A document written for a person tolerates gaps. The reader has judgment: if a fact is missing, they look it up, ask, or decide it does not apply. The text can afford to be ambiguous because the reader fills it in.
At Elytra, the documents that hold up the operation are read by an agent before it acts, and sometimes nobody reviews the result until later. That changes what the text has to do. The agent fills the gaps too, but not with judgment. It fills them with whatever is plausible.
A few weeks ago we asked one to write a follow-up email to a client. In that client's file, the job titles of the two contacts are listed as pending confirmation. The agent wrote the email without mentioning them, and said so: it does not invent them because the file marks them as unknown.
It is a small thing. An invented job title in a follow-up email does not sink a deal. It is the kind of mistake the client notices and nobody corrects afterwards.
The rule that came out of it: an empty field and an unknown field are the same thing to a person, and opposites to an agent. Empty reads as "does not apply". Unknown reads as "do not invent". You have to write down the absence, not just the fact.
And that costs. Our client files got to nineteen data points marked as pending. The easy, fast thing would have been to keep leaving them blank, and nobody would notice until the day an agent fills them in on its own.
We made it a step in the process: a client file is not closed until what we do not know is marked as such. The nineteen pending items are still there, unresolved, and that is how they should stay until someone confirms the data.