Intake and the world bible

Intake runs once, properly, before production. Two assets come out of it. Re-running intake later because it was rushed costs more than doing it slowly the first time, since every draft produced in between has to be re-checked.

Budget two to three hours of founder time, once. This is the only large block the engagement should ever ask for.

Asset one: the voice corpus

Genuine samples the fingerprint gets built from and every draft gets checked against.

What to collect:

• 10 to 20 pieces the founder wrote themselves, unassisted. Posts, internal memos, long messages in the team chat, investor updates, conference talk transcripts.
• Recordings of them talking about their work, unscripted. Podcast appearances are ideal. Speech patterns transfer to the page better than an invented style does.
• Anything they have written that they dislike, with the reason. Negative examples calibrate faster than positive ones.

What to extract:

• Rhythm: typical, shortest, and longest sentence length. Whether they fragment. Paragraph shape.
• Openers and closers, as a distribution across the samples rather than a single preferred form.
• Lexicon: words they use, words they never use, jargon tolerance, formality, profanity.
• Formatting habits and refusals. Emoji, bold, lists, line breaks.
• Their standard analogies and the examples they return to.

Never mix assisted drafts into the corpus. Habits picked up from a model get recorded as the founder's voice and then defended on every future draft.

Re-extract from the corpus when edit rate rises. Absorbing edits one at a time produces a voice that drifts toward the ghost's preferences without anyone deciding to.

Asset two: the world bible

The facts the drafts are allowed to stand on.

Offers. What the company sells, at what price points, to whom, in the founder's own framing. Include what they do not sell, which is the more common source of an inaccurate post.

Nameable clients. An explicit list of clients that may be named, and under what conditions. Default is that nobody is nameable until they are on the list. Note any that may be described but not named.

Citable numbers. Every metric the founder is willing to state publicly, with the date it was true and the source. A number without a date becomes a liability the moment someone asks. Refresh the list quarterly.

Collaborators. Partners, investors, advisors, portfolio companies. Who may be tagged, who may be mentioned, who prefers neither.

Competitor policy. Whether competitors are named at all, named only neutrally, or engaged directly. Most engagements should pick one and hold it, because inconsistency here reads as opportunism.

The never-list. People and topics that never appear under any circumstances. Former co-founders, live litigation, a family member's business, an industry fight the founder has decided to stay out of, politics, anything under NDA.

The never-list saves the engagement more often than the rest of the bible combined. It is also the section founders fill in fastest, because they already know what it contains. Ask directly: who should never see their name here, and what topic would make you regret a post.

Maintenance

The bible is a living document with one owner on each side. Update it when a client relationship changes, a number goes stale, or something joins the never-list. Anything added to the never-list applies retroactively to the drafts already in the queue, so check the queue at the same time.