Achieving Undetectable AIGC: Case Studies on How to Bypass AI Detectors and Fall Below the 10% Flag

Friends, many students asked the editor, "teacher, the aigc rate of my paper can not be reduced. I have used all kinds of online prompts. Why is it still greater than 15%. Is there a thorough solution? ”First of all, no one can guarantee that every chapter can be reduced from 100% to 0% by any method, but I can responsibly say that it can be reduced to less than 15% by my method. Looking at my last screenshot, my duplicate checking rate of 90% has been reduced to 3.8% for the first time. It is easy to pass the Turnitin, iThenticat.

1 Correct process for detecting aigc rate

Focus first: do not use the school system for preliminary inspection (most schools only have 1-2 free opportunities, which is a pity to waste). Follow the three steps of [preliminary inspection → modification → review] to save time.

Step 1: initial check . The first check of free tools to locate high-risk paragraphs gives priority to the use of Turnitin+Grammar (Academic Version) combination: Turnitin: one free check per day (within 5000 words, super words are checked in chapters) focuses on the "high-risk paragraphs" marked in red/yellow, copies and pastes them into the document to mark them, and then modifies them accordingly. Grammar (Academic Edition): the free edition can detect the repeated use of "first, second, therefore, so" and other words in the "aigc mechanical sentence", and can also check grammar errors, and modify the aigc problems and the language expression of the paper.
Step 2: modify . After the initial inspection, the first round of modification can be started without entanglement of specific values, and the high-risk section of the red mark can be circled.
Step 3: review . After the modification of high-risk paragraphs, use the school designated duplicate checking system (HowNet/VIP/Grida, etc.) to re detect: ① when uploading, be sure to adjust to the school format setting (font/line spacing/reference mark), otherwise Grida will misjudge "nonstandard reference" as aigc, and the standard reached is "low risk", and the threshold of most schools is "aigc rate ＜ 15%". If there is a "medium risk" paragraph, just modify it again according to the following method.

2 Nine practical methods to reduce aigc

2.1 AI auxiliary descent AI

Instead of "Rewriting" Ai directly, it gives precise instructions to make AI produce more artificial academic content. Instruction template: help me rephrase this passage, requiring: ① keep the logic of XX (core point, such as "the impact of aigc on academic writing") unchanged, ② add a specific detail, such as "contact aigc in XX literature to generate content repeatability problem", ③ do not use rigid cohesive words such as "first, second, and addition", and the sentences are staggered in length. [paragraph to be rewritten]

Before revision: aigc is widely used in academic writing. It can quickly generate the first draft and help to find literature, but it also has the problem of content duplication.
After modification: the application of aigc in the academic writing scene can help researchers quickly generate the first draft. For example, a 3000 word literature review can be completed in one hour, and the literature can also be located through keyword retrieval. However, according to the research of digital academic norms, the content generated by aigc has the problem of "logical repetition of paragraphs", which needs to be modified manually.

2.2 Supplementary subject

AI sentences often lack a subject or have unclear reference, such as "draw a conclusion through experiments", "this method is better" and so on. It is OK to add a subject.

Before modification: the data analysis shows that it needs to be further tested (lack of subject+unclear reference)
After modification: a conclusion is drawn through the analysis of 200 questionnaires. This study believes that this conclusion still needs to be verified by subsequent experiments. (add the subject "this study" to clarify the meaning of "this conclusion")

2.3 Optimize connectives

AI likes to use the parallelism of "first/second/additional/Therefore"+comma, and replace it with flexible connectives+integrated short sentences.

Before modification: the users of the short video platform grow rapidly and have high user stickiness, so it is suitable for marketing.
After modification: "the short video platform has both high user growth (12% in 2024) and high stickiness (90 minutes per day). Because of these two characteristics, it is very suitable for enterprises to do precision marketing". Use "both/and" to replace the parallel conjunctions, and combine short sentences+supplementary data.

2.4 Reconstructing the total starting sentence pattern

AI likes to use "the definition of XX: XXX" and "the characteristics of XX: XXX" as the starting sentence. If the Turnitin reaches the elevation risk, it can be changed into a coherent sentence pattern of "one is/two is/three is".

Before modification:

Definition of aigc: content assisted by AI technology
Characteristics of aigc: high speed and high repetition rate.
Application of aigc: academic writing and marketing copywriting.

After modification: aigc can be divided into three parts, one is concept explanation: AI generates content (text, image, etc.), the other is characterized by efficient

2.5 Supplementary examples/Literature

AI content is mostly vague conclusions. For example, aigc is risky. Adding a literature point of view or case will reduce the aigc rate and increase the academic nature.

Before revision: aigc has great ethical risks in academic writing. ”(empty)
After revision: aigc has obvious ethical risks in academic writing. As Li et al. (2023) said in academic ethics research, 32% of students directly use AI to generate content without indicating the source, which does not meet the originality requirements of academic integrity. "(supplementary literature+data cases)

2.6 Scene narration+humanized tone

Think of yourself as "giving a paper to the students", explain the professional content of AI in popular words, and integrate it into the paragraph (without major changes, the risk rate drops quickly).

Before modification: recommendation system based on collaborative filtering algorithm, personalized recommendation through user behavior data, (too mechanical)
After modification: when we usually brush the shopping app, the "frequently viewed goods" pushed by the system is actually a collaborative filtering algorithm. In short, the system will first collect the data of goods you have browsed or added to the shopping cart, and then push something you are interested in based on this information, that is, personalized recommendation, (add scene+popular explanation)

2.7 Deletion method

If the total number of words in the paper exceeds the requirements of the school (for example, the school requires 8000 words, and I wrote 10000 words), delete some unnecessary paragraphs directly (such as "irrelevant background introduction" and "repetitive conclusion summary" written by AI), and check them directly after deletion.

For example, the paper focuses on "the risk of aigc in academic writing". AI has written a "development history of generative AI (gpt-1 in 2017 to gpt-4 in 2023)" - this has little to do with "risk" and is deleted directly (save time+reduce aigc rate).

2.8 Sentence upgrading

Replace the simple words of AI with academic synonyms, and make short sentences longer (don't pile up rare words, it's OK naturally).

Before modification: aigc has fast generation speed but poor quality, "simple words+short sentences"
After modification: the speed of aigc's production content is much faster than that of manual work, but in terms of quality, the texts produced by aigc tend to have chaotic logic and groundless views, which can't be directly used in academic writing. "

2.9 Expansion method

When encountering texts written by AI that explain professional terms, such as "blockchain technology", we will follow the method of "delete duplication+expand examples": delete: delete the repetitive parts of AI, such as "blockchain is a distributed ledger, that is, a decentralized ledger, that is, an ledger without a central node", delete the words "decentralized ledger/an ledger without a central node", and expand: explain professional terms in vernacular and examples.

Before modification: blockchain technology is a distributed ledger technology, that is, a decentralized ledger technology, that is, a ledger technology without a central node, which can ensure that data cannot be tampered with, "(repeated explanation)
After modification: "the essence of blockchain technology is distributed ledger technology - in short, unlike the bank ledger, which is managed by a bank organization, but maintained by miners all over the world like bitcoin ledger, so the recorded data is difficult to tamper with." (delete duplicate+fill in examples)

3 Summary and supplement

Self checking logic: don't make sentences illogical just because you want to reduce the duplicate checking rate of aigc. Just read it once to ensure that the paragraph is logical and smooth. The citation format can't be sloppy: Turnitin and iThenticate misjudged "unmarked AI citations" as high-risk. Citations should be marked with "author+year+page number" in strict accordance with the requirements of the school. 0% is not the only standard: most schools can pass if<15%, and excessive modification will waste time.