Somewhere between the drafting of an email and the clicking of send, a strange new transaction is taking place. AI tools like Superhuman and Grammarly are no longer just fixing your typos or suggesting better word choices. They’re learning how you write — your cadence, your quirks, your tendency to sign off with “Best” instead of “Cheers” — and then generating entire messages in your voice. The result is a communication layer where the person receiving the email may never know whether a human or a machine composed it.
That’s the pitch. And it’s working.
Superhuman, the premium email client that charges $30 a month, has leaned hard into AI-generated drafts that mimic a user’s writing style. Grammarly, the ubiquitous writing assistant with tens of millions of users, has moved well beyond grammar correction into full-blown AI composition. Both companies are betting that professionals want to spend less time writing and more time doing — and that AI impersonation, for lack of a better term, is the feature that will get them there. But as these tools become more sophisticated, an uncomfortable question is surfacing among the very technologists building them: what happens when nobody can tell who actually wrote anything anymore?
On a recent episode of The Verge’s Vergecast, the hosts put Superhuman and Grammarly through rigorous testing, examining how well each tool could replicate a user’s personal writing style. The verdict was mixed but telling. Superhuman’s AI drafts, trained on a user’s sent messages, produced replies that were eerily close to how the person actually writes. Grammarly’s approach was broader, offering tone adjustments and full rewrites that felt polished but sometimes generic. Both tools, the reviewers noted, had crossed a threshold: they weren’t assistants anymore. They were ghostwriters.
Shishir Mehrotra, the CEO of Coda and a longtime product thinker in Silicon Valley, joined the same episode to discuss the broader implications. His framing was sharp. Mehrotra argued that AI writing tools are creating a new kind of social contract problem. When you receive an email from a colleague, there’s an implicit assumption that the person wrote it. If an AI composed the message — even one trained on that person’s style — does the communication still carry the same weight? Mehrotra didn’t offer easy answers, but he was clear that the industry hasn’t grappled seriously enough with the question.
He’s right.
The technology itself is advancing at a pace that makes philosophical hand-wringing feel almost quaint. Superhuman’s “Write with My Voice” feature analyzes patterns across hundreds or thousands of a user’s previous emails to build a stylistic profile. It picks up on vocabulary preferences, sentence structure, even the way someone handles pleasantries versus getting straight to business. The AI then uses that profile to generate first drafts of replies that, in many cases, need only minor tweaking before they’re sent. For power users who process hundreds of emails a day, the time savings are substantial.
Grammarly has taken a slightly different path. Its AI features, including the GrammarlyGO assistant launched in 2023, focus on adapting tone and formality across different contexts. A user can ask Grammarly to make an email sound more confident, more diplomatic, or more casual. The tool can also generate entire drafts from bullet points or brief prompts. While it doesn’t build as deep a personal voice model as Superhuman, Grammarly’s scale — over 30 million daily active users as of its most recent disclosures — means its impact on professional communication is arguably larger.
And the competition is intensifying. Google has integrated its Gemini AI into Gmail, offering smart compose features that go far beyond the predictive text of a few years ago. Microsoft’s Copilot, embedded across Outlook and the broader Microsoft 365 environment, can draft emails, summarize threads, and suggest responses. Apple announced Apple Intelligence features for Mail that prioritize and summarize messages, with writing tools that can rewrite text in different tones. Every major platform is converging on the same idea: your email shouldn’t require you to actually write it.
The business logic is straightforward. Email remains the dominant communication channel in professional settings, with estimates from the Radicati Group suggesting over 360 billion emails will be sent daily in 2025. Most of those emails are routine. Scheduling. Follow-ups. Brief acknowledgments. The argument from companies like Superhuman is that AI should handle the routine so humans can focus on work that actually requires thought. It’s a compelling pitch, especially for executives and salespeople who live inside their inboxes.
But the cultural implications are thornier than the product demos suggest.
Consider the trust dimension. Professional relationships are built, in part, on the texture of communication. The way a manager responds to a difficult question. The warmth — or lack of it — in a client’s follow-up. These signals matter. When AI generates those signals, calibrated to sound exactly like the sender, it creates a kind of emotional forgery. The recipient responds to what they believe is a genuine human expression. It isn’t. Not entirely.
This isn’t a hypothetical concern. Researchers at Cornell University published work in 2024 examining how people perceive AI-generated messages versus human-written ones. Participants consistently rated AI-generated messages as less sincere once they learned the origin — even when they couldn’t distinguish the two in blind tests. The implication is that disclosure matters, but almost no one is disclosing. There’s no “Sent with AI” tag on Superhuman drafts. No asterisk on Grammarly-composed emails. The tools are designed, by intent, to be invisible.
Mehrotra’s point on The Verge’s podcast touched on this directly. He drew an analogy to earlier waves of automation in communication — spell check, autocomplete, even the shift from handwritten letters to typed ones. Each transition, he noted, initially felt like a loss of authenticity. People adapted. But Mehrotra also acknowledged that AI-generated writing is qualitatively different from autocorrect. It’s not fixing what you wrote. It’s writing what you would have written. That distinction, he suggested, deserves more scrutiny than it’s getting.
The legal and regulatory angles are starting to come into focus as well. The European Union’s AI Act, which began phased implementation in 2024, includes transparency requirements for AI systems that interact with humans. Whether an AI-composed email sent by a human user falls under those provisions is still being debated, but the direction of travel is clear. Regulators in Brussels are thinking about this. In the United States, the conversation is less advanced, though the FTC has signaled interest in AI-generated content that could be deceptive in commercial contexts.
For enterprise buyers, the calculus is different. Companies adopting Superhuman or Grammarly at scale are primarily measuring productivity gains. And the numbers are persuasive. Superhuman claims its AI features save users an average of four hours per week. Grammarly’s enterprise tier, Grammarly Business, markets itself on communication consistency across teams — ensuring that every customer-facing email meets a certain standard of clarity and professionalism. For a sales organization with 500 reps, that consistency has real revenue implications.
So the tools are selling. Fast.
Superhuman raised $75 million in a Series C round in 2023, valuing the company at $825 million. Grammarly was valued at $13 billion in its 2021 funding round and has continued to expand its enterprise business. Both companies are profitable or near-profitable, a rarity in the AI sector where burn rates remain staggering. Their success reflects a genuine market demand: professionals want help with email, and they’re willing to pay for it.
The competitive dynamics are worth watching. Superhuman’s advantage is depth — it builds a richer model of individual writing style because it controls the entire email client experience. Grammarly’s advantage is breadth — it works as a browser extension, a desktop app, a mobile keyboard, and integrates with virtually every writing surface a professional touches. Google and Microsoft have distribution advantages that dwarf both, but their AI writing features are more generic, optimized for the average user rather than the power user.
There’s also a philosophical split in how these companies talk about what they’re building. Superhuman leans into the idea of augmentation — the AI as an extension of you. Grammarly frames its tools around communication effectiveness — the AI as a coach that makes you better. Google and Microsoft tend to emphasize efficiency — the AI as a time-saver. These aren’t just marketing differences. They reflect genuinely different product philosophies that will shape how AI writing evolves over the next several years.
The most provocative question, though, may be the one that none of these companies are eager to answer publicly. If AI can write emails that are indistinguishable from human-written ones, and if those emails are sent without disclosure, what is the status of email as a medium? Does it become a protocol for AI-to-AI communication, with humans merely approving or rejecting drafts? And if so, why are we still writing emails at all?
Some in the industry are already pushing toward that endpoint. Startups like Lindy AI and Relevance AI are building autonomous agents that can handle entire email workflows — reading incoming messages, drafting responses, scheduling meetings, and following up — with minimal human oversight. These aren’t writing assistants. They’re replacements for the act of writing altogether. The human becomes a supervisor, reviewing AI-generated correspondence the way a manager reviews a junior employee’s work.
That future may be closer than most professionals realize. But it raises a final, uncomfortable truth that Mehrotra alluded to in his conversation with The Verge. Communication isn’t just about transferring information. It’s about signaling attention, care, and presence. When a friend writes you a long, thoughtful email, part of what you value is knowing they spent time on it. When a client sends a carefully worded response to a sensitive negotiation, the care in the language is itself a form of commitment. AI can replicate the words. It cannot replicate the intention behind them.
Whether that distinction matters to the market is another question entirely. Right now, the market is voting with its wallet, and the wallet says: write my emails for me. Superhuman, Grammarly, and every major tech platform are happy to oblige. The technology is impressive. The productivity gains are real. And the slow erosion of authentic human communication in professional life continues, one AI-drafted email at a time.
Nobody’s sending a disclosure notice about it.
WebProNews is an iEntry Publication