I work on a rewriting tool. This is the question we get asked most and answer least comfortably, and it comes up in dev circles as often as it does in academic ones now that a model drafts the first version of half the prose anyone ships.
The assumption underneath it is that a disclosure requirement and a detector output are the same object. They are not, and the whole thing falls apart once you separate them. Everything below is what our own public pages commit to, including the parts that are bad for us.
The short answer
No. If your institution or your journal requires you to disclose AI involvement in producing a document, running that document through a rewriting tool afterwards does not remove the requirement. Rewording changes how something is expressed; it does not change what happened while the document was being written. Our own ethics page states this: "The tool changes how something is worded; it does not change the fact that AI was involved."
The assumption worth naming
The unspoken version goes something like: if the detector no longer marks it, there is nothing left to declare.
Those are two different questions that happen to arrive at the same moment. One is about what a piece of software outputs. The other is about what you did. A disclosure requirement is written about the second, and no operation performed on the text afterwards reaches back and changes it.
It is worth saying plainly because the confusion is comfortable, and comfortable confusions do not get examined until somebody is sitting opposite you.
What we say about it on our own site
We sell the rewriting. So the relevant thing is what we commit to in public rather than what we might say in a sales conversation. From our ethics page, verbatim:
"if your institution requires disclosure of AI use, a manuscript processed here is still within that requirement. The tool changes how something is worded; it does not change the fact that AI was involved."
The same page is equally direct about not being the authority on your case: "We do not interpret those rules, and we do not declare any use 'fine'." And about outcomes: "We do not promise you will pass any detector. Detection systems keep updating; text that passes today may not after the next iteration."
It also declines to pretend about what the product is for: "the humanizer is used to lower the scores detection tools return, and that is what many people pay for. Denying it would cost this page whatever credibility it has."
You do not have to take any of that on trust. It is on the site, logged out, and you can go and read the paragraphs around it.
The two things that are genuinely separate
Keeping these apart makes the rest straightforward.
What happened while you wrote. Did a model draft a section, produce an outline, generate a summary you built on, translate a passage, or write nothing at all? That is a fact about your process. It is fixed by the time you are reading this.
What the document looks like now. Word choice, sentence construction, structure. That is what any rewriting operates on, and it is the only thing it operates on.
A disclosure rule asks about the first. A detector guesses at the first by inspecting the second — imperfectly, which is the entire reason this field exists. The imperfection is not a loophole in the rule; the rule was never about the detector.
What this does not mean
Three corrections, because the point above gets overextended in both directions.
It does not mean editing is disclosable. Nobody declares a spellchecker. Where institutions draw the line between assistance and generation varies enormously, and it is their line to draw, not ours and not a tool vendor's.
It does not mean you should declare more than is asked. Volunteering a detailed account of every keystroke is not honesty, it is anxiety. Read the actual requirement.
It does not mean a rewrite is illegitimate. Revising your own prose is ordinary academic work. The question here is narrow — whether the revision changes an obligation that attached earlier — and the answer to that narrow question is no.
So what do you actually do
- Find the requirement, in writing. Your institution's academic integrity policy, the assignment brief, or the journal's author guidelines. Those are the documents that govern; anything else is commentary, including this page.
- Work out what the rule is asking about — the drafting, the editing, or both. Rules that only cover generation and rules that cover any AI assistance ask you for different things.
- Write the declaration about your process, not about your file. "A model produced a first draft of the discussion, which I then rewrote" describes what happened. "The document was processed by software" does not answer the question that was asked.
- Put it where it is supposed to go. Publishers differ on placement and on what is exempt, and we compared several of them in publisher AI disclosure policies.
- Ask if it is genuinely unclear. A question asked in advance and a question asked after submission are received very differently.
Where a rewriting tool does fit
In one place, narrowly. A report exists, it carries highlights over particular paragraphs, those paragraphs need work, and you would rather the untouched ones stayed untouched. HumanPen works from the report and edits what it marks.
That is a scope decision about text. It is not a position on your disclosure, and by design we do not hold one — "Your institution sets the rules, not us" is the heading it sits under on our own page.
If the honest answer to your situation is that you need to declare something, declare it. There is no version of this article where a rewriting step is the alternative to that.
Frequently asked questions
If the AI score drops to zero, do I still have to disclose? If your institution or journal requires disclosure of AI use, yes. The requirement is about what happened while the document was written; a score is a measurement of the text as it stands now.
Does using a rewriting tool itself need to be declared? That depends on how your institution defines the scope of its rule, and it is their definition that governs. Some rules cover generation only, some cover any AI assistance. Read the wording rather than reasoning by analogy.
Do you tell users whether their use is acceptable? No. Our published position is that we do not interpret those rules and do not declare any use acceptable; the institution's policy and the journal's guidelines are the authority.
Can any tool promise the text will not be flagged? Ours does not, and the published reason is that detection systems keep updating, so text that passes today may not after the next iteration.
A scope decision, not a disclosure decision
Where a report marks specific passages, it can define what gets edited. The declaration is a separate question with a separate answer.
Originally published at HumanPen: https://humanpen.net/en/blog/does-rewriting-change-what-you-declare?utm_source=devto&utm_medium=article&utm_campaign=d66
