Give your chatbot its own email address so it answers customers who write in, and replies in the same thread.
The email channel gives your chatbot an address of its own. Anyone who writes to it gets an answer drawn from your data sources, in the same thread, usually within a minute. Every exchange lands in Chat logs beside your website and help page conversations.
It suits the mail that already arrives at support@ or hello@: the repeat questions your docs already answer. Forward those to your agent's address and let it take the first pass.
Turn it on
Open your chatbot dashboard, switch to Deploy, and select the Email card. Click Set up.
Switch Enable email channel on.
While it is off, mail sent to your agent is still saved to Chat logs — it is simply never answered. That makes it a safe way to watch what arrives before you let the agent reply.
Under Mailbox → Addresses, copy Your agent's address. It looks like:
agent@k7m2xq9pbr4t.denser-mail.comThe subdomain is issued once and does not change. What sits in front of the
@ is yours — support@, sales@ and billing@ at the same subdomain all
reach the same agent, so you can publish a different address per audience
without setting anything up.
Send it a question from your own mailbox. A reply should arrive in the same thread, and the conversation should appear in Chat logs with an Email label.
When you are happy with the answers, forward your real support address to the agent's address, or publish the agent's address directly.
Forwarding is the usual setup
Most teams keep their existing support@ address and add a forwarding rule to it, rather than asking customers to write to a new address. Every mail provider supports this — in Google Workspace it is Settings → Forwarding, in Microsoft 365 it is Mail → Forwarding.
Mailbox
Addresses
| Field | What it does |
|---|---|
| Your agent's address | The address that receives mail. Issued automatically, permanent, and safe to publish. |
| Reply addresses | Addresses the agent replies from, each added through the setup below. If mail arrived at one of them, the reply comes from that same one; otherwise the first listed. With none set up, replies come from the agent's own address. |
| Deliverability alert contact | Where we tell you about mail that did not reach the customer. See Delivery alerts. |
Replying from your own address
By default replies come from agent@….denser-mail.com. To have them come from support@yourcompany.com instead, add it under Mailbox → Reply addresses → New address. It takes three steps, and the address is not used until all of them are done.
Name the address. Type the address your customers already write to. The preview shows the From: line replies will carry.
Forward it to your agent. In your email provider's settings, add your agent's address as a forwarding address, then press Verify automatic forwarding.
We send a test message to your address and watch for it to arrive at the agent. There is no way to check a forwarding rule from outside a mailbox, so this is a real end-to-end test rather than a stored setting.
Where the confirmation code goes
Most providers confirm a forwarding address by mailing it. That mail arrives at your agent, so the code or link is in this chatbot's Chat logs.
Authenticate the domain. Add the DNS record shown, then press Validate.
One record is required — a TXT record holding a DKIM public key. It authorises us to send as your domain and signs each reply in a way that aligns with it, which is what stops recipients filtering mail that claims to be from you.
Two more are shown as recommended. MX and SPF records on a mail. subdomain align SPF as well; a _dmarc record at p=none asks recipients to report on mail claiming to be from you. Neither is needed to start sending.
DNS can take up to 72 hours to propagate. The records stay on this screen until they resolve, and Validate re-checks them.
Replies from a configured address carry a Reply-To pointing at the agent, so a customer's follow-up comes back to the agent and the conversation continues.
Removing an address also revokes our authorisation to send as its domain, once no other address is using it.
One account per domain
A domain can be authenticated by one Denser account. If yours is already set up elsewhere — another workspace, a colleague's account — adding it here is refused. Contact support if it belongs to you.
Filtering
| Field | What it does |
|---|---|
| Blocked senders | Mail from these is saved to Chat logs but never answered. Enter a full address to block one person, or a bare domain (newsletter.example.com) to block everything from it. |
| Spam detection | Suspected spam is saved without a reply. |
Mail that is saved rather than answered appears in Chat logs with a line saying why — the channel was off, the sender is blocked, it was judged spam, or it was sent automatically. Suspected spam is capped at five messages per sender per day, since the agent's address is public and storing everything would let anyone fill your log.
Spam detection uses the receiving server's own verdict first, which is free and immediate. Only when that verdict is inconclusive does the AI read the message and decide — and when it is unsure, it answers. A wrongly answered piece of spam costs one reply; a wrongly ignored customer looks exactly like the agent being broken.
Mail carrying a virus is discarded outright, whatever these settings say. Mail to an address that no chatbot owns is discarded too, so publishing the agent's address does not expose you to traffic aimed at made-up addresses at the same domain.
Mail the agent never answers
Machine-sent mail is recognised and left alone, with no setting to turn it off:
- Out-of-office and vacation replies are saved to Chat logs, not answered. Answering one is how a loop starts: the reply lands back in the mailbox that sent it, which replies again.
- Newsletters and mailing-list mail are saved, not answered.
- Bounce reports and read receipts are discarded. They are a mail system's note about a message you sent, not a message from anyone — delivery problems reach you through the alert contact instead.
Forwarded mail is not treated as automatic, so pointing your existing support address at the agent works exactly as described above.
Composition
| Field | What it does |
|---|---|
| Display name | The name on the reply, e.g. Acme Support. Defaults to the chatbot's name. |
| AI composition disclaimer | Adds a short note that the reply was written by AI. Edit the wording to match your tone. |
| Email signature | A rich-text signature added to the bottom of every reply. |
The preview beside the form shows how a reply will look with these applied.
AI Settings
The AI Settings tab is the same model and prompt configuration as every other channel: LLM model and AI prompt. Changing it here changes it everywhere, and the dialog says so before it saves.
Citations are not offered here. A citation is a footnote whose link a mail client cannot resolve, so email answers never carry them however the setting is left — the assistant is asked to name a page in a sentence instead. The toggle stays on Customization, where it governs the channels that do render them.
If you want email answers to read differently from your website chat — longer, more formal, signed off — say so in the AI prompt rather than maintaining a second one.
Every email becomes a ticket
Email is a ticketing medium in a way live chat is not: the customer has already left, and the answer has to survive being read later. So every message the assistant answers opens a ticket in the Helpdesk, whichever way it was answered, carrying the sender and the thread.
Mail that arrives but is not answered — filtered as spam, sent by a machine, or classified as something the assistant should not touch — is still stored, so nothing silently disappears.
Review replies before sending
By default the assistant answers on its own. Turn on Review replies before sending on the AI Settings tab to put a person in front of every reply instead.
With it on:
- The assistant still writes the answer, immediately.
- Nothing is sent. The message is held as a draft.
- A ticket waits in the Helpdesk for someone to approve it.
Only the sending is withheld — the drafting still happens the moment the mail arrives, so approving is a read-and-click rather than a write.
Which mode to run
Review mode is the safer default while you are still learning what your content covers. Once the drafts stop needing edits, turning it off is what makes the channel actually save time.
How threads work
A reply carries the identifiers a mail client needs to file it under the original message, so a customer's follow-up stays in the same thread — and in the same Denser conversation. Everything a person writes from one address is grouped as one visitor, however many threads they start.
Start a new thread and you get a new conversation, which is what a fresh question should be.
Delivery alerts
An answer being written is not the same as an answer arriving. A conversation in Chat logs shows the reply either way, so a message that never landed is invisible unless something tells you.
Set Deliverability alert contact and we email that address when:
- The send fails. The reply never left our servers — most often because it was addressed from a domain we are not authorised to send as. The answer is in the conversation log; you can reply by hand.
- The message bounces. The recipient's provider refused it. A permanent bounce means the address does not exist; a temporary one means we retried and gave up.
- The recipient reports it as spam. Stop replying to that address. Complaints count against sending reputation, and enough of them pause sending for everyone on the domain.
Each alert names the recipient, the reason the receiving server gave, and the original subject. Leave the field empty and none are sent — the failures are still recorded on our side, but nothing reaches you.
Limits
- Attachments are not read. A message with one is answered from its text; the attachment is ignored.
- Replies are text and HTML. The agent does not send attachments back.
- One address per chatbot. Run a second chatbot if you need a second address.
- Automatic mail gets no reply. See Mail the agent never answers.
- Quoted history is trimmed. A long reply chain is cut back to what the person actually wrote this time, so the agent answers the new question rather than re-reading the thread.