A Review of Grammarly:
Nine Years of Not Noticing

The hardest thing about reviewing Grammarly is that I never think about it. Nine years in, I have no ritual around it, no opinion I reach for, no moment where I decide to use it. It is simply present in whatever box I am typing into, and when something is wrong there is a red line under it.

That turns out to be the entire product, and the reason it survives.

Three and a Half Years Before I Paid Anything

My subscription page says my first payment was 3 June 2020. That is not when I started using it.

My email archive has Grammarly messages going back to 7 October 2016, including an "upgrade to Grammarly Premium" prompt — which they only send to people who already have an account. There is no welcome email left, so the real signup may be earlier still. Grammarly itself does not show an account start date anywhere.

So: roughly three and a half years on the free tier, and it was fine. I was using it for spellchecking and nothing else, and I never looked at what else it did. Which, I'll admit up front, is exactly the accusation I level at other people later in this review.

What It Has Actually Cost

The gap between June 2023 and September 2023 isn't a lapse — it's the handover, when work took over paying. I'd assumed for a while that the cheaper $79.20 charge was some kind of win-back discount, because that's a pattern I've been caught by elsewhere. It wasn't. It was just the transition.

Which leaves the obvious problem with me reviewing this at all: I haven't paid for it myself since June 2022. So let me answer that directly. If work stopped paying tomorrow, would I buy it back at $144 a year?

Yes. Without much thinking about it. The price went from $139.95 to $144 along the way and I'd pay the increase too. The next section is why.

The Wave

When the writing is going well, an idea moves forward at a steady pace — like a boulder rolling, or a wave you're riding. Everything is present at once: the thing you meant to say, the knowledge you already had, the shape it wants to take. You are not creating something amazing. You are creating exactly what was in your head, which is a different and rarer thing.

And every interruption puts you behind it.

Fix a spelling. Check whether that comma is right. Go and look up the date something launched. Verify a company fact. Each one is small. Handle enough of them and you look up, ready to sprint and catch back up, and the wave is gone. Not ahead of you — gone. You can't see it, you don't know where it went, and all you have left is the memory of what it felt like to be inside it.

Anyone who has made anything knows that feeling, and knows it's the worst one there is.

That is what I pay Grammarly for. Not to be the best editor available — it isn't. To remove a category of interruption without making me leave the sentence I'm in. I glance up, see red, hover, accept, and I'm still in the wave.

It is the same reason I dictate these reviews into a voice recorder instead of typing them. Anything that slows down getting the idea out risks losing the idea.

Every Single Day for a Year

I pulled 363 consecutive days of activity records straight out of Grammarly's analytics endpoint, from 29 July 2025 to 31 July 2026. Three numbers:

It was running every single day for a year, without exception. I wrote something on three days in four. It changed something on one day in two.

The weekday breakdown is the part that made me laugh at myself. Monday is my heaviest day at 87%, and it declines gently across the week — to 62% of Sundays. There is barely a weekend in this data. That's what having a job and a website at the same time looks like on a chart.

The longest unbroken run of writing days was 56, from 21 December to 14 February. January 2026 was 31 days out of 31. The quietest stretch was November 2025 — nine days out of twenty-nine, including a full week where I apparently wrote nothing at all.

I want to be careful about what those numbers do and don't show. They're daily yes/no flags: did this account write today, did anything get improved. They aren't a measure of how much. There's a reason I'm being careful, and it's coming.

The Habit That Outlived Its Own Logic

Here is the specific thing I do, which I'd never examined until I sat down to write this.

When I need to write a business email — something where the tone matters, where I don't want to put things the wrong way — I open Outlook. Then I copy what I've written, paste it into Grammarly, use the tone controls to make it read properly business-formal, copy it back, and send it from Outlook.

Grammarly is already running inside Outlook. It has been the whole time. It is fixing the grammar in place as I type.

I leave an application that already has the tool in it, to go and use the tool, and then come back.

And I could do the same job in any frontier assistant — better, probably, and in one place. I know that. I've started to think that way more often. But nine years of muscle memory says email, go to Grammarly, fix it up, copy, paste, send, and muscle memory keeps winning.

That's the convenience argument proving itself by breaking its own rule. The habit is so established that it survives being less convenient than the alternative.

Where It Gets In My Way

I don't write long-form in Grammarly. I've tried. The problem is that I can't ignore a red underline — if I see something wrong, I want to fix it, so I stop, right-click, fix, right-click, fix. For a short message that's fine. For three thousand words it's the exact interruption I described earlier, delivered by the thing that's supposed to prevent it.

So I write in a plain text file. Head down, usually looking at the keyboard rather than the screen, blast the idea out end to end, and go back afterwards to clean it up — where Grammarly is waiting anyway, because it's in the text editor too.

Which means the part of the product Grammarly would most like me to use — their own editor, on their own site — is the part I actively avoid. I have no reason to go to grammarly.com. It lives everywhere else.

What I Go To an Assistant For Instead

Two things, and they're both after the writing rather than during it.

Validation. Anything that needs checking — a date, a figure, whether a thing I remember is actually true. That always means leaving the page, and no in-line tool is going to solve it.

Cohesion. Once I've finished, or hit a milestone, the useful question is: I've written a lot, but does it hold together? That's an editor's job, and it's genuinely valuable when the editor understands what you were going for.

So the division is clean, and it isn't really a competition. Grammarly works inside the wave. An assistant works after it. The only thing that would collapse that distinction is an assistant that watches you type where you already type, rather than one you have to visit — and even then, it would only cover writing, not the dozen other boxes I put words into every day.

The Numbers That Cannot Be True

Grammarly sells analytics as a reason to buy the expensive tiers — tie the data to business outcomes, measure the impact. So I pulled mine as raw JSON rather than reading it off the dashboard.

Across 279 daily records covering one year on one account, it reports:

• 45 days out of 279 — 16% — where the recorded session time is longer than a day. The worst is 27 July 2026 at 106.9 hours, which is four and a half times a twenty-four hour day.
• Three days where "sessions improved" is larger than "sessions total" — 233%, 150% and 133%. That's what's behind the dashboard chart drawing a point above 100% on an axis that runs to 120%.
• A member count between 2 and 6 on an account with one member. Never 1.
• A summary that disagrees with its own detail. The header reports 1,884 sessions; the daily rows add up to 3,768 — exactly double.
• A suggestion acceptance rate of 0.0%, reported alongside 1,154 sessions it says were improved.
• 1,734,076 words across 279 days. That's 6,215 words every single active day for a year, at a claimed average of 9.9 hours a day.

On 26 July 2026 it records 237,426 words and 64.3 hours of activity — inside a twenty-four hour day.

Two honest caveats. This is one account, over one year, from one endpoint. And a one-person organisation is an odd shape that may well hit code paths nobody tests. I'm not claiming every Grammarly customer's dashboard is wrong.

But I am saying that mine is, demonstrably, and that I only found out because I stopped reading the chart and looked at the numbers underneath it. Anyone buying a plan for the reporting should check theirs the same way — the export button and the browser's network tab are enough.

What survives the audit is the simple stuff. Days active — 279 — is corroborated three separate ways: this file, a second endpoint, and the console. A yes/no flag per day is hard to get wrong. It's the volume metrics that collapse, and those are exactly the ones a dashboard is bought for.

Why I Never Worried About the Privacy

Grammarly lives in every text box on both my machines. It sees everything I type. I self-host my own transcription specifically so voice notes never leave my hardware, and I've written a whole editorial about the cost of that choice — and for nine years it never once occurred to me to apply the same question here.

I know why, and it isn't carelessness.

Word had a dictionary file. It lived on your machine. When it didn't know a word you clicked "add to dictionary" and your copy got better. That's the model I formed in the nineties, and when Grammarly turned up doing a recognisably similar job, my brain filed it in the same drawer without ever asking the question.

A mental model from the previous generation of software carried forward and hid the change. That's worth being alert to generally — a lot of things that used to run on your computer now run on somebody else's, and they look identical from where you sit.

What I do have is a hard rule that predates any of this: nothing regulated or customer-related goes through Grammarly, or through any AI assistant. That material moves through encrypted channels and nowhere else. It's reflexive at this point.

One thing genuinely has changed, though. Grammarly raised a billion dollars in May 2025, and has since acquired an email client and a document platform, with the acquired founder now running the company. It's still privately held — but a profitable private company has less reason to go looking for new revenue in its data than one carrying that much capital and, plausibly, heading somewhere public.

Nothing in the terms changed. The incentive did. It doesn't make them worse than the frontier AI companies — it puts them on the same footing, and means I watch them the same way I watch everyone else. And it's worth noticing that a company whose whole proposition is being in every text box you use now also owns your email client and your documents.

Who This Is For

Not for you if your grammar is already immaculate and you take pride in it. You know who you are — you're the one correcting strangers mid-argument online. This tool has nothing to offer you. It might suggest a paragraph restructure. That's it.

Not for you if you want something to write things for you. That isn't what it's for, and if that's the job, a frontier assistant does it better for $20 a month.

It's for you if you want the rough edges taken off writing you've already done. Spelling, punctuation, structure, the sentence that says the thing but says it clumsily. None of that changes what you're saying — it makes what you're saying easier to take in. Sentences are just ideas. This makes the ideas go down easier.

And if you're buying it for the analytics, check the numbers before you trust them.