Blue bubbles
Writing messages that land
Current documentation
Formatting and tone rules for messages sent from Beam, including what AI-written text gets wrong.
Keep formatting minimal
Bold, italics, and heavy punctuation are supported but they read badly in a text thread. A message with several bold phrases and underlines looks like a newsletter, not a person. Plain sentences win.
If an AI writes your messagesInstruct it explicitly to avoid markdown. A model left to its own devices will produce bold headers and bullet lists inside what is supposed to be a text message, and the recipient can tell instantly.
Rules worth putting in your assistant's instructions
- One or two sentences per message. If it needs a paragraph, it needs a call.
- No markdown, no bullet points, no headers.
- Contractions always. Lowercase-casual is fine.
- One question at a time.
- Never send links in a first message.
- Vary the wording between recipients. Identical text sent to many people at once is the clearest spam signal there is.
On deliberate imperfection
A little informality reads as human. Overdone, it reads as illiterate, and it is your sales rep's name on the message. If you want an assistant to write loosely, give it specific examples of what that means, rather than a general instruction to make mistakes. A misplaced letter in a long word is human. Mangled basic words are not.
What good looks like
hey, saw you grabbed the guide. what made you start looking into this? yeah we handle the follow-up so leads don't sit. want me to walk you through how it'd work for your setup? monday around 11 or wednesday at noon, which is easier?
Short, specific, one question, no formatting, no links.