1. You write an essay.
2. [Word-level randomization] An algorithm looks at it, as a first cut, replaces uncommon words with common equivalents.
3. [Sentence-level randomization] Changes each sentence to a randomly selected grammatically correct equivalent.
4. [Paragraph-level randomization] Adds spurious repetitions, replace some paragraphs with automated summarized versions.
i.e. destroy the rhetorical effect, but get the message across so as to defeat NLP-based identification. If the content of the message is important, and the final version is okay to the author, this might be safer for anonymous disclosures.