The AI's writing style, and why it never uses em dashes

Every AI feature in SOPCompass writes to the same house style, so drafted SOPs read like your team wrote them.

The rules

• Steps are actions. A step title is a short imperative phrase with no trailing punctuation: “Send the deposit link”, not “The deposit link is sent.” Reference material that is read rather than done goes in a section’s body, not into fake steps.
• Specifics are kept exactly. Names of systems, field names, menu paths, amounts, thresholds and who to contact are copied as given. They are the reason the SOP exists.
• Vague stays vague. Where your source was unclear, the AI says so rather than inventing a resolution.
• Plain language. Short sentences, second person where natural, no filler.
• Lead with the answer. In Ask, a one line question gets a one line answer, with detail after.
• Simple formatting. Drafted sections use plain HTML only (paragraphs, lists, bold, tables). No scripts, styles or embedded frames.
• SOP, not “playbook”. The AI calls a procedure an SOP.

No em or en dashes

The AI doesn’t use the long em dash or the medium length en dash. In written output this is enforced twice:

1. Every AI prompt tells the model not to use them.
2. Because a prompt rule is a request, not a guarantee, every written AI feature also passes the model’s output through a filter before you see it. En dashes in ranges become plain hyphens (“5-10”). An em dash used as a break becomes a comma. A dash that ends a line is removed.

The filter only changes AI output. It doesn’t touch text you write yourself. In a voice interview the interviewer’s lines are spoken, so only the prompt rule applies there.