# How your assistant writes > The formatting rules every Beam assistant follows, why each one exists, and the research behind them. The difference between a text that gets a reply and one that gets ignored is usually how it is written, not what it says. Every assistant on Beam follows the same set of writing rules. They are on automatically, they are not switchable, and they are tuned for one thing: getting a reply. This page explains each one and the evidence behind it, so you can see why your assistant writes the way it does. Two kinds of rulesThese control how your assistant writes and apply to every business on Beam. Your own house rules control what it says, and you write those yourself. Where the two disagree, formatting wins. ## Never end a text with a period This is the single most effective rule on the list, and the best evidenced. Researchers at Binghamton University showed people short text exchanges where the reply was a word like "Okay", "Sure" or "Yeah". When that reply ended in a period, readers rated it significantly less sincere than the identical word without one. A follow-up study found the same thing and added that it reads as abrupt. The detail that makes this convincing: when the same messages were shown as handwritten notes, the effect vanished. The period is not doing grammar in a text message, it is doing tone, and the tone it conveys is annoyance. ``` Sounds good. <- reads as irritated Sounds good <- reads as friendly ``` Gunraj et al., Texting insincerely: The role of the period in text messaging, Computers in Human Behavior, 2016. Follow-up: Punctuation in text messages may convey abruptness. Period, 2017. ## Spell everything correctly, and never fake a typo A popular piece of advice says to have your AI misspell a word now and then so it seems human. The research says this is a bad trade, and we do not do it. A 2022 study had a hundred people rate short messages for trustworthiness. Spelling errors cost about three points of trust each, and the damage stacked: two errors cost roughly six points, five errors cost thirteen. There was no free allowance where the first mistake was harmless. Your customer is deciding whether your business is competent enough to hand money to. There are better ways to sound human than looking careless, and the rest of this page is those ways. Witchel et al., Spelling Errors in Brief Computer-Mediated Texts Implicitly Lead to Linearly Additive Penalties in Trustworthiness, Frontiers in Psychology, 2022. ## Sentence case, not all-lowercase Writing entirely in lowercase reads as relaxed and friendly between friends, and it is genuinely how a lot of people text. But the same research that documents that habit also documents people switching back to capitals in professional situations. A business texting a customer entirely in lowercase risks reading as careless rather than casual, particularly to older customers. Sentence case with the final period removed is the safe middle: it looks like a person, not like a form letter. ## Keep it short, and vary the length A study of real text messages found an average of about 65 characters, roughly fourteen words, when people had 160 available. Nobody was running out of room. They were choosing to be brief. More useful than the average is the variation. The spread was nearly as large as the average itself, which means real conversations are lumpy: a four-word message next to a longer one, then a fragment. Messages that are all a similar length are one of the clearest signs of automation, which is why your assistant deliberately varies. Thurlow, Generation Txt? The sociolinguistics of young people's text-messaging, 2002, corpus of 544 real messages. ## No emoji unless the customer uses one first A 2026 study found that messages without emoji made the sender appear most competent and professional. Positive emoji helped only alongside neutral or good news, and notably failed at the thing people most often use them for: softening bad news, where they made the sender seem insincere. So your assistant stays clean by default, and never puts an emoji next to a price or a disappointment. If your customer opens with one, it can mirror them. Collabra: Psychology, Emojis at Work: The Effects of Emoji Use on Perceptions of Competence and Appropriateness, 2026. ## No em dashes, semicolons, or bold text None of these appear in normal texting. A semicolon in a text message is the punctuation of a press release, and bold text does not even render in most messaging apps, so it arrives as stray asterisks. The em dash deserves a note because it has become famous as an AI giveaway. The honest version: measurements do show AI using it around three times more often than people, but the human range is wide enough that plenty of real writers use it more. It is a weak signal, not proof. We remove it anyway, because in a sixty-character text it has no job to do and a large share of your customers now read it as machine-written. ## Avoid the words that give AI away Some words have become so associated with AI writing that they now work against you. The clearest case is measured: an analysis of millions of research abstracts found the word "delve" appearing 28 times more often after AI writing tools became common. Your assistant avoids that whole family, including seamless, robust, elevate, leverage, unlock, meticulous, and openers like "I would be happy to", "Certainly", and "Great question". It also avoids lists of three ("faster, easier, and more affordable"), which is a rhythm AI falls into constantly and people rarely use in a text. Kobak et al., Delving into LLM-assisted writing in biomedical publications through excess vocabulary, Science Advances, 2025. ## One exclamation mark at most The same research that found periods read as insincere found the opposite for exclamation marks: a reply with one was rated more sincere. It is a genuine warmth signal, which is exactly why it stops working when overused. One per message, never two, never three. ## What this looks like in practice ``` Before: "Hello! I would be happy to help you with that. Our seamless solution delivers faster, easier, and more affordable results. Let me know if you have any questions!" After: "hey! depends on what you need honestly. most of our clients run between 1-3k a month depending on call volume" "are you handling a lot of inbound leads right now?" ``` Shorter, no sign-off, no em dash, no marketing vocabulary, no closing period, and two messages of different lengths instead of one block. That is the whole difference, and it is applied to every message your assistant sends without you doing anything. ## What you still control These rules govern style. Everything about substance is yours: what you sell, what to mention, what never to bring up, which questions to ask first, what words you use for your own business. See Coaching your assistant for how to set those, and try any change in the test chat before a real customer sees it.