eliminate AI traces—yes, using AI to reduce AI detection rates, or "fighting fire with fire." I recommend ChatGPT for this purpose’s overly florid style, ChatGPT generates text that is much closer to human writing.

Avoid AI-Style Transition Words

Prompt: Replace complex transition words (e.g., "therefore" , "at the same time") with simple ones (e.g., "so", "in addition" ). Remove unfamiliar transition words and delete concluding paragraphs (e.g., sections starting with "in summary" ), retaining only core arguments to ensure content stays tightly focused on the topic.

Optimize the Thesis’s Logical Structure

Prompt: Add logical connectors (e.g., "because" , "it can be seen that") at the beginning or end of sentences. This clearly expresses causal, progressive, or contrasting relationships, making the text easier for readers to understand and improving its fluency.

Highlight Paragraph Key Points

Prompt: Each paragraph should revolve around a central thesis. Delete background information or content irrelevant to the topic. Add connectors or phrases (e.g., "next", "on the other hand") between paragraphs to enhance logical coherence and prevent readers from feeling disoriented.

Adjust Paragraph Structure

Prompt: Start each paragraph with a topic sentence that clearly states the central idea, followed by 2–3 supporting pieces of evidence (e.g., data or examples). Keep paragraphs under 5 lines in length to avoid information overload and reader fatigue.

Eliminate Redundant Expressions

Remove repetitive adjectives (e.g., "important and crucial" ) and adverbs with no practical meaning (e.g., "very" , "obviously"). Retain only necessary modifiers to make the language more accurate and concise.

Simplify Jargon and Long Sentences

Prompt: Replace professional jargon with plain language (e.g., "regression analysis" simplified to "data analysis" ). When necessary, add a brief explanation in parentheses (e.g., "data analysis (i.e., regression analysis)" ). Split complex long sentences into 2–3 short ones, with each short sentence conveying only one piece of information.