Most AI writing tools forget a correction the moment you close the window. ReplyPen writes down what it learns, in a place you can read, so a fix you make once shapes the next Draft too.
What is in Memory?
Three parts, each in plain language:
| Part | What it holds | Where it comes from |
|---|---|---|
| Business profile | What your business does, who it serves, your role, and the facts every reply should get right | Your website, read during setup, and what you tell ReplyPen |
| Tone profile | How you write: greeting, sign-off, length, formality | The replies you have already sent |
| Knowledge log | Facts and answers ReplyPen picked up, each with the day it learned it | Your sources, your edits and your Feedback |
Nothing in Memory is hidden from you. If ReplyPen uses it to write a Draft, you can read it.
What does “learns your tone” actually mean?
It means ReplyPen studies the replies you already sent and writes down how you write: how you greet, how long your answers are, how formal you are and how you sign off. That description is the Tone profile. During setup you see a preview of it (“Here’s how you sound”) before the first Draft.
Every Draft follows the Tone profile, in the language your customer wrote in. It is a description of your style, not a set of templates, so replies read like you without repeating the same phrases. For why canned wording puts customers off, see reply templates that don’t sound canned.
What happens when you edit a Draft?
When you change a Draft before sending it, ReplyPen notices the difference between what it wrote and what you sent, and learns from the change. What it learned goes into the Knowledge log, so you can see it. The same happens when you give Feedback on a Draft (“too long”, “we don’t offer that”) and regenerate it: ReplyPen remembers what you told it.
This is also how the answers to repeat questions settle. The first time a customer asks something new, you fix the Draft, and ReplyPen learns from the change so the next similar email goes better. More on that in handling repeat customer questions.
How do you correct something it learned?
Open Memory, find the fact or the trait, and change it. A wrong price in the Business profile, a sign-off you never use in the Tone profile, an outdated answer in the Knowledge log: each is a line of plain text you can correct.
If the wrong fact comes from one of your sources, such as an old help article, fix the source as well. The Check note under each Draft shows which source a reply relied on, so you can see where a mistake came from.
Is Memory used to train AI models?
No. Memory belongs to your account and is used to write your Drafts. The AI providers we work with are not allowed to train their models on your email. See where your data goes.