Founders who publish in both rooms usually land on one of two bad habits. They paste the LinkedIn post into X and watch it get cut off mid sentence, or they run two personalities and find the X account sounds like a stranger who happens to share their job.
Sonar takes a third position, and the whole platform model is built on it. One identity, per platform registers. Same person, same opinions, both rooms. What changes is how the room lets you talk.
One identity, two rooms
The architecture is literal about this. One voice model, learned from your real posts wherever they live. Each platform then gets a register: rules for how that one voice expresses itself in that room.
With enough real posts the register is learned rather than assumed. Twenty X posts produce a measured X register; five LinkedIn posts beside an X anchored voice produce a measured LinkedIn register. Below those floors the register is translated at draft time from a set of written rules, and that instruction is emphatic about translating rather than transplanting. A LinkedIn feed of one line tweet fragments is a transplant.
Registers survive a platform switch. Turn X off and its register stays on file, dormant, still there a year later. What you edit is split the same way: the lesson learned from a rewrite is filed against the platform it was made on, so an edit to an X post never teaches your LinkedIn drafts anything.
What never changes
Three things cross the boundary untouched.
Substance. The X register instruction is explicit that the voice model still owns substance, stance and vocabulary, and that the register changes only compression and rhythm. Your opinion does not change because there are fewer characters available to state it in.
The rules you have set. Voice rules you have saved, from your edits or from something you told the chat, ride every register. A rule saved for both rooms is stored once and folded into each register at draft time.
The people rule. Private individuals are never named in anything publishable, and the instruction lists X versions alongside posts, captions, comments, carousels and scripts. Compression is no loophole. Mechanics in how Sonar keeps your colleagues out of your LinkedIn posts.
What changes: the physics
Some of the difference between the rooms is what the platform will accept, before anyone makes a style choice.
LinkedIn allows 3,000 characters. Roughly the first 210 show in the feed before the see more fold, across about three lines, so the opening has to earn the click on its own. No threading. Carousel documents and quote cards both work as visuals.
X allows 280 characters per post on a free account, or up to 25,000 on a paid one. No fold, because the whole post is shown. Posts thread. Quote cards work as visuals, and carousel documents do not, since the document post is a LinkedIn feature.
The 280 limit is enforced rather than suggested. When a draft is saved for X, the body is split on lines of three dashes into segments, and any segment over 280 characters on a non premium account is repacked into a thread at sentence boundaries before it is stored. The function is applyXPhysics, and it runs when a draft is created, when it is edited and in the chat's final pass, with a backstop at publish. A premium account keeps its long post untouched, since X will accept it.
The threading maths packs whole paragraphs first, so a sentence break never gets promoted into a paragraph break, and only a paragraph too large for one post breaks down into sentences and then words. A thread is one item in your calendar with its segments held inside it, so a six post thread never counts as six posts against your weekly cadence.
Cadence differs too. LinkedIn's default weekly target is three posts and X's is six, which reflects each room's native rhythm rather than any claim about what works. Both are starting points; your own goal lives in settings, and that is what the meter on Home measures against.
What changes: the register
Under the physics sits the register, which is mostly about compression.
For X, the instruction asks for the same person compressed: direct, hook first, short sentences, line breaks between beats, no hashtags, emoji only if your own writing uses them. The take or thread decision is made before the writing starts. One sharp claim or single beat story is one post under 280. A setup with a turn and a payoff, a numbered argument, or anything carrying two or more distinct beats, is a thread of three to seven, where the first post earns the read on its own and the last one pays it off. Cramming a story into one crowded post and padding a one liner into a thread are named as the two failure modes.
For LinkedIn, when the voice was learned from X posts, the instruction is the mirror image. Same person, slower room. Keep the vocabulary, the stance and the directness, and write real paragraphs of two to four full sentences. A LinkedIn post can hold a setup, a story and a payoff, so use the room honestly: do not compress like X and do not pad to fill.
One more rule matters if you publish in a single room. The model is handed a channel contract naming that platform, and it then never raises the other one: no cross post suggestions, no variants, no threads, not even as a follow up question.
The same idea in both rooms
Here is one idea written twice. It is an invented example, and the point of it is that the two versions share an argument and share no sentences. In the X version, a line of three dashes is the break between one post in the thread and the next, which is the same separator Sonar writes.
LinkedIn.
We turned off our free trial last month and closed more deals in the four weeks after than in the eight before it.
The trial was cheap enough to run. What it cost us was the first conversation. Someone would sign up, poke around for a day, and arrive at the call having already decided what the product was, usually based on the two screens they happened to open first. I spent the next thirty minutes correcting an impression instead of finding out what they were trying to fix.
Now there is no trial and there is a call. It is a worse experience for the buyers who genuinely want to evaluate software alone on a Sunday, and a better one for everyone else. I would not have believed that a year ago.
X.
we killed the free trial and closed more in the 4 weeks after than the 8 before
---
the trial was cheap. the damage was to the first call.
people showed up having decided what the product was, from whichever two screens they opened first.
---
so the call became 30 minutes of correcting an impression instead of 30 minutes of finding out what they were actually trying to fix
---
now: no trial, just the call. worse for the people who want to evaluate alone on a sunday. better for everyone else.
i would not have believed that a year ago
Same argument, same opinion, same stake in the ground. The LinkedIn version spends the room's space on a middle paragraph that carries the reasoning. The X version breaks that reasoning across four beats, because on X the reader chooses to continue after each one. Neither one names a customer.
The rhythm inside the sentences is covered in measured prosody, where the material comes from is in what to post on LinkedIn after a client call, and the connection itself is on the X integration page. The rest of the writing is on the Sonar blog, the product is at Sonar and explained on how it works, and there is one plan on pricing.
Questions
Do I have to use both platforms?
No. Choose one at onboarding or change it in settings with one tap. A single platform founder is never shown or told about the other room, and that is enforced by an instruction the model receives rather than by a filter on its output.
What happens if a draft is too long for X?
It is repacked into a thread at sentence boundaries when it is saved, unless your connected X account is premium, in which case the long post is stored as written.
Does a thread count as several posts against my weekly target?
No. A thread is one content item with its segments inside it, so the cadence meter counts it once.
Do my edits on one platform change the other?
No. An edit made on an X draft teaches the X register only. The two registers learn separately on purpose.