Grammarly Announces New Authorship Verification Tool

Last week, Grammarly announced its new tool, Grammarly Authorship, which it claims will help users and institutions spot AI-generated writing.

Grammarly is best known as a writing assistance tool. It helps writers, including students, proofread their work and can also recommend suggestions for style, tone, brevity and more. The service also offers a plagiarism detection tool as part of its core product.

However, Grammarly has come under fire in recent months for allegedly enabling or encouraging plagiarism. That’s because, in April 2023, the company introduced a generative AI tool to write text for authors.

This has made Grammarly very controversial in education circles. Some claim their work was erroneously flagged as AI writing without using Grammarly’s AI tools. Others, however, have encouraged the use of Grammarly, especially for students who are struggling with writing.

However, Grammarly hopes its new tool could mitigate those concerns by providing a new way to detect AI writing, particularly in student work.

The approach is fairly novel, but we will not know how effective it is until the product launches in September.

Still, it’s worth looking the approach and seeing why it could be a game changer.

How Grammarly Authorship Works

The idea behind Grammarly Authorship is relatively simple. Rather than trying to determine if a work is plagiarized or AI-generated after it’s finished, it instead monitors the writing process.

To do this, Grammarly Authorship connects with the author’s word processor, in particular Google Docs or Microsoft Word. Then, Authorship marks the text into one of four categories: Human-written, AI-generated, AI-modified or pasted.

It also detects if a section was typed with an “unnatural typing pattern” and how long it took to complete the work, including whether that is reasonable for the length.

To present the findings, Grammarly Authorship has three separate approaches:

1. Authorship Analytics: This is simply the tool’s analytics. It highlights the percentages from each category and other data about the writing.
2. Authorship Report: This provides the full text of the writing and uses color coding to highlight the various types of writing contained within.
3. Authoring Reply: This works similarly to versioning in that it replays the writing of the work. It shows how long the author spent on each section and provides text highlights.

The service also proactively reminds authors/students to cite content from external sources.

Grammarly says the tool complements Grammarly Originality, which provides more traditional plagiarism and AI detection.

All in all, it’s an interesting idea and one that may have some promise for editors and instructors.

An Interesting Approach to AI Detection

As we’ve discussed recently, detecting AI writing is fraught. Though AI detectors may be improving, they are still far from perfect, leading to false positives and negatives.

However, regular plagiarism detection has that same issue. A plagiarism scan, even by a high-end service like Turnitin, is often wrong. The difference is that, with a plagiarism scan, a human can look at the report and make the final determination. With AI detection, there is no insight into how or why the tool reached the conclusion it did.

This makes it difficult for instructors or editors to act on AI detectors’ findings. It would be problematic even if they were 99% reliable, which is unrealistic.

As such, the hunt has been underway to find a way to verify the findings of AI detectors. This has included both digital and analog solutions. Some instructors quiz students about a paper in the classroom, require non-written elements to an assignment, or check the document’s version history.

Grammarly Authorship works similarly to the version history approach but with much more information. However, it has one major limitation: The author has to have it enabled as they are writing.

This is likely why Grammarly is heavily targeting the educational sector with this feature. According to their press release, they already have partnerships with 3,000 educational institutions. Many of those institutions likely provide students with Google Docs or Microsoft Word, making it easy to install Grammarly on student accounts and mandate its use.

While this makes a great deal of sense, it does limit the usefulness of this tool outside of controlled environments like workplaces and schools.

If you can’t compel someone to use this tool, it won’t be able to verify the work at all.

Bottom Line

Grammarly says they will release Grammarly Authorship next month for Google Docs and later this year for Microsoft Word. When that happens, testers, including myself, can try it and see how effective it is.

Until then, we can only discuss the concept.

To that end, it is an exciting idea. Verifying a work as it is written opens up new ways to validate its authenticity. However, the limitation of running it on the author’s word process limited the scenarios where it will be helpful.

That said, a service like Grammarly could provide third-party verification that a work is AI-free. I could see sites and publications promoting such verification to AI-weary readers.

Though the service does raise some privacy issues, those are largely moot. Grammarly already reads and parses every word it checks. As such, Grammarly Authorship doesn’t raise any privacy issues that the core product has not already raised.

In short, I’m interested to see how effective this tool is. If effective, it could be a powerful but limited way to detect author misconduct.