Since the inception of mainstream AI content generation back in 2022, there has been a constant clamor to find a solution to people passing this content off as their own, human-written work.
While other brands have used the term ‘AI content detector’ as a marketing ploy for a poor-at-best product, Originality.ai has always strived to provide the most in-depth and detailed tool possible in an attempt to bring honesty and transparency to the world of AI content detection.
However, that doesn’t mean that false positives don’t occur from time to time, which can cause significant issues for brands, writers, students, and many more.
From our discussions with users, many of these issues seem to stem from the use of tools like Grammarly, which is why we decided to put it to the test ourselves.
Grammarly Plagiarism Checker Review
Can Grammarly Detect AI Writing
Does Using Grammarly Get Content Detected as AI Generated?
Grammarly is a writing assistance tool that helps users improve their writing by checking for grammatical errors, punctuation mistakes, spelling errors, and style issues.
Additionally, it provides insights into sentence structure, vocabulary usage, tone, and readability, aiming to assist writers in crafting clear and compelling content.
It’s a brilliant tool, and one that can greatly support writers in any vocation, whether it be book writing, marketing, or academia.
Grammarly also provides rephrasing suggestions, using it’s AI to identify a more common way of structuring a particular sentence, or a way that is less wordy, or one that is more suitable to your selected target audience.
A false positive refers to a situation in which a test or evaluation incorrectly indicates the presence of something that is not actually there or incorrectly identifies a condition as being present.
For example, you have written a piece of content entirely on your own, only to see it flagged as AI-generated when you run it through Originality.ai.
This is an issue that some of our users were experiencing, whereby their original content would pass through Originality’s detectors with no issues, but the version that had been run through Grammarly for amends first was flagging as AI-generated.
To test this theory and identify what specific element of Grammarly could be causing the problem, we decided to do our own testing. After all, Grammarly isn’t just a spell-checker, it also offers AI-generated rephrasing suggestions and wording changes, so it’s important to identify which aspects are causing false positives to occur.
To do this, we gathered ten known human IELTS essay answers as our benchmark and ran each of those pieces of content through Originality Standard 2.0 for clarity (note: this model is retired, for details on our latest model visit our AI Detector Accuracy Review). We then proceeded to break down our amends into two different categories:
Lightly edited: With this version of the content we only accepted Grammarly grammar suggestions, and not any rephrase or rewrite suggestions.
Heavily edited: With this version, we accepted all Grammarly insights, including rephrasing and rewriting suggestions.
You can find the data for our test here.
As you can see from the results, light edits resulted in almost all of the examples continuing to display as human-written, whereas the heavy edits had a significant impact on whether or not the content was flagged as AI-generated.
This makes sense when you consider that the more you rely on the AI suggested edits and phrasing changes from tools like Grammarly, the less the work reads like your original draft, and the more of it is based on AI suggestions.
While this is only a small sample size, it gives us a good indication of how much impact overly relying on these tools can have on your AI content detection score.
To clarify Originality.ai’s content policy on Grammarly and AI detection, we’ve included a brief update to this article as of June 20th, 2024.
Please note that as Grammarly releases new features and as our machine learning engineers at Originality.ai continue to improve accuracy and lower false positives, this policy may change and be updated at a later date.
False positives occur when copy or content that a person writes is incorrectly identified as AI-generated.
To lower the risk of false positives when using the Originality.ai detector:
For additional information on reducing false positives, read these articles:
Grammar and spelling suggestions from Grammarly aren’t considered AI writing. The grammar and spelling correction features are editing tools that are unlikely to trigger AI detection.
Although grammar and spelling suggestions are unlikely to trigger AI detection, Grammarly’s Rephrase, Rewrite, and Use our best version — prompt AI detection.
On Grammarly’s features page, they note that they integrate ‘generative AI’ to power certain features. Although AI can be an excellent tool for improving writing skills, it’s important to be aware that if you edit with Rephrase, Rewrite, and Use our best version, it could trigger AI detection.
To avoid issues, it’s best practice to clarify a client’s guidelines on the use of AI-powered writing and editing tools like Grammarly at the beginning of a project.
Additionally, Grammarly also recommends providing proper citations when using AI rewriting tools, such as their Paragraph Rewriter.
Grammarly has launched its own AI detector which is reviewed here.
So, in summary, when users of Grammarly use the "Rephrase" or "Rewrite" suggestions, it is likely to trigger AI detectors to identify the content as AI-written. If it is used for spelling, not so much.
That is why it’s essential always to be transparent with what tools you use and maybe even save an original version of the document before using Grammarly to make any changes. That way, if any issues arise, you can provide the original document for clarity.
Have you seen a thought leadership LinkedIn post and wondered if it was AI-generated or human-written? In this study, we looked at the impact of ChatGPT and generative AI tools on the volume of AI content that is being published on LinkedIn. These are our findings.
We believe that it is crucial for AI content detectors reported accuracy to be open, transparent, and accountable. The reality is, each person seeking AI-detection services deserves to know which detector is the most accurate for their specific use case.