Skip to main content

​
What shapes Swan’s writing

When Swan drafts an email or LinkedIn message, it draws on three sources, in this order of context:
  1. Org knowledge — your value proposition, competitive positioning, and ICP. This is the baseline context Swan uses for any GTM writing or reasoning, not just outreach. See org knowledge.
  2. Inbox-specific instructions — style and tone guidance tied to one connected inbox (one person’s email or LinkedIn account), so messages from different inboxes can sound like different people.
  3. Message examples — real messages Swan keeps: your past messages, drafts you approved or edited, and messages that got a positive reply. Swan learns from them; it does not show them to the agent while it writes.
Before writing, Swan reads the instructions of each inbox it sends from, per channel, and writes in that voice. The angle, the offer and the facts come from the task and your skills.

​
Inbox instructions

Inbox instructions are free text tied to a specific inbox, and they fully replace whatever was there before when updated — there’s no merging of old and new instructions. When you ask Swan in chat to change them, it asks you for confirmation first, since instructions apply to every future message sent from that inbox, not just the current conversation. The one change Swan makes without asking is the daily review of your own edits, described below. You can also set these directly from Tools → Email or LinkedIn → Configuration → the inbox’s actions menu → Instructions; see inboxes. Instructions stay separate for each inbox. When you ask Swan to update them, it can check whether you want the same change on your other connected inboxes; changing one does not automatically change the others. Swan also keeps each inbox’s instructions current from the edits you make to its drafts. Once a day, for every inbox where someone edited a draft the day before, Swan reads the last 30 days of edits and adds habits that show up in two or more of them, such as a shorter opener or a line you always put back. It ignores fact fixes, notes to colleagues and content tied to one campaign, and it keeps the text you wrote. You can change or remove anything it adds; if you keep making the same edit, Swan may add that habit back. When you connect an inbox that has no instructions, Swan drafts them for you from messages you sent before connecting (see below). The draft starts with “Drafted by Swan from your sent emails on [date]. Edit freely.” (or “sent LinkedIn messages”), and you get a Slack DM about it if your Slack account is linked. Swan drafts only when it found at least five good messages, and it never touches instructions that already exist, including ones you write while the draft is being prepared. The draft describes how you write (tone, length, greeting, how you ask), not what you say. Messages that look AI-written, such as ones full of stock phrases like “no small feat” or “synergy”, only teach length and layout, and Swan never turns a stock AI phrase into a rule even when you use it. When your inbox has an email signature in Swan, the draft leaves out your sign-off because Swan adds the signature; otherwise, and on LinkedIn, it keeps how you sign off.

​
Message examples

Examples are stored per organization and tagged as email or LinkedIn; the two channels never mix. They come from three places:
  • Positive replies. When a contact first replies to a Swan sequence and the reply moves the sale forward (asks for a meeting, a demo, or pricing), the message they answered becomes an example, together with the reason for reaching out and their reply.
  • Approvals. The first time a contact’s first message in a sequence is approved, it becomes an example. If you edited it before approving, your edited version is saved and marked as edited. Approving a later follow-up step doesn’t add an example.
  • Your own past messages. When you connect an email inbox or a LinkedIn account, Swan reads up to 1,000 messages you sent before connecting and keeps up to 20 of your best: written to prospects, not inbox warm-up traffic, at least 25 words, cold messages that got a positive reply first, then follow-ups that got one, then other cold messages, and one per template. This runs in the background and takes a few minutes. Messages sent through Swan are left out, so these examples are your own writing.
Swan does not give examples to the agent while it writes. When it did, the agent copied their lines and facts, and examples from teammates mixed in other people’s voices.

​
How to steer Swan’s writing

Examples are kept for learning, not shown to Swan while it writes. To change how Swan writes from an inbox, change that inbox’s instructions: ask Swan to update them when you’re unhappy with a draft, edit them yourself in Tools, or edit the drafts before you approve them (an edit you make twice or more reaches the instructions in the daily review). When you like a message, tell Swan what you like about it, and Swan adds that habit to the instructions.

​
What this doesn’t include

Swan personalizes per contact using research and reasoning, not through structured A/B testing — there’s no built-in mechanism to split traffic between two message variants and compare results. If you want to compare approaches, the practical workaround is running two separate sequences and comparing their outcomes manually in analytics.

​
Common questions

If I edit a message before approving, does that count as feedback? Yes, in two ways. Your edited version is kept as an example for a contact’s first message, marked as edited. And every edit, on any step, feeds the daily review: a change you make to two or more drafts for different contacts is added to that inbox’s instructions the next morning. To make a change stick right away, ask Swan to add it to the inbox’s instructions. Does Swan learn from which messages got replies? Swan keeps the message a prospect answered positively as an example, together with their reply. It does not change how Swan writes on its own. What does Swan do with my inbox when I connect it? It reads messages you sent before connecting, keeps up to 20 of your best as examples, and drafts writing instructions if the inbox has none. It doesn’t send anything or change existing instructions. Do inbox instructions apply to LinkedIn and email both, or just one channel? They’re tied to one inbox, and an inbox is either an email account or a LinkedIn account, so instructions apply to that one channel. If you want the same style on both your email and your LinkedIn, set instructions on both inboxes. Can I ask Swan to write more like a specific past message? Yes. Share the message and tell Swan what you like about it. Swan adds that habit to the inbox’s instructions, which it reads before every draft.