How Grammarly Fails to Understand Human Writing
A deep dive into the software's limitations
There’s a chance you, like me, are inundated with Grammarly’s ads everywhere you browse. They pop before videos, occupy Google AdSense spaces, and slide into inboxes. It’s surprising how much money a grammar checkup tool can spend on marketing.
Grammarly promises to help you produce “mistake-free, bold, and clear writing.” To get there, and according to a 2021 Reuters report, the company has raised $200 million at the valuation of $13 billion, all in the service of their AI-powered* editing software. In a Bloomberg interview, the Ukrainian founders said their immigrant background gave them “a unique perspective on the complexity and power of language in helping people reach their goals.”
It’s an accurate statement. Language is indeed complex and powerful—especially when it comes to writing. There are hundreds of small decisions I’m making while writing this piece. These decisions will affect how you, the reader, perceive me. You could leave this page right now and I’d be financially hurt by whatever negative impression my words have left on you. Nerve wracking.
And in a couple of minutes or hours, you’ll assume my role: you’ll write something to someone, hoping it’ll sway them to do what you want, fearing it might rub them the wrong way—or worse, not rub them at all.
That’s why Grammarly and my newsletter exist. Despite having different approaches, we both aim to improve your writing. The biggest difference is that Grammarly, a 600-employee company founded 13 years ago, asks you to let a bot handle it: according to their landing page, the software flags grammatical mistakes, makes writing bold and clear, and provides style and tone suggestions.
Yet does Grammarly deliver on these promises? The answer is an emphatic no.
Whenever I used it, Grammarly left a lot to be desired: it ignored some of the mistakes I made and changed some of the writings I was proud of.
But let’s not use my writing as an example; I could be a self-centered individual who wouldn’t fault himself. To see how Grammarly outright botches the task of editing—wreaking havoc on not only accurate, but also beautiful writing—we’ll paste some of the best prose I could find—lines carefully written by recognized writers and extensively revised by respectable editors—into the app.
The results were—and I’m sorry for typing this cliched adjective—jaw dropping.
We’ll start with a paragraph I personally adore. It’s a brilliant text from Technics and Civilization by Lewis Mumford:
The new attitude toward time and space infected the workshop and the counting house, the army and the city. The tempo became faster: the magnitudes became greater: conceptually, modern culture launched itself into space and gave itself over to movement. What Max Weber called the "romanticism of numbers" grew naturally out of this interest. In time-keeping, in trading, in fighting men counted numbers; and finally, as the habit grew, only numbers counted.
What do you think about this piece? Any issues with it? I don’t know about you, but Grammarly found 9 suggestions—including 6 you can access with a $30 monthly membership.
Here are some of Grammarly’s suggestions:
(1) counting house: It appears that counting house is missing a hyphen. Consider adding the hyphen(s).
Awful mistake. Grammarly, for an unknown reason, considered counting house a compound adjective, thus suggesting a hyphen.
But how? Where’s the noun this “compound adjective” modifies? Even assuming a parser’s logic, counting house is followed by a comma and a determiner (the), which means there’s no way it could be an adjective. Counting house is, simply, a compound noun.
(2) and: it seems that this sentence contains a series of three or more words, phrases, or clauses. Consider inserting a comma to separate the elements.
Grammarly here refers to the second and in “the workshop and the counting house, the army and the city.”
Two things. First: no. You shouldn’t add a comma because “the army and the city” are two words, not three.
Second: Grammarly, If you think I should add a comma before the second and, why didn’t you flag the first one as well? According to the app’s computation, the workshop and the counting house should satisfy the same requirements for flagging.
(3) conceptually: It seems that conjunction use may be incorrect here. Consider adding and.
I’m still scratching my head at this one.
Why on earth would you add and?
Here you find three rhythmic sentences connected by a unified sphere of meaning—hence tied with two colons. So how is this suggestion accurate, not to mention relevant?
You might argue this passage is too clever for Grammarly: whoever writes like Mumford shouldn’t need a grammar tool. Let’s insert then a popular article from The Economist. Why is everyone so busy? is one of the finest essays I’ve ever read—the writer(s) and editor(s) didn’t put a foot wrong.
Let’s paste the whole piece and see what Grammarly thinks.
And… it went insane: the app detected 79 “mistakes”, graded the piece as “unclear”, and, overall, gave it a quality score of 75 out of 100.
Here’s one of the sentences that didn’t set well with the tool:
Ever since a clock was first used to synchronize labour in the 18th century, time has been understood in relation to money.
What seems to be the problem, officer? In relation to. It’s “wordy.”
Any suggestions? Sure, consider changing it to about. So it becomes “…time has been understood about money."
I’ll let you try each suggestion and see how the new sentence reads—I’m sorry for the impending aneurism.
Most of the feedback I received from Grammarly for my text or others’ has been flawed, confusing, irrelevant, or entirely wrong.
Back in college, I used Grammarly infrequently just to make sure I didn’t miss an apostrophe, so it was surprising to come back and see the little progress they have made.
To be clear, I don’t doubt their dedication or effort, but the results I’ve shown here speak to how incredibly hard it is for software to grade, and improve, human writing. It also proves that writing is so much more than some arbitrary, surface features.
What makes this even more problematic is the app’s garish claims: on the landing page you find a bolded headline that reads Great Writing, Simplified. The Features section touts a piece of software that will help “an entire organization communicate with skill and confidence.” Start using the app, however, and you find a completely different reality: the premium suggestions, cleverly promoted within the editor to speak to the user’s fear of missing out on important suggestions, are mostly useless. The explanations provided for the highlighted “errors” are hardly relevant. And by giving judgments irresponsibly, the tool doubles down on writers’ insecurities about their writing.
Some might argue Grammarly’s job is simpler than that: the editor flags spelling and basic grammar errors, and that’s what the people want. Knowledge workers write basic sentence structures and make predictable grammatical errors, and they’re just looking for a tool to remind them that they forgot to capitalize Wisconsin.
Two problems with this argument. One: grammar and meaning are inextricably linked. If the app doesn’t get your point on a human level, grammar will surely suffer. There will be things it’ll get right and others, not so much. And you’re not paying $30 a month for a 70% accuracy rate.
Two: approaching writing (and editing) the Grammarly way severely inhibits its transformative potential: Following Grammarly’s edits blindly limits your creative freedom and dulls your text. It makes everyone sound the same.
The subtle nuances of human expression, the ones that go undetected by grammar tools, are what makes great writing—and using Grammarly confirms this: there's a chasm between understanding the structure of language and grasping its soul.