How ClearPen's AI Humanizer Works: Block-Based Rewriting, Explained

ClearPen is often described as an "AI humanizer," but under the hood it is closer to an editing pipeline than a rewriting prompt. This post walks through what happens between pasting a draft and getting a result back — and why each step exists.

Why a Single Rewrite Prompt Is Not Enough

The simplest way to "humanize" text is to ask a chatbot to rewrite it in a more human style. It rarely works well, for a structural reason: AI detectors look for the statistical fingerprints of language-model output — highly predictable word choices and very even sentence rhythm — and asking another language model to rewrite the text tends to produce more of the same fingerprints.

The second problem is length. A single pass over a long document spreads its effort unevenly. In our testing, whole-document rewriting consistently improved the first part of a document and left the later sections largely untouched, so the result read human in the introduction and machine-written by the conclusion.

ClearPen's pipeline is built around both problems.

Step 1: Score the Input First

Before rewriting anything, ClearPen scores your text with its built-in AI detector. The before score streams to your screen within a few seconds.

If the text already reads as human — a score in the green band — ClearPen stops. Nothing is rewritten and no words are used. You are told how many individual sentences, if any, still lean AI, and you can choose to humanize anyway. We explain why in ClearPen's fair-billing safeguards.

Step 2: Split the Document Into Flow-Preserving Blocks

ClearPen divides the document into blocks of roughly 250 words. The split rules are strict:

• Blocks never break mid-sentence.
• Paragraph flow is respected wherever possible.
• Headings pass through unchanged.
• Rejoining the blocks reproduces the document's structure exactly.

Each block is then processed independently. This is what keeps quality consistent from the first paragraph of a long document to the last. In one internal test on a 3,835-word article, whole-document rewriting left the sentence-level AI average near the top of the scale, while block-based processing brought it to single digits with no block left reading strongly as AI.

Step 3: Rewrite in Layers, Sentence by Sentence

Within each block, ClearPen applies several rewriting stages in sequence, each using a different model and a different kind of change:

• Restructuring that changes sentence construction and rhythm, not just vocabulary.
• Natural variation in word choice and sentence length — the "burstiness" that human writing has and generated text usually lacks.
• Targeted paraphrasing applied only to the sentences that still score as AI after the earlier stages.

That last point matters. After the first stages, every sentence in the block is scored individually. Sentences that already read as human are left alone; only the ones that still read as AI get further rewriting, and each candidate rewrite is checked for similarity to the original so the meaning does not drift.

Layering outperforms any single technique. In our experiments, each stage on its own made a modest difference; combined, they made a large one.

Step 4: Repair Grammar With a Light Touch

Multi-stage rewriting can leave small artifacts: a slightly awkward phrase, an article in the wrong place. ClearPen runs a minimal-edit repair pass that fixes those and nothing else. It is explicitly not a second rewrite — heavy "improvement" at this stage tends to undo the natural variation the earlier stages created.

Only blocks that remain stubborn after all of this get a heavier rewrite as a fallback.

Step 5: Rejoin, Polish and Apply Your Spelling Preference

The blocks are rejoined and the whole document gets a final grammar and spelling polish in the convention you chose: US English or UK English.

Step 6: Re-Score and Compare

The finished text is scored again and compared with your original. Three checks happen here:

• A quality gate. If the result does not meet ClearPen's internal quality bar — overall score, how much of the text is still flagged, and whether the end of the document scores much worse than the beginning — the pipeline automatically escalates with additional parallel attempts and keeps the best.
• Input protection. If the best rewrite would still score worse than your original, ClearPen returns your original text and tells you so.
• The heatmap. You get the before and after scores plus a per-sentence colour map for both versions, so you can see exactly what remains.

Short inputs work slightly differently: ClearPen generates three candidate rewrites and keeps the one with the lowest AI score that still preserves your meaning.

The Controls You Have

What ClearPen Does Not Do

Being clear about limits is part of the design:

• English only. Text in other languages is detected and returned unchanged rather than rewritten badly.
• No promises about third-party detectors. ClearPen measures its own before-and-after score. Other detectors use different methods, change without notice and frequently disagree with each other.
• No change to authorship. ClearPen changes how text reads, not who wrote it. Use it where AI assistance is permitted and within your organization's policies.

Related Reading

• What is ClearPen?
• Why ClearPen shows a before-and-after AI score
• How GPTZero works — and why it gets it wrong