Learn AI Reply Writer — guides & blog.
Step-by-step instructions to get the most out of every reply, plus tips and updates from our blog.
Guides & instructions
Getting started — install the extension & sign in
Add AI Reply Writer to Chrome from the Chrome Web Store, pin it to your toolbar, then sign in with your account. Pinning matters more than it sounds: the toolbar icon is where you check which plan you are on and how much of this month’s allowance is left, and an unpinned extension hides both behind the puzzle-piece menu.
Signing in once is enough
Your plan, your remaining replies and your language preference live on your account rather than in the browser, so they follow you to a second computer the moment you sign in there. Nothing is stored in the extension itself, which is also why signing out on one machine does not disturb the other.
If you use more than one Chrome profile — a personal one and a work one, say — sign in on each. They are separate browsers as far as an extension is concerned.
You do not need the extension to start
The AI Chat on this site uses the same models and the same writing controls, so you can try the product before installing anything. It is also the better place for a long piece: a cover letter, a proposal, a post you want to shape over several passes.
The extension earns its place when you are already inside somebody else’s composer and do not want to leave it — which, on a busy day, is most of the time.
Write your first reply
Open any composer — an email, a direct message, a comment box — and click the “✦ AI Reply” button that appears beside it. It reads the conversation that is on screen, matches the tone the other person is using, and drafts a reply. Review it, change anything you like, then insert and send.
It reads the thread, not just the last line
The draft is written from the conversation as a whole, so a reply to “sounds good, what time?” knows which meeting is being talked about and what was already agreed. That is the difference between a reply that moves things forward and one that asks a question the thread has already answered.
When the newest message is your own, it writes a follow-up instead of a reply — a polite nudge that does not pretend somebody answered you.
Nothing is ever sent for you
The draft lands in the composer and stops there. You read it, edit it, and press send yourself. There is no setting that changes this, deliberately: the moment a tool can send on your behalf, every draft has to be right rather than merely good, and that is not a promise worth making.
Choose a model
Pick the right brain for the job from the model menu. The regular models are fast and available on every plan, including the free one. The advanced models are slower, noticeably better at nuance and long context, and draw on the advanced credits a paid plan gives you.
When the faster one is the right one
Short replies, scheduling, acknowledgements and anything where the answer is obvious are better on a regular model — you get the draft in about a second and you were going to edit one line of it anyway.
Save the advanced models for the messages that are actually difficult: a complaint you need to answer carefully, a proposal, a reply in a language you do not write fluently, or a thread long enough that the details matter.
One subscription, no keys
Every model in the menu is billed through your plan. You never paste an API key into the extension, and there is no separate account to keep topped up with any provider.
Control tone & length
Two menus shape every draft. Writing style sets the register — Professional, Casual, Friendly, Persuasive or Concise — and Length sets how much of it there is: Short, Medium, Long, or Auto, which matches the length of what you are replying to.
Auto is the one to leave on
Auto reads the thread and answers a one-line message with one line. A fixed Medium on a chat app produces a paragraph where a sentence was expected, which is the single most common way an AI draft announces itself as one.
Set a fixed length when you know better than the thread does — a first outreach message that needs room, or a reply you want kept deliberately short.
Style is a starting point, not a cage
You can also type an instruction into the box before generating: “politely decline”, “ask for the deadline”, “agree but push the call to Thursday”. The style menu sets the voice; the instruction sets what the message has to achieve, and the two work together.
Reply in another language
The language menu decides what the draft is written in, independently of the language the conversation is in. Left on Auto it matches whatever the other person is writing, which is what you want for an inbox that receives messages from anywhere.
Answering in a language you do not write
Set the language explicitly when you want to answer in the other person’s language rather than your own — a buyer writing to you in Spanish, a supplier in German. The draft comes back in that language, and you still read it before it goes anywhere.
Scripts that read right to left — Arabic, Hebrew, Persian, Urdu — are written and displayed in the right direction, in the composer and in the panel.
Give it context it cannot see
Knowledge is the place to put the things that are true about your work but are nowhere in the conversation: what you sell, your delivery times, the name you trade under, the two things you never agree to. It is attached to every draft, so you write it once.
What belongs in it
Short, factual, and the kind of thing you find yourself typing repeatedly. Prices you quote, the turnaround you promise, the way you spell your own product. Facts, not instructions about tone — tone belongs in the style menu, and mixing the two makes both weaker.
Leave out anything you would not want in a draft. The knowledge you save is sent with the request that writes your reply, which is exactly why it makes replies more accurate — and exactly why a password does not belong in it.
Pictures in the thread are read for you
When somebody sends a screenshot, a receipt or a photo of a product, it is examined and described before the reply is written — so the draft can answer what is actually in the picture instead of talking around it.
Have replies read aloud
Any draft can be read out in a natural voice, which is more useful than it sounds when you are working through a long inbox: you listen to the reply while your eyes are on the next message, and you catch the sentence that reads oddly far faster than you do on screen.
Where the allowance goes
Spoken replies are counted separately from written ones, so listening to a draft never spends the allowance you were saving to write with. Running out of one does not stop the other.
Your device’s own voices remain available whatever your plan says, and cost nothing.
Invite friends & earn free replies and voice chats
Open Refer & earn, press Share Now, and send your link. Your 1st friend earns you 5 free replies, every friend after earns 10 — and every friend earns you 5 voice chats. There is no limit on how many people you invite.
It is a permanent raise, not a top-up
Both rewards are added to your monthly allowance and stay there, so the tenth invitation is still worth what the first one was. A one-off bundle of credits would be worth nothing by the following month; this is not that.
A referral counts when the person you invited creates a genuine new account through your link. Inviting yourself, or an account that has already been referred, does not count — the programme has to stay worth running.
Plans & credits
Free gives you 20 replies a month on the regular models, which is enough to find out whether the product belongs in your day. The paid plans unlock unlimited regular replies and add advanced credits.
What a credit actually buys
A regular reply and an advanced reply are not the same unit of work, which is why they are counted separately. Regular replies are unmetered on a paid plan. Advanced credits are spent when you pick an advanced model, and the heavier the model the more of them one draft costs — so the credit figure is an approximation of replies, not a promise of a fixed count.
Elite is $14 a month with about 2,000 advanced credits. Ultra is $29 with about 5,000, and is the one that makes sense for an agency or anyone drafting all day.
Allowances reset, they do not roll over
Everything resets at the start of each billing period. Unused replies and credits do not carry forward and have no cash value, which is worth knowing before you save them up for a busy week that never comes.
Troubleshooting basics
Most problems are one of three things, and all three have a quick check.
The button is not there
Refresh the page after signing in — a tab opened before you signed in does not know you have. Then check the extension is enabled and pinned, and that the site is one of the ones it supports. On a single-page app the composer sometimes mounts late; giving it a second and clicking into the message box brings the button with it.
The draft will not generate
Wait a few seconds and press it again. When a provider is busy or down, the request moves to the next model automatically rather than failing, and that takes a moment longer than usual. A draft that never arrives at all is usually a sign the allowance for that kind of reply has run out — the panel says so when it has.
The reply is right but the tone is wrong
That is a settings answer, not a bug. Change the writing style, drop the length to Concise, and put the thing you actually want into the instruction box. If it keeps missing the same way, the fix usually belongs in Knowledge — it is answering without a fact it needed.
Still stuck? Support → Report issue sends it to us with the details we need, and we answer.
From the blog
5 ways to reply faster without sounding like a robot
Speed and warmth are usually treated as a trade-off. They are not — most slow replies are slow because of the blank page, and most robotic ones are robotic for reasons that have nothing to do with how quickly they were written.
Lead with the answer
The single biggest improvement to any reply is putting the answer in the first sentence and the context after it. People read the first line and decide whether to read the rest; a reply that opens with three lines of preamble gets skimmed no matter how good the rest of it is.
Let the length match the message
A paragraph in answer to a one-line chat message reads as effort in the wrong place. Leave the length on Auto so a short question gets a short answer, and override it only when you have a reason.
Match their register, not your habits
If they wrote “can you do fri?” then “Friday works — shall we say two?” is the right reply and “I am writing to confirm my availability on Friday” is not. The draft follows the tone of the thread by default, which is usually more casual than the one people default to when they start from nothing.
Say what happens next
A reply that ends without a next step makes somebody else decide what it was. Finish with the concrete thing — the time, the file, the question — and the thread moves forward instead of going round again.
Add one line that only you could write
This is the one that does the work. A sentence about the thing they mentioned, the deadline you know is tight, the holiday they are about to take. It takes five seconds and it is what makes the rest read as a person rather than a template.
None of this is about typing faster. It is about not starting from nothing — and then spending the time you saved on the one line that matters.
Free vs Elite vs Ultra — which plan fits you?
The plans differ on two axes, and only two: how many replies you may write, and how good the model writing them is allowed to be. Everything else is the same product.
Free
20 replies a month on the regular models, no card. It is a genuine trial rather than a demo: the same reading of the thread, the same tone controls, the same languages. If you reply for a living you will run out; if you are deciding whether this belongs in your day, it is enough to decide with.
Elite
$14 a month. Unlimited regular replies — so the counting stops — plus about 2,000 advanced credits for the messages that need the better model. This is the plan for one person with a busy inbox, and it is the one most people stay on.
Ultra
$29 a month, with about 5,000 advanced credits and the reasoning models. The case for it is volume and difficulty together: an agency answering for several clients, a seller handling a marketplace inbox all day, anyone whose hard messages outnumber their easy ones.
How to choose without guessing
Start free and watch where you run out. If it is the reply count, Elite solves it. If it is that the drafts are fine but not good enough on the messages that matter, that is an advanced-credit problem and the jump to Ultra is the one worth making. Upgrading and downgrading both take effect from the Pricing page, and you keep what you have paid for until the period ends.
How AI Reply Writer handles your data
A tool that reads your inbox has to be clear about what it does with what it reads. This is that, in plain terms.
Your conversation goes to the model, and nowhere else
When you press the button, the conversation on screen is sent to the AI provider so a reply can be written, and the draft comes back to you. It is not sold, it is not used for advertising, and we do not keep the content of your request once your result has been returned.
You never handle a key
The provider keys live on our server, not in the extension and not in your browser. That is why a single subscription covers every model in the menu, and why there is nothing for you to rotate, store or leak.
Nothing is published without you
The extension writes into the composer. It does not click send, and it has no way to. Every message that leaves your account was read and sent by you — which is also the honest answer to “what if it writes something wrong”: you will see it first.
What you choose to add
Anything you save in Knowledge is sent with the requests that write your replies, because that is what makes them accurate. Treat it as you would a note stuck to your monitor: useful facts about your work, and nothing you would mind a supplier seeing in a draft.
The full detail — what is stored, for how long, who processes it and the rights you have over it — is in the Privacy Policy, and it is written to be read rather than to be survived.
Need a hand? Visit the Help Center or head back to the app.