AI Detector Bypass - Learn to Manually Humanise AI content with me!

Summary

The presenter explains how AI-writing detectors may assess text and demonstrates manually revising an AI-generated academic paragraph. The stated goal is to make AI-assisted writing less likely to be flagged. The presenter argues against automated “AI humanizer” tools because, in their view, these can reduce writing quality.

How the presenter says AI detectors assess writing

The presenter says detection involves more than spotting particular words. Factors discussed include:

• Perplexity: How predictable a text is to a language model. The presenter characterizes human writing as generally less predictable than AI output.
• Burstiness and variation: Differences in sentence length, structure, and word choice. AI writing is described as more uniform.
• Sentence structure and polish: Consistently flawless grammar and style may be a signal, according to the presenter.
• Repeated wording and transitions: AI may overuse phrases or connectors such as “therefore,” “however,” and “in conclusion.”
• Vocabulary: AI is said to favor safe, predictable word choices.
• Depth and coherence: Text may sound plausible on the surface while lacking an original argument or making statements that are not meaningful in context.
• Comparison with language databases: The presenter says some detectors compare writing with large collections of human- and AI-generated language.

The presenter emphasizes that simply replacing a few familiar AI phrases may not change the text’s overall structure enough.

Revision approach presented

For academic writing, the presenter recommends revising meaning and structure rather than mechanically swapping words:

• Use appropriate qualification or “intellectual hesitation.” Avoid presenting uncertain claims as established facts. Examples include “can,” “may,” “appears,” and “is likely.”
• Add relevant critique or qualification. Acknowledge limitations, inconsistencies, or other viewpoints where they genuinely fit the argument.
• Vary sentence openings and syntax. Use different sentence patterns, introductory clauses, dependent clauses, or inverted structures instead of repeating the same construction.
• Check how sentences fit together. The presenter says patterns across a paragraph—including similar sentence lengths and structures—matter more than any one sentence.
• Simplify and remove vague or redundant material. Make sure each statement contributes to the point rather than merely sounding academic.
• Understand the passage and restate its meaning in your own words. The presenter says they would ideally set the original aside, retain its meaning, and write a new version from that understanding.
• Use references where claims need support. In the demonstration, the presenter adds a sample citation while qualifying a claim.

Demonstration

The presenter revises an AI-generated paragraph about migrants’ self-esteem and communication. They soften absolute claims, remove or simplify broad phrases, rearrange sentence structures, and combine ideas to reduce repetition.

They then submit the revised passage to the same detector used earlier. The original is reported as 100% AI-generated, while the revised version receives a 0% result. This is presented as one example, not as a guarantee that the approach will work in every case.

Speakers and sources featured

• Speaker: The video’s presenter, identified by the channel as Dr Kriukow.
• Tools and sources mentioned or shown: ChatGPT (called “Chajbd” in the auto-generated subtitles), Microsoft Word, and the Detect AI detector.
• Citation mentioned in the example: “Jackson 2020,” presented as a sample reference.