We measured thirty announcement posts. The problem is not tone

When a draft sounds corporate, the advice is always the same: shorten the sentences, cut the jargon, be more human. We had thirty drafts and an engine that measures text, so we checked whether that advice is aimed at anything real.

It is not, or at least not at the thing it names. Here is what actually separates the two versions.

The setup

Fifteen announcement moments, from a product launch to a price increase to an outage to going quiet for six months. For each one, two drafts of the same announcement:

• the default, written to match the register that the ranking "25 examples of X post" listicles actually publish, and
• the rewrite, which obeys one rule specific to that moment and changes nothing else.

Both drafts of all fifteen were then read by the measurement engine behind our free brand voice generator. It is arithmetic over the text rather than a judgement: sentence counts, word counts, first and second person, whether the text addresses a reader at all. Every figure below can be checked by counting the drafts, which are all published on the fifteen pages themselves.

What moved

First-person mentions fell on 15 of 15. Average 5.07 per post before, 1.67 after. Not most of them. All of them, without that being the rule for any of them.

Not one of the fifteen default drafts addressed the reader at all. Zero out of fifteen. The engine flags a text as reader-addressing when it uses second person in a way aimed at the person reading, and the standard version of every single moment failed that test, including the ones whose entire purpose is to tell a customer something that affects them.

What did not move, which is the part worth publishing

Average sentence length went from 13.71 words to 13.83. It got shorter on eight of fifteen, which is close enough to a coin flip to mean nothing.

The rewrites are longer. 48.9 words on average before, 50.3 after.

So the two most common pieces of advice for a draft that sounds corporate, shorten it and say less, are not what distinguishes these thirty texts. You could apply both perfectly and land exactly where you started.

What is actually going on

The default announcement is about the announcer. That is usually said as a matter of taste, and it turns out to be arithmetic: five first-person mentions and zero reader-addressing, in a post whose entire purpose is to tell somebody else something.

Everything people notice about these posts follows from that one property rather than from tone.

Stock phrases are the clearest case. Three of the fifteen defaults tripped the generic-phrase detector and none of the rewrites did, and not one rewrite was edited to remove a phrase. A sentence about the reader has nowhere to put "excited to announce". The phrase needs a subject who is excited, and once the subject is the reader, the construction stops being available. The jargon was a symptom.

Length works the same way. The rewrites are longer because being about the reader costs words: a date, a figure, what happens to their existing work, what they need to do. "We are thrilled to announce" is short precisely because it commits to nothing.

The five hardest moments, and why they are hard

Not every announcement fails the same way, which is why one rule for all of them does not work.

A price increase is the only announcement where the reader wants the opposite of what you want. Warmth reads as manipulation there, and the fix is to put the old number, the new number and the date in the first sentence.

An outage or a mistake is the only one written while the facts are still arriving, so speed and accuracy trade against each other in real time. Hedging is the instinct and it is what produces "some users may have experienced".

An award contains no reader benefit at all. Nothing changes for anybody, which is why these read as hollow even when the recognition is real.

A new hire is about a person who did not write it, cannot easily object to it in their first week, and will be introduced by it for a year.

A rebrand is the only one graded on the same axis as its own subject: a vague rebrand post is evidence that the rebrand is vague, because the identity is the thing that produced the writing.

Those five cannot share a rule, and that is the case against every "10 templates for your announcement" page: the template encodes a register, and the register is not what is wrong.

The thing this does not tell you

The engine measures properties of text. It cannot tell you whether an announcement was worth making, and that is the more common failure by some distance. Every one of the fifteen pages carries a section arguing for not posting at all. That is a required field in the registry, and it is required because writing the fifteen made it obvious: there is no announcement here without a real case for silence.

It also cannot tell you whether the post sounds like you. Second person and a date do not add up to a voice. What produces a voice is a set of things you consistently do and consistently refuse, which is what the twelve voice cards are worked examples of, and what a stored brand persona exists to hold so it does not get re-invented per post. Describing your brand once is the argument for that, and it is a different argument from this one.

If you want to run it on your own draft

Paste it into the brand voice generator. It is free, ungated, and the measurements are the same ones above. Two numbers are worth looking at before anything else: how many first-person mentions you used, and whether it registers as addressing a reader at all.

If the answers are "a lot" and "no", the draft has the problem all thirty of these had, and shortening the sentences will not touch it.