ForthWrite Review - An AI Email Writer That Learns Your Actual Voice

Welcome to this ForthWrite review ✨

Every AI email writer makes the same promise and breaks it in the same way. You paste a thread, you get back something fluent, competent, and completely unlike anything you have ever written. It says "I hope this email finds you well." It says "I'd be happy to." You spend two minutes rewriting the thing you asked a machine to write for you, and you wonder what exactly you saved.

ForthWrite's pitch is that the problem is the training data. Generic tools learn from the internet's idea of a professional email. ForthWrite learns from your sent folder, then keeps score by comparing what it drafted against what you actually sent, and treats the gap as a signal. That is a meaningfully different bet, and it is the one thing I wanted to test properly.

So I connected a real Gmail account, ran the onboarding end to end, let it read my sent mail, generated drafts on real threads, and ran its own benchmark against my own history. Below: what the voice matching actually does, what drafting in Gmail is like, where the product reaches beyond the inbox, what it costs, and the parts that are still rough.

Getting Started & First Impressions

There is no app to learn. You sign up free, add a Chrome extension, connect a mailbox, and the product lives inside Gmail and Outlook on the web. A dashboard exists at forthwrite.ai for the settings, analytics, and the power tools, but the daily surface is your inbox.

The onboarding is where I first saw the idea working rather than just being claimed. After connecting Gmail, ForthWrite reads your sent mail and then asks a short set of persona questions: your name, your company, your role, your industry. Standard stuff. The interesting part is the Advanced panel, which is not asking, it is showing you what it already worked out.

Default tone came back pre-selected as Casual, tagged "Suggested from your sent email." Under a section titled "Phrases you never want to see," it had already ticked "per my last email" on my behalf. The free-text box beneath it was pre-populated with two more it had inferred: "I hope this email finds you well" and "Just wanted to reach out." My sign-off was filled in from my actual mail. None of that is a questionnaire. It read my sent folder and came back with a first draft of me, and it was broadly right.

Then it spends about 25 seconds building a system prompt from the persona answers plus a sample of your sent mail. The result was 759 words and specific enough to be uncomfortable:

Your tone is casual and direct, sometimes blunt or playful, but always terse. Sentences are short, often just a few words. Paragraphs are short and punchy, never dense.

It had also written per-type playbooks I never asked for. On customer support: "direct answers first. Acknowledge the issue in one line if needed, then solve it or point to the fix. Skip apology-heavy language, you rarely apologize." That last clause is an observation about me, drawn from my mail, and it is accurate. That is the moment the product stopped feeling like a wrapper.

The prompt is fully editable, which matters, because it is not flawless. One line reads "Do not use contractions is not required either way, match the clipped, plain style shown in the profile," which is a template merge that went wrong and never got proofread. It is cosmetic, but it sits in the artifact that drives every draft you generate, so it is a strange thing to leave in.

Voice Matching, And How Honest It Is About Itself

This is the headline feature, so it deserves the scrutiny.

The mechanism has three parts. A voice profile built from your sent mail, refreshed roughly every 50 captured emails so it tracks your voice as it drifts. Smart retrieval, which pulls your most relevant past replies into the context when drafting a new one, so it has live examples rather than a description. And edit supervision, which watches the diff between the draft it gave you and the mail you actually sent.

What I did not expect is that ForthWrite will grade itself against your history, on demand, without sending anything. In Prompt Lab there is a Benchmark button that re-drafts up to 100 of your most recent sent emails using the active prompt and scores each draft against what you really sent. I ran it. It tested 22 samples and returned a Voice Match of 56%, a Lab Score of 20%, and a distribution of 1 excellent, 14 good, 7 needs work.

Now, the honest reading of that number: it is not a verdict on ForthWrite. My test account's sent history was mostly throwaway mail with subjects like "hello", "ag" and "ergregreg", and the tool had 11 training samples to work with. You cannot learn a voice from that, and the product says so, repeatedly and without spin. The Training Data page rated my voice profile Apprentice and told me plainly it was at "0% to Trainee at 25 captures." Voice signatures showed 0 patterns with the note "We typically need ~10 edited drafts before one shows up here." Prompt Lab's Optimize feature was locked outright: "you're at 0/50 sent pairs."

I want to be fair about what that means in both directions. It is genuinely refreshing that a paid AI tool ships a button that tells you it is currently at 56% and shows its working, instead of a marketing number. Most tools in this category would never let you see that. But it also means the promise has a real cost of entry: ForthWrite is not good on day one, by design and by its own admission. It needs roughly 25 captures to reach the next tier, 50 sent pairs before it can self-optimise, and ~10 edited drafts before coaching says anything. If you send five emails a week, you are looking at a long runway before the thing you paid for turns on.

There is also a bug worth flagging here, because it undermines the feature that is supposed to be the guardrail. Prompt Lab showed me a "Voice profile mismatch" warning: "Your captured voice never uses I hope this email finds you well, Just wanted to reach out, per my last email, and 2 other phrases, but the active prompt asks for them." The active prompt does not ask for them. It explicitly forbids them, in a sentence that begins "Never use." The detector is matching on the string and missing the negation, so it is flagging the prompt for containing the very rule that protects you. Harmless today, but Optimize and Auto-Pilot use that same profile as a guardrail when proposing prompt changes, which makes a false positive there more than cosmetic.

Drafting In Gmail, And What It Saves You

Open a thread, hit Reply, and a ForthWrite widget docks into the compose box: a tone preset dropdown, a box for a context-specific prompt, a microphone for dictating that context instead of typing it, and a Generate Draft button.

I tested it on a real thread, a Booking.com message about a special request on a hotel reservation. The draft came back in around 25 seconds:

Hello,

The reservation is now over (checkout was July 16). No action needed on my end.

Thanks for the updates either way.

That is a better answer than I expected, and the reason is not the prose. It read the checkout date buried in the booking table, compared it to today, worked out the thread was moot, and said so in three lines instead of writing me a polite paragraph about a request that no longer mattered. The marketing calls this "context-aware replies" and it is doing more work than the phrase suggests. It also obeyed the persona: short, no em dashes, opened with "Hello," which is one of the openers the prompt had inferred from my mail.

The catch: the inbound email was in French and the draft came back in English. That is not supposed to happen. ForthWrite has a Drafting language setting whose default is literally "Match inbound (default) — Reply in the same language the sender wrote in." Mine was on the default. It did not match inbound. There is fine print admitting extension drafts can diverge from the proxy here and that "extension parity is on the roadmap," which suggests the team knows.

The fix was quick, and it showed me the best small feature in the product. The widget switches after generation into a refine mode: "What would you like to change?" I typed "Reply in French, the sender wrote in French" and got back:

Bonjour,

Le séjour est terminé depuis le 16 juillet, donc ce n'est plus un problème.

Merci pour les mises à jour.

Same reasoning, same brevity, correct language, and a Draft 2/2 counter appeared with arrows to flick between versions. Refining in plain language beats rewriting by hand, and keeping both drafts to compare is the kind of detail that only shows up when someone actually uses their own product.

On the time-saved claim, I have to be straight with you: I cannot confirm it from one session. The Performance page tracks Time Saved as a headline metric and mine read 0min after a single draft, which is honest arithmetic. The mechanism is plausible and the drafts were good enough to send with light editing. But "saves you hours" is a claim that needs a month of real sending to earn, not an afternoon, and I would rather tell you that than pretend otherwise.

Beyond Gmail And Outlook

This was pitched to me as the differentiator, since most rivals stop at Gmail and Outlook, so I went looking for it. The picture is more partial than the pitch.

Slack is real but not connected. There is a proper Slack card in the dashboard: "Draft replies to Slack DMs and mentions without leaving the app," with a Connect Slack button and instructions to right-click any message in a connected workspace and choose "Draft with ForthWrite." I could not test it end to end, and my understanding from the team is that public distribution is still pending Slack's approval, which means for most people it is not yet a thing you can just switch on.

LinkedIn I could not verify at all. There is no LinkedIn surface anywhere in the dashboard, no card in Connections, and no mention on the marketing site. It may well ship inside the extension on linkedin.com, but I could not confirm it in this session and I am not going to tell you a feature exists because I was told it does.

Outlook is real and shipping, alongside Gmail, both on the web. That is the honest current state: two surfaces live, one built but gated, one I could not find. Worth knowing if the multi-surface story is why you are considering it, because today the product you would actually buy is a Gmail and Outlook product.

The Wider Product

Plenty else worth knowing, most of it Pro-tier:

• Auto-draft and batch. Replies get written in the background as mail lands, so drafts are waiting when you open Gmail. A "Generate Drafts" button parks itself in the corner of your Gmail inbox to clear dozens of threads in parallel.
• Ten AI providers. OpenAI, Claude, Gemini, Grok, DeepSeek, Mistral, Perplexity, Cohere, LibreChat and Plusvibe, switchable per account. The site says nine; the dashboard lists ten. No other email tool I know lets you pick your model.
• Prompt Lab. Version control for your system prompt, with a Benchmark, a Compare view, Duplicate, and Push to extension. My account had two versions after onboarding alone.
• Training data management. Every sample it has learned from is listed individually, with an Export button, so you can walk away with your corpus. ForthWrite also says you can wipe the lot with one click, which I did not test.
• Recipient-aware and intent-aware drafting. Tone adapts to who you are writing to; a follow-up is treated as a different job from an intro.
• Voice dictation, so you can speak the context of a reply rather than type it.
• Still cooking: the dashboard Inbox is marked SOON, Notifications are COMING SOON, and Optimize and Auto-Pilot stay locked until you hit 50 sent pairs.

Pricing & Plans

Four tiers, and the logic is unusually clean.

Free is $0 and more generous than it needs to be: 10 drafts a week, no card, no API key, and crucially voice matching is included. The thing that makes ForthWrite different is not paywalled, which is a confident choice. You lose the custom persona, auto-draft, analytics and dictation.

Standard is $15/mo billed annually ($180/yr), or $18 monthly. This is the tier for someone who just wants good email: a sharper model, custom persona prompt, recipient and intent-aware drafts, smart retrieval, voice dictation, unlimited drafts, and unlimited connected Gmail and Outlook inboxes.

Pro is $30/mo annually ($360/yr), or $36 monthly, and it is where the interesting machinery lives: auto-draft, batch, edit supervision, performance analytics, Prompt Lab, training data export, and a premium frontier model. Everything I found most compelling to test was Pro.

Teams starts at $59/seat/mo annually ($79 monthly), two seats minimum, with a shared team persona, a shared Prompt Lab library, team analytics and admin seat management. Enterprise adds SSO/SAML, SCIM and audit logs.

Two things to actually understand. Paid plans start with a 14-day free trial, no card, and Teams gets 30 days. And BYOK is a discount, not a tier: bring your own OpenAI or Anthropic key, pay that provider directly, and take ~17% off. Feature-for-feature the two variants are identical, and the pricing FAQ says so plainly instead of upselling you. On top sits a 60-day "sound like you" guarantee: if the drafts do not sound like you, they will do a 30-minute call to tune your voice profile, and refund everything if it still is not clicking. Given that the product's whole thesis takes weeks to prove, a 60-day window rather than a 14-day one is the right shape.

Who Should Use ForthWrite?

It is a strong fit if you are:

• Someone who sends a high volume of professional email from Gmail or Outlook and edits every AI draft you have ever generated.
• A founder, exec or account manager whose voice is part of the job, where a generic-sounding reply costs you something real.
• Anyone who wants to choose their model, or already has an API key and would rather pay the provider directly.
• The kind of person who will actually look at a voice-match score and tune a prompt, because Pro rewards that.

It is a harder sell if you send a handful of emails a week. The product needs volume to become itself: 25 captures for the next profile tier, 50 sent pairs to unlock Optimize, ~10 edited drafts before coaching speaks. Low senders will pay for months of a tool that is still learning. It is also Chrome plus Gmail or Outlook on the web only, so desktop Outlook or Apple Mail people are out, and if you are buying it for Slack and LinkedIn today, buy it for the email instead.

Conclusion

That's the end of this ForthWrite review. What won me over was not the drafting, which is good but which several tools do adequately. It was the onboarding handing me back "you rarely apologize" as a fact about my own writing that it had inferred from my sent folder, and it was Prompt Lab volunteering a 56% score against my own history instead of a marketing number. A tool that measures whether it sounds like you, shows you the number, and tells you it is not there yet is a tool making a falsifiable claim. That is rare enough in this category to be worth something.

What holds it back is that the claim takes weeks to pay off, and a few edges are unfinished: a broken sentence sitting in the generated prompt, a mismatch detector that cannot read its own negations, a Connections page and a Settings page that disagree about whether my Gmail is linked, and a language setting that did not do what it said. None of it is fatal. All of it is the kind of thing you notice in an hour.

Try the free tier first. Ten drafts a week with voice matching included is enough to find out whether the thing that comes back sounds like you, and that is the only question that matters here. If it does, Pro is what turns it into a system.

What I liked:

• Voice matching is included on the free tier. The differentiator is not paywalled.
• Onboarding reads your sent mail and hands back a specific, editable 759-word prompt, not a questionnaire result.
• Benchmark grades drafts against your real sent history on demand, without sending anything.
• The product is honest about not being ready yet: profile tiers, locked features, and a visible score.
• Refine-in-plain-language with version history beats hand-rewriting a draft.
• Ten model providers, switchable. BYOK is a straight discount, not a worse plan.
• A 60-day guarantee sized to how long the thing actually takes to work.

Things to keep in mind:

• It is deliberately not good on day one. Low-volume senders will wait a long time to see the value.
• Slack is built but gated on Slack's approval; I could not find a LinkedIn surface at all. Today it is a Gmail and Outlook product.
• My French thread got an English draft despite the language setting defaulting to "match inbound."
• Small unfinished edges: a garbled sentence in the generated prompt, a false-positive mismatch warning, Settings and Connections disagreeing on inbox status.
• Chrome plus Gmail or Outlook on the web only. No desktop mail clients.

If you have ever rewritten an AI draft to sound like yourself and wondered what the point was, ForthWrite is aiming squarely at you.