Reference lists and in-text citations are often the parts of a manuscript you must not paraphrase. Detectors may still skim them; billing systems that charge by “whole document words” punish you for leaving them alone.

HumanPen meters work by rewritten continuous prose. Frozen citations, numbers, and formulas are outside the rewrite bill by design. That matches the academic constraint: keep the evidence trail intact while you revise only the flagged narrative sentences.

Why this matters on a real report

• A similarity instrument and an AI-writing instrument are different scores. Cleaning a bibliography rarely fixes an AI highlight on a methods paragraph.
• If you “humanize” a reference list, you risk broken DOIs, wrong years, and silent claim drift.
• Paying for unchanged citation tokens is wasted spend and wasted attention.

Workflow with HumanPen

Import the AI report → map highlights to document spans → rewrite only flagged continuous prose → keep the reference list and numeric claims frozen → re-export DOCX with layout preserved as far as the pipeline allows → re-read the new report before another pass.

No undetectable claim. No “guarantee you will pass.” Just a narrower edit surface and metering that does not charge you for the bibliography you correctly left alone.

— Charles Li

Try the product: https://humanpen.net · https://lnk.bio/humanpen

When an AI writing report comes back with highlights, the useful move is usually not a full-document rewrite. Open the report, jump to the highest-confidence spans, and decide whether each span is boilerplate, over-smooth paraphrase, or a claim you still stand behind.

HumanPen is built around that order of work. You import a Turnitin or iThenticate-style AI report (or paste the spans yourself), and the tool focuses on the passages that were marked — instead of washing every paragraph “just in case.” Citations, numbers, and formulas stay frozen; continuous prose that was never flagged stays alone.

A practical sequence

1. Read the report as a map of highlights, not as a single guilt score.
2. Rank flagged spans by how much they matter to your argument.
3. Rewrite only those spans; leave clean sections untouched.
4. Re-export, re-check, and only then decide whether another pass is needed.

That sequence matches how most detectors present evidence: span-level marks first, aggregate percentages second. Treating the percentage as a “rewrite everything” alarm tends to erase voice and burn words you did not need to change.

What HumanPen does and does not claim

HumanPen is a document-level rewrite assistant for authors who already have a report in hand. It is not a guarantee of any score, not an undetectable mode, and not a bypass of institutional rules. If a recheck still shows flagged prose at or above the product threshold, the product can re-run on the still-flagged spans without charging those rounds again — that is a workflow promise about remaining marks, not a promise that a school will accept the paper.

Author: Charles Li (Goldsmiths alumni)

Product: https://humanpen.net · links: https://lnk.bio/humanpen