This question arrives at our client desk more often than any other, usually phrased carefully, usually by a researcher writing in their second or third language who has been told conflicting things by a supervisor and a journal website. It deserves a direct answer rather than a reassuring one.
Language editing is permitted. It is routine, it is expected, and for authors who do not write in English as a first language it is close to standard practice. Journals ask that it be acknowledged, not that it be avoided. What is not permitted is anything that makes someone other than the named authors responsible for the substance of the work.
The line, stated once
Authorship is about responsibility for the content. Editing is about the clarity of its expression. The line falls between those two things, and it is more workable than it sounds.
The widely used framing comes from the ICMJE's four criteria for authorship, which an author is expected to meet all of: substantial contribution to the conception, design, data acquisition, or analysis; drafting or critically revising the intellectual content; approval of the final version; and agreement to be accountable for the accuracy and integrity of the whole. An editor meets none of them, and should not.
So the practical test on any single change is: does this change what the paper claims, or only how clearly it says it?
- Fixing a tense inconsistency in the methods — expression.
- Rewriting a sentence so the subject and verb agree — expression.
- Converting a reference list to the target style — expression.
- Softening proves to is consistent with — arguably substance, and therefore a query rather than an edit.
- Writing the discussion section — authorship.
- Adding a citation the author has not read — substance, and a fabrication risk.
- Generating a limitations paragraph the author did not write — authorship, whether a person or a model produced it.
An editor who is unsure which side a change falls on should not make the change. They should ask. That is what an author query memo is for, and the volume of queries in a returned manuscript is a reasonable proxy for how seriously a service takes this line.
What crosses it
The terms overlap in everyday use, so it is worth being precise about what actually causes problems:
Ghostwriting is when someone who materially wrote the paper is not named. That is a research-integrity problem regardless of who paid for it, and it is the thing publishers act on.
Guest or gift authorship is the inverse: someone named who did not contribute. Also a problem, and often found alongside the first.
Paper mills sell authorship positions on manuscripts written to order, sometimes on fabricated data. Publishers now retract these in batches and increasingly flag the institutions involved. Any service that offers to write your paper, guarantee acceptance, or sell a slot on someone else's manuscript is in this category, whatever it calls itself.
Undisclosed generated text is the newest version of the same issue. The major publishers and COPE have converged on a consistent position: a language model cannot be an author, because it cannot take responsibility or be accountable; and any substantive use of one in producing the manuscript must be disclosed. Note the asymmetry — models cannot be authors, but their use must be declared, which means the obligation sits with the humans either way.
What you are expected to disclose
Most journals ask for one or more of the following. Check the specific instructions for authors, because they differ and they change.
- An acknowledgement naming the editing service or individual, and what they did — typically a line such as The authors thank [name] for language editing.
- A statement in the cover letter that the manuscript was language-edited, sometimes with a certificate attached.
- A declaration of AI use, if any tool was used in drafting or editing, usually in a dedicated section rather than the acknowledgements.
- Confirmation that all named authors approved the final version — which is a real obligation, not a formality, and the reason edits should return tracked rather than applied.
Disclosure is protective. An acknowledged language edit is unremarkable to an editor; an unacknowledged one that becomes visible later is a problem you have created for yourself.
Why tracked changes matter more than they seem
If edits come back already applied, three things become impossible. You cannot see what was changed, so you cannot approve it in any meaningful sense. You cannot tell whether a change altered your meaning. And you have no record to point at if anyone later asks what the service did.
Returning work as tracked changes is not a courtesy or a workflow preference. It is the mechanism that keeps authorship where it belongs: every change is visible, every change is rejectable, and the final text is the one the authors accepted. A service that returns a clean file has quietly moved a decision from you to them.
The same argument applies to automated passes. A macro that normalises two hundred references should produce a log of what it touched, not a silently improved document.
Questions worth asking any editing service
Before sending a manuscript anywhere, these six answers tell you most of what you need to know:
- Do you return tracked changes, or a clean file? If clean, ask why.
- Will you write any part of the manuscript? The correct answer is no, without qualification.
- What do you do when a sentence is unclear because the underlying claim is unclear? The correct answer is that they query the author rather than guess.
- Do you use language models, and how? Flagging and checking is a defensible use. Generating replacement text is not.
- Will you provide a certificate describing exactly what was done? You may need it, and a service that will not describe its own work in writing is telling you something.
- Who owns the manuscript and what happens to the file afterwards? Ask about retention, confidentiality, and whether anything is used to train anything.
Anything promising acceptance, peer-review outcomes, or a specific impact factor is selling something that cannot be sold.
The grey areas, honestly
Two cases genuinely sit close to the line, and pretending otherwise helps nobody.
The first is heavy language editing of a manuscript whose English is very difficult. At some point, restructuring a paragraph for clarity starts to shape the argument. The honest handling is for the editor to work in tracked changes at the sentence level, raise the structural problems as queries rather than fixing them, and let the authors rewrite. If a service instead returns fluent prose that the authors could not have produced and cannot fully defend, the line has been crossed, whatever the invoice says.
The second is the discussion section, where language and interpretation are hardest to separate. Our rule is that an editor may sharpen how a claim is stated and must not adjust how strong it is. Turning demonstrates into suggests changes the paper's claim. That is a query.
Where we stand
Edmacro is an editing service and not a writing service. We edit language, mechanics, formatting, references, and documentation of work you wrote. We do not draft content, generate text, or add data. Every change returns tracked, with a log of the automated passes and a memo of the questions we could not answer for you — and a certificate describing exactly what we did, for the journals that ask.
That is not a marketing position. It is the only arrangement under which the paper stays yours.
Related service