How to avoid AI detection

People arrive at this question from two very different places. Some used AI and want to hide it. Others wrote their own work, got flagged, and want to stop being accused. The techniques overlap; the ethics do not.

What actually lowers a score

• Vary sentence length deliberately. Follow a long sentence with a short one. This raises burstiness, the single biggest lever.
• Add concrete specifics, names, numbers, dates, examples. Specificity raises perplexity in a way a thesaurus never will.
• Cut stock transitions and dead phrases: moreover, furthermore, it is important to note, in today's fast-paced world.
• Vary how sentences open. Three consecutive sentences starting the same way is a strong marker.
• Keep your own voice: contractions, asides, the punctuation you actually use.

What does not work

• Synonym swapping. It is what early humanizers did and what detectors were retrained to catch.
• Unicode tricks and invisible characters. Detected trivially, and treated as deliberate deception rather than a stylistic matter.
• Running text through three humanizers. Output degrades faster than the score falls.
• Any tool promising a guaranteed bypass rate. Nobody can guarantee that against systems they do not control.

Notice what the working list is

Vary rhythm, add specifics, cut dead phrasing, keep your voice. That is not evasion advice, it is the standard advice in every book on writing ever published. The techniques that survive detector updates are the ones that genuinely improve the prose, because they change the text rather than disguising it.

The better question

Not how do I avoid detection, but which of my habits make my writing read as machine-generated. That question has a fixable answer, and fixing it helps whether or not anyone is running a detector.