Grammarly After the AI Pivot: A Tool in Search of a Purpose

In the spring of 2023, a product manager at a mid-sized tech company in Austin named Dana Weisberg opened her quarterly performance review and found a line item she had not expected. Her manager had flagged, under “areas for development,” a pattern of comma splices in her written updates. The feedback had been surfaced, the manager explained, by a new AI tool the company had licensed. That tool was not Grammarly. It was a generic LLM the team was using to summarize written communication, and the summarizer had noted, in passing, that Weisberg’s sentence-level construction was occasionally awkward. Weisberg had been a Grammarly Premium subscriber for four years. She canceled the subscription the next week.

This is the kind of moment that matters for a company like Grammarly. Not a splashy competitive launch, not a market-share chart, but a quiet substitution in a single user’s workflow. Grammarly was founded in 2009 in Kyiv by Max Lytvyn, Alex Shevchenko, and Dmytro Lider, originally as a grammar-checker aimed at students worried about academic integrity. It grew slowly through the 2010s, pivoted to a browser extension that inserted itself into every text field on the web, and by 2019 had raised at a billion-dollar valuation. By 2021, the valuation was thirteen billion. The product had become nearly ubiquitous among knowledge workers, writers, and ESL professionals. It was, for a while, one of the clearest product-market fits in consumer software.

Then ChatGPT shipped. The timing was brutal. In November 2022, a free tool arrived that could do everything Grammarly did — fix grammar, suggest rewrites, check tone, rephrase for clarity — and several things Grammarly could not, like drafting entire emails, summarizing long documents, or explaining why a sentence was wrong rather than just flagging it. The value proposition Grammarly had spent thirteen years building was absorbed, in a matter of months, into a broader category of text assistance that was commodifying fast. The question was not whether Grammarly would respond. The question was whether a response existed.

The response was GrammarlyGO, launched in 2023, which grafted a generative-AI layer onto the existing extension. It drafts emails. It adjusts tone. It rewrites paragraphs. It is, functionally, a wrapper around a large language model, with Grammarly’s distribution and typing-surface integration as the moat. The company has spent two years since then reframing itself. It acquired Coda in late 2024. It rebranded parts of the product as a “communication assistance” layer rather than a grammar checker. The leadership has talked publicly about the shift from correction to composition. The marketing suggests the company has a clear thesis about what it is becoming. The product does not yet make that thesis visible.

Here is the difficulty. Grammarly’s original proposition was specific and bounded. It caught your grammar mistakes. It was accurate enough, fast enough, and quiet enough to sit in the background of every document you wrote. The extension was load-bearing infrastructure for people who worried about the way their writing was received. A subscription was worth the price because the problem it solved was narrow and recurrent.

The new proposition is broader and blurrier. Grammarly now wants to help you write, and wants to do so in competition with ChatGPT, Claude, Gemini, Microsoft Copilot inside Word, Google’s Help me write inside Docs, and a pile of smaller AI writing tools. AI writing assistants for academic work have proliferated to the point that choosing among them has become its own small research project. The broader proposition has a harder time justifying a thirty-dollar monthly subscription when the user can get more capable generation from a free ChatGPT account or bundled generation from a Microsoft 365 license they already pay for.

What Grammarly still does that the generic LLMs do less well is the narrow thing. It sits inside every text field. It flags mistakes in real time as you type. It does not require you to context-switch to another tab and paste your paragraph into a chat. For users whose core problem is confidence in the grammatical correctness of short-form writing, emails, Slack messages, internal memos, this is still genuinely useful. The integration is the product, and it would take a significant engineering effort for any of the generic LLMs to replicate it across the range of browsers and native applications Grammarly covers.

The question is whether that narrow usefulness justifies a standalone company at a thirteen-billion-dollar valuation. The honest answer is probably not, and the company’s behavior since 2023 suggests the leadership knows this. The Coda acquisition was an attempt to expand surface area. The rebranding efforts are an attempt to reposition away from a category that is collapsing. Grammarly is trying to become a productivity platform because it cannot remain a grammar checker, and the productivity-platform market is considerably more crowded and considerably harder to win.

There is also a specific technical problem with Grammarly’s position that gets less attention than it should. The company’s original advantage was a large corpus of training data about writing mistakes, gathered over a decade of user submissions, combined with specialized models trained on that corpus. This advantage has been almost entirely erased by the scale of foundation-model training. GPT-4 and its successors have seen orders of magnitude more text than Grammarly’s entire training history. For the specific task of identifying whether a sentence is grammatical, a general-purpose LLM is now at least as accurate as the specialized tool, often more so, and can additionally explain its reasoning in ways Grammarly’s rule-based explanations cannot. The proprietary advantage is gone.

What remains is the distribution advantage. Grammarly’s extension is installed on roughly thirty million devices. The extension sees what users type. It has the right to suggest edits before the text is submitted. This is a genuinely valuable position, and it is not one that ChatGPT can easily replicate. A company with that distribution and a clear generative-AI thesis could build something meaningful. A company with that distribution and an unclear thesis risks becoming a widget that users install once and forget to cancel.

The user-research question Grammarly has to answer is what its core job-to-be-done actually is in 2026. When a writer encounters the need for help, do they want a silent sentence-level checker, a proactive writing coach, a draft generator, a tone adjuster, or a full AI assistant that can do all of these? The answer probably varies by user and by context, which is a hard product problem. The framing of AI as a partner rather than an outsource is especially relevant here, because the pedagogically responsible version of a writing tool is one that flags issues and prompts reflection, not one that drafts for you. Grammarly has historically been on the responsible side of this line. It is not clear the new version stays there.

For students, which is where Grammarly originally made its name, the question is sharper. A student who uses Grammarly to fix grammar in their own writing is getting the same kind of low-grade feedback they might get from a TA. A student who uses GrammarlyGO to rewrite their paragraph or draft their thesis statement is doing something different, and the line between the two inside a single extension is thin. Universities have spent the past three years trying to articulate policies on AI writing assistance. Grammarly has not made those policies easier to write.

Where this leaves Grammarly as a product is in a strange middle position. The old product is largely obsolete, but it still works and still has customers. The new product is unfinished and competes in a crowded category. The company has the resources to keep iterating and the distribution to matter if it figures out what it wants to be. It has not yet figured that out. Tools that survive category disruption usually do so by narrowing, not by expanding. Grammarly is currently trying to expand. Whether that works is one of the more interesting open questions in the AI-assistant market right now, and the answer will probably be visible in the renewal rates, not in the press releases.

Photo via Unsplash.