How to Use ChatGPT and Grammarly for Article Editing Without Violating Journal Rules

You open your draft, copy-paste a paragraph into ChatGPT, and hit enter. Within seconds, the model rewrites your entire section with polished phrasing. You paste it back into your manuscript, hit submit, and hope for the best.

Three weeks later, your editor sends a message: your article is being reviewed for an ethical compliance violation. Undisclosed AI editing had replaced the mandatory AI declaration. Your submission is returned.

This scenario is more common than you might expect. The same 2026 policy updates that tightened journal disclosure rules across major publishers (Elsevier, Springer Nature, Wiley, Taylor & Francis) caught researchers who assumed “basic editing” meant “no disclosure needed.” The gap between what an AI tool does and what a journal considers acceptable is wider than most authors realize.

Here’s what actually changed, why Grammarly and ChatGPT for academic editing are treated differently, and the exact workflow that keeps your manuscript in compliance with every publisher’s requirements.

Grammarly and ChatGPT Are Not Treated the Same by Publishers

The core distinction that determines your compliance obligation is functional, not stylistic. Publishers classify AI editing tools into two categories: assistive line-editors and generative editors.

Grammarly is classified as an assistive line-editor. It operates on a paragraph-by-paragraph basis, offering inline suggestions for spelling, grammar, punctuation, and tone. Most major publishers treat this category as a minor editing exemption. Grammarly does not rewrite entire sections; it flags specific errors and lets the author decide whether to accept each suggestion. That limited scope is what gets it exempted from formal disclosure across Elsevier, Wiley, and Taylor & Francis.

ChatGPT is classified as a generative editor. It operates at the sentence and paragraph level, rewriting entire passages rather than flagging individual errors. It generates new text, restructures arguments, and can alter the meaning of your original draft. This is what triggers mandatory disclosure across every publisher.

The distinction matters because each publisher defines “basic editing” narrowly. Springer Nature’s traffic-light framework calls this Green-level use—”Assistive AI use that supports expression, organisation, or efficiency without influencing scientific or scholarly judgement.” Grammarly fits this definition. ChatGPT’s broader rewrites fall into Amber or even Red territory, depending on how much of your draft it alters.

This is why you can use Grammarly freely, submit it without a formal declaration, and still use ChatGPT on the same manuscript—so long as you document the ChatGPT use separately. The same tool is exempt at one publisher but flagged at another. Here’s how the four major publishers actually classify them.

Practical Implications for Mixed-Tool Workflows

The practical implication is straightforward: if you use Grammarly for a 10,000-word manuscript and ChatGPT for three paragraphs of style adjustment, only the ChatGPT use requires declaration. But if you paste an entire section into ChatGPT and accept its rewrite, that entire section is subject to disclosure. This is why tracking which tool was used for which section matters—it is the single most reliable way to write an accurate disclosure statement.

The Publisher Comparison Matrix

Every major publisher draws its own line between exempt editing and required disclosure. The table below maps the policies as of June 2026, based on each publisher’s official author guidelines.

Every publisher requires the same four pieces of information: the tool name, version number, purpose of use, and how the author reviewed the output. You can satisfy all four with a single paragraph in your Acknowledgments section.

Note that even publishers that exempt Grammarly from formal disclosure still require human accountability. Elsevier’s policy states that authors must review and verify all content generated by AI, even if the tool is exempt from declaration. The accountability obligation does not change; only the declaration requirement does. This is why the distinction between exempt and required disclosure matters. The responsibility for accuracy remains yours regardless of whether you file a declaration.

Editing Quality: Why More Corrections Don’t Mean Better Editing

You might assume that a tool that catches more errors produces higher-quality editing. The evidence contradicts that assumption.

A peer-reviewed study published in PLoS One in February 2026 directly compared the editing quality of ChatGPT, Grammarly, and a human editor on two draft manuscripts written by Ugandan researchers (August et al., 2026). The results were clear.

ChatGPT made roughly ten times more corrections than Grammarly. It also made about three times the number of corrections the human editor produced. But quantity did not translate into quality. Only 61% of ChatGPT’s corrections were judged as genuine improvements. Fourteen percent of the corrections actually made the text worse, and roughly a quarter were judged neutral.

Grammarly made fewer corrections, but they were narrowly scoped to spelling and grammar. The human editor made the fewest overall corrections, but 90% of its readability edits were judged as improvements.

The finding matters because it reveals a pattern many authors encounter: generative models tend to over-edit, producing stylistic changes that sound polished but don’t improve the scientific meaning. A higher correction count is not a quality signal. It is a noise signal.

Practical takeaway: If you use ChatGPT for editing, you need a verification step. Review each suggested edit against your original text. Do not accept changes blindly. The human editor in the PLOS One study did not make more corrections; it made smarter ones. The same principle applies to AI—only accept corrections you can verify.

The Risk of Lost Content

The implications extend beyond quality. The PLOS One study also found that ChatGPT removed 10 pieces of key information from one of the manuscripts, including in-text citations and a reference to one of the tables. This is not an edge case. It is a documented risk of generative editing tools that no line-editor can produce. When a tool rewrites entire sentences, it loses the author’s original intent alongside its errors. This content loss is another reason why verification is not optional—it is a compliance requirement as well as a quality safeguard.

Why This Matters for Journal Submissions

These quality results have direct implications for journal submission. A manuscript that passes through ChatGPT before submission may contain stylistic improvements that weaken the author’s original argument. Editors may notice the difference between the author’s voice and the AI’s voice, which can raise suspicion about undisclosed editing. The PLOS One study’s recommendation—that authors should “consult journal guidelines and disclose their use”—is the only safe approach.

What Changed in 2026: The Disclosure Updates

Two policy developments shifted AI disclosure from optional to mandatory across biomedical and general science publishing. Understanding these updates prevents compliance gaps at submission.

ICMJE’s Dual-Disclosure Requirement

The International Committee of Medical Journal Editors (ICMJE) updated its Recommendations in 2026 to require dual disclosure. Authors must now disclose AI-assisted technology use in both the cover letter and the submitted manuscript. Previously, disclosure was optional or limited to the manuscript alone. The cover letter disclosure ensures the editorial office sees it during initial screening, before peer review begins.

The ICMJE disclosure must include: the specific tool used, its version, the purpose it served (e.g., language polishing, data analysis, figure generation), and an affirmation that human authors verified the output. Failure to disclose may be treated as scientific misconduct by member journals. The ICMJE framework is the most widely adopted standard in biomedical publishing, and many general-science journals follow it as well.

The Lancet’s Three-Element Documentation

The Lancet Group’s 2026 AI guidelines introduced a stricter framework: tool name and version, purpose, and extent of author review. These three elements replace the vague “AI-assisted editing” language that authors used in 2024 and 2025. The Lancet also requires that AI-generated content be verified against original sources, and that AI cannot be used as a cited reference. The full Lancet author guidance is available at thelancet.com.

These two updates represent a shift from “permission” to “documentation.” Publishers are not banning AI tools; they are enforcing transparent declaration. The COPE position statement on authorship and AI tools, first published in February 2023, established the foundational principle that AI cannot be an author and that authors retain full accountability. The 2026 updates built on that foundation by adding procedural requirements. The full COPE position statement is available at publicationethics.org.

The Transparency Gap

A related but distinct issue is that one-third of researchers never formally disclose their AI use when submitting, according to Springer Nature Group data cited by TheSify’s 2026 guide. This “transparency gap” persists even though disclosure requirements have tightened across the industry. The gap is most pronounced among researchers who use AI tools for what they perceive as “minor” editing tasks. They assume that basic language polishing falls outside the scope of mandatory disclosure. The publisher response is increasingly that perception does not override policy, and the compliance burden is on the author to know their journal’s specific requirements.

Your Safe Editing Workflow

You can use AI tools for manuscript editing without violating journal rules. The workflow below is built around the four elements every publisher requires and the PLOS One study’s findings about editing quality.

Step 1: Choose Your Tool and Document the Intent

Start by identifying which tool you need. If you only need spelling, grammar, and punctuation fixes, Grammarly (or the built-in spell-checker in Word) is sufficient. If you need sentence restructuring, readability improvements, or style adjustments, ChatGPT may be appropriate—but it will require disclosure.

Document your intent upfront. Write one sentence that states: “I used [Tool Name] for [purpose].” This sentence becomes the foundation of your disclosure statement.

Step 2: Use the Tool and Track Its Output

When using Grammarly, accept only corrections you verify. When using ChatGPT, paste text paragraph-by-paragraph, not section-by-section. This limits the scope of each edit and makes it easier to review.

After each editing pass, create a comparison document. Use Word’s “Compare Documents” feature to generate track changes. This is the same method the PLOS One study used and the most reliable way to identify what the tool changed versus what you wrote.

Step 3: Verify Every AI Suggestion

This is the step most researchers skip. Open the track-changes document and review each suggestion. Ask yourself: does this edit preserve the scientific meaning? Does it improve readability? Does it introduce new errors?

If a suggestion changes technical terminology, flag it. If a suggestion rewrites a sentence in a way that obscures your original argument, reject it. Do not accept a change just because it sounds better.

Step 4: Write Your Disclosure Statement

Your disclosure statement needs four pieces of information. Here is a sample you can adapt for any journal submission.

Sample disclosure statement (adapt this to your context):

“The authors acknowledge the use of [Tool Name, e.g., Grammarly] for language editing and [Tool Name, e.g., ChatGPT] for readability improvements during manuscript preparation. The tool(s) were used for [purpose: grammar correction, sentence restructuring, style adjustment]. Human authors reviewed and verified all AI-generated edits against the original draft and accept full responsibility for the accuracy and integrity of the final manuscript.”

Place this statement in: the Acknowledgments section or the Methods section, depending on your target journal’s preferred location.

This statement satisfies Elsevier, Springer Nature, Wiley, Taylor & Francis, the ICMJE, and The Lancet simultaneously. It is not thesis-level disclosure; it is journal-level compliance.

Step 5: Submit With the Declaration Attached

If submitting to an ICMJE-member journal, paste your disclosure into the cover letter and the manuscript. If submitting to a Wiley journal, use their submission declaration form as described in their AI guidelines. If submitting to a Taylor & Francis journal, include the statement in the manuscript body as required by their AI policy.

Check the “Instructions for Authors” page on your target journal’s website. Some journals have journal-specific wording requirements. The four core elements do not change; the framing may.

The same logic applies at Elsevier, where authors must insert their declaration immediately above the reference list. And Springer Nature’s policy requires disclosure through their green/amber/red framework, which you can find at group.springernature.com.

Common Mistakes That Lead to Compliance Violations

Researchers consistently make the same errors when using AI editing tools. These mistakes result in returned manuscripts, editorial inquiries, and sometimes formal investigations. The most common compliance failures involve misunderstanding the distinction between exempt and required disclosure, or failing to document generative tool use even when the tool is permitted.

Mistake 1: Assuming “Basic Editing” Means “No Disclosure Needed”

Grammarly is broadly exempt. ChatGPT is not. Using the same tool on the same manuscript does not trigger the same compliance requirement. If a paragraph passes through Grammarly and a separate paragraph passes through ChatGPT, you disclose only the ChatGPT use. But the reverse is also true: if you use ChatGPT on one paragraph and Grammarly on the rest, the ChatGPT paragraph is subject to disclosure regardless of what Grammarly did elsewhere.

Mistake 2: Copy-Pasting a Disclosure Statement Without Tailoring It

Every publisher expects tool-specific information. A generic statement that says “AI tools were used for editing” does not satisfy any journal. You must include the tool name, version, purpose, and verification step. Generic statements are the most common cause of returned disclosures. The Lancet’s declaration form requires the exact name of the model and version, not a generic reference to “a large language model.”

Mistake 3: Using ChatGPT for Entire Sections Without Reviewing the Output

The PLOS One study found that ChatGPT over-edited roughly 39% of its suggestions. Accepting entire rewrites without review introduces stylistic changes, alters meaning, and creates compliance risk. If you cannot verify the output, you should not use it. The study also found that ChatGPT removed 10 pieces of key information from one manuscript, including citations and table references. This is not a hypothetical risk. It is a documented outcome.

Mistake 4: Uploading Confidential Manuscripts to Public AI Tools

Springer Nature’s policy explicitly prohibits uploading unpublished manuscripts into unsecured or public AI systems. Even if Grammarly exempts you from disclosure, you must verify that the tool’s terms of service protect your unpublished data. Grammarly’s privacy policy states it does not sell user data or share it for model training. ChatGPT’s free version does not offer the same guarantee. Always check the tool’s data handling policy before submitting, not after.

Where to Disclose: Cover Letter, Manuscript, or Both?

The ICMJE’s dual-disclosure rule is the clearest signal. If your target journal is an ICMJE member—which includes most biomedical and many general-science journals—disclose in both locations. If the journal does not follow ICMJE, check its author guidelines. Some journals accept a single declaration in the manuscript body; others require a separate form at submission.

The Lancet’s guidelines explicitly state that AI disclosure belongs in a separate declaration statement at the end of the manuscript. Springer Nature prefers the Methods section or Acknowledgments. Wiley uses a submission declaration form. Taylor & Francis places the statement in the manuscript body.

The location varies. The content does not. And in every case, the content must include four elements: tool name, version, purpose, and verification. Nothing else is sufficient.

A practical approach: start by checking the journal’s “Instructions for Authors” page. Most major journals now include a dedicated section on AI policy, often linked from their ethics guidelines. If no dedicated section exists, search for “AI” or “generative AI” in the site’s search bar. Most journal sites have updated their author guidelines to address AI use. If you still can’t find guidance, contact the editorial office directly. It is always better to ask upfront than to risk non-compliance after submission.

How This Compares to Literature Review Guidelines

You might be wondering whether the rules for literature reviews differ from editing rules. They do not. The same disclosure principles apply across all manuscript preparation tasks. Using ChatGPT to summarize a literature review requires the same disclosure as using ChatGPT to edit a results section. Both are generative tools, and both require documentation.

For a deeper look at how to use ChatGPT ethically across different academic tasks, see our guide on the ethical use of ChatGPT for literature reviews. The principles of transparency, verification, and disclosure carry across the entire writing process.

If a reviewer has flagged your language quality and you need to choose an editing path, our action plan for language editor comments covers the full decision tree: self-edit, AI first pass, or professional editing, plus the disclosure requirements for each option.

How International Teams Rely on AI Editing

The need for language editing is not evenly distributed. A 2026 study analyzing 135 AI declarations across 8,633 articles published in arXiv journals found a highly significant association between team composition and AI-use purpose (p = 0.0008). International research teams placed significantly more emphasis on grammar assistance (31.8%) compared to non-international teams (13.2%).

This statistic matters because it explains why journal disclosure policies are becoming stricter. International teams, particularly researchers in low-income countries, are the primary users of AI editing tools. They rely on these tools to overcome the language barrier in academic publishing. The compliance gap is not just an ethical concern; it is a participation barrier.

The PLOS One study addressed this directly. Its authors, working with Ugandan researchers, found that professional editing services cost nearly half of the average salary for early-stage researchers in Colombia. AI tools like Grammarly and ChatGPT are free. They democratize access. But they also require compliance documentation. The equity gap in editing is real; the compliance gap is the next one.

The ethical implications go beyond compliance. The PLOS One authors found that ChatGPT’s content filter blocked editing of passages that included words related to women’s health and reproductive topics. This is not a minor glitch; it is a systemic barrier that disproportionately affects researchers in fields like sexual and reproductive health. The same tool that democratizes editing for some researchers blocks editing for others. That is a second equity problem that publishers are only beginning to address.

What To Do If Your Draft Was Already Submitted Without Disclosure

If your manuscript has already been submitted and you realize you used ChatGPT or another generative tool without disclosure, the safest action is to email the editorial office immediately. Most editors will allow you to submit a correction or an addendum. The ICMJE treats nondisclosure as misconduct only when it is intentional. Good-faith corrections are the standard expectation.

Do not wait for the editor to flag the issue. Proactive disclosure is always treated more favorably than caught nondisclosure.

The Bottom Line

AI editing tools are permitted across every major publisher. The rules are not about banning tools; they are about documenting them. Grammarly is broadly exempt. ChatGPT requires disclosure. The 2026 ICMJE update requires dual disclosure in both cover letter and manuscript. The Lancet’s guidelines require three specific elements in every declaration.

Follow the workflow: choose your tool, document your intent, verify every edit, write a statement that includes tool name, version, purpose, and verification, and submit with the declaration attached. Your manuscript will pass compliance screening, your editor will not flag you, and your peer-review process will start on time.