Grammarly Updates Authorship, Improves Labeling

Last month, I published an article entitled, “How Grammarly Launders AI-Generated Content.” The article examined Grammarly authorship and highlighted the confusing way it flagged content rephrased by Grammarly’s “Humanize” tool.

Shortly after the article’s publication, a representative from Grammarly reached out to me, saying they would address the issues I raised and that an updated version would roll out shortly.

Now, a month later, I can confirm that Grammarly has addressed the biggest issue with its Authorship tool. It no longer allows people to “launder” AI text using its “Humanize” tool. 

This is a substantial improvement that makes Grammarly Authorship much more effective at its stated goal. 

However, the tool still does have some notable shortcomings. Though Grammarly says those are being addressed, they have not rolled out as of this writing, leaving a few gaps in its coverage.

Recapping the Problem

Grammarly Authorship was first announced in August 2024. It was billed as a tool that works with either Microsoft Word or Google Docs to track how a document is written. By monitoring what was copied and pasted into the document, it could see when content was pasted in, either from other websites or various AI systems.

The goal was to fill a gap in the current writing landscape. AI detection remains imperfect, and even small amounts of AI polishing or editing can trigger AI detectors. However, since false positives are also an issue, albeit much less common, both students and teachers are seeking ways to prove how a piece was actually written.

This has led to the rise of controlled writing environments, such as Turnitin Clarity, that track how a work is written and allow a third party, such as a teacher, to see the document’s creation process.  

Grammarly Authorship is such an environment, but it works with existing word processors, namely Google Docs and Microsoft Word. 

However, the initial release had a significant issue. Any text that was run through Grammarly’s “Humanizer” AI tool would be listed in the report as “Typed by a Human.” Although the subhead indicates that it was “Rephrased with Grammarly’s AI,” the text would still be highlighted in green.

Grammarly has now fixed that issue, and it did so simply by changing how “Humanized” or otherwise rewritten text is presented.

The Fix

When showing how a report was drafted, Grammarly Authorship presents the text in three broad categories: “Typed by a Human,” “Copied from a Source,” and “Unknown.” Each category has three subcategories that break down how the content was created in more detail.

Previously, the subhead “Rephrased with Grammarly’s AI” was under the first category and “AI-Generated” was a subcategory of the second. Now, “AI-Generated” Reads “AI-Generated or Rephrased with AI.” Now, if you use the Humanizer tool, or any similar AI rewriting tool from Grammarly, it falls into that subcategory.

Previously, if you generated a bunch of AI text or brought that text in from another source and then used the humanizer or paraphrasing tool on it, the rewritten text would be marked as “Typed by a Human.” With this change, the rewritten text is marked as “Copied from a Source.”

So now, if you use Grammarly’s AI tools to rewrite AI text, it will remain labeled as 100% AI-generated.

Grammarly Authorship also tracks edits made through its spelling and grammar checking tools. However, those remain marked as “Typed by a Human.”

All in all, this is a much better balance, and it fixed the biggest issue with Authorship. However, there are still other problems that aren’t as easy to fix.

Into the Unknown

The third category of Grammarly Authorship is “Unknown”, this serves as a catchall for text that Grammarly Authorship can’t determine how it was inserted. 

This includes everything from text typed before Authorship was enabled, text pasted from applications other than the web browser, and, perhaps most importantly, text generated within the document itself.

When you open a new Document in Google Docs, it offers “Generate Document” using Google Gemini. If you do this, Grammarly Authorship tracks this as “Unknown” since the text was neither typed nor pasted. 

This is a technical limitation. Though Grammarly has said that they are working on this, it’s a much more difficult issue than recategorizing “humanized” text.

In the end, the best approach to “Unknown” text may simply be to treat it as it’s described, an unknown. It doesn’t prove anything either way. However, this requires the person reviewing the report to understand it, and, as we’ve seen with other plagiarism reports, that can be difficult to convey.

Bottom Line

Grammarly is in a difficult position. Grammarly, or rather, their parent company Superhuman, is an AI company. However, they are at least in part targeting academia, which is understandably wary of AI’s potential for cheating.

A tool like Grammarly Authorship can be very useful in an academic setting, especially for students who are doing legitimate work but are worried about being falsely flagged for using AI. 

To that end, if you are a student (or any other kind of writer) with that concern and you use Grammarly, I recommend turning on Authorship. As I discussed last month, it does a decent job of showing that a particular work was human-written, even if there are things that can be improved.

Other than the aforementioned technical issues, the next steps will be more educational and documentation. Although the company has created FAQs and other documents for educators, if someone receives an Authorship report out of the blue, there’s little guidance on how to read and interpret it.

Because if these reports are used correctly, they can provide a valuable data point when trying to determine how a work was created. However, they are also potentially misinterpreted. This makes clear instructions all the more critical.

Still, if I were a student in 2025, I would consider using Grammarly Authorship or a similar tool. Having evidence that a piece is human-written could be valuable. Though the concern over Grammarly is understandable, especially in schools, this is a tool that, if used correctly, can help both teachers and students fight cheating.