Every AI email tool sells you on time. Save 30 minutes a day. Reply 5x faster. Inbox zero in half the time.
This is true for almost all of them. It is also the wrong way to evaluate them.
Speed is table stakes. Any tool that lets you generate a draft instead of typing one will save you time, because typing is slow. The interesting question, the one that determines whether the tool is actually useful, is different:
How many of the drafts it produces would you send without reading?
Not glance at. Not skim. Send. The way you send a one-line "sounds good, talk Tuesday" reply. Click the button, look away, move on.
For most AI email tools, the honest answer is zero. You read every draft. You edit most of them. The "saved time" gets quietly clawed back by the proofreading tax.
Why send-without-reading is the only metric that matters
If you have to read every AI draft carefully before sending, you have not actually delegated the email. You have just changed the medium. Instead of typing the email yourself, you are reviewing somebody else's email and deciding whether to ship it. The cognitive load is similar. The risk is higher (you might miss something the AI got wrong). The relationship cost of a robot-flavored sentence reaching your client is real.
The promise of AI email was supposed to be different. The promise was that for the predictable 70-80% of your inbox (status updates, scheduling, follow-ups, acknowledgments, straightforward requests), the AI would draft a reply that you could trust to represent you. You would skim, occasionally tweak the last sentence, and send. The mental loop closes in 8-10 seconds, not 90.
That requires confidence. And confidence requires something almost no AI email tool actually delivers: output that is reliably indistinguishable from what you would have written yourself.
Why most AI email tools fail the confidence test
Three reasons, in order of how often they bite.
1. The output sounds like ChatGPT, not like you
This is the dominant failure mode. The draft is grammatically excellent, structurally clean, and tonally generic. It contains the linguistic markers that tip recipients off that the email was AI-drafted: the over-polished sentence cadence, the unnecessary "I hope this email finds you well," the em-dash that you never use. You spot it immediately because you know your own voice. So you edit. The send-without-reading rate is zero.
2. The tool does not know the thread
Most AI email tools start cold on every draft. You paste the original email, you write a brief about what you want to say, you get a response. The AI has no awareness that you and this person have been emailing for two years, that the last call covered a specific topic, or that the previous three exchanges established context this reply needs to honor.
So the draft is competent but contextually shallow. You read it carefully because you have to verify that nothing important was missed. You edit because the AI greeted them like a stranger. The send-without-reading rate stays at zero.
3. The tool does not know the relationship
Even if the tool reads the thread, it does not know that this client prefers short replies, that this colleague hates bulleted lists, or that you have an inside reference with this contact that has appeared in every third email between you for two years. Without that, the draft cannot match the texture of the relationship. It is technically right and relationally wrong, and you can feel the difference in a single read.
What it takes to clear the confidence threshold
Two architectural things, neither of which is solved by "a better prompt."
The tool has to read your sent email history, not just your prompt. The difference between describing your voice ("warm but professional, prefers short paragraphs, casual sign-off") and showing the AI your last 200 sent emails is the difference between a sketch and a photograph. Voice description converges on a stylistic average. Voice evidence captures your actual patterns: the specific openers you use with new contacts vs. ongoing ones, the closers you reach for when you are pressed for time, the sentence length distribution that makes a paragraph feel like you wrote it. This is the architecture behind embeddings-based voice matching, and it is the one path to a draft that reliably reads as yours.
The tool has to live inside the email client. Not because in-inbox saves seconds, though it does. Because the AI needs to read the thread automatically, surface the right historical context, and condition the draft on the actual conversation rather than on a brief you wrote. Tab-switching to a chat window is what forces you to re-establish context every time, and that re-establishment is what makes the AI miss things that a tool with native thread access would not.
Speed is the floor; quality is the ceiling
Every AI email tool will save you minutes per email. That is the floor. The thing they are competing for is the ceiling: how often the draft is good enough that you can actually let go.
When the ceiling is high enough that you trust the tool with most of your inbox without re-reading, the time savings compound in a way that "5x faster typing" cannot. You stop checking the AI. You stop second-guessing the draft. You stop carrying the email in your head as a thing you still have to think about. The mental space that email used to occupy frees up.
That is the promise. Speed is just the price of entry.
ForthWrite optimizes for the ceiling
ForthWrite reads your sent email history (encrypted, isolated to your account), stores it as embeddings, and conditions every draft on the relevant past correspondence. It runs inside Gmail and Outlook, so the thread context is always present. The result is drafts that read recognizably like you on most everyday email, which is the only way the send-without-reading number gets above zero.
Not every draft. The first few weeks before the system has enough data are the weakest. Genuinely novel topics where your sent history provides thin guidance are still going to need an edit. But for the high-volume, predictable email that fills most inboxes, the goal is for you to stop reading the drafts and start trusting them.
The first 10 drafts per week are free, no API key needed. Try it on an email sitting in your inbox right now. The honest test is whether the draft is good enough that you would send it without rewriting.
Build a persona prompt that captures how you actually write →