Eight fields, one or two sentences each, that turn generic AI output into output that sounds like your organisation. The first thing any team should build, whatever stage they're at.
Name, type, and scale. ie. "A family foundation giving $15M a year, team of eight" or "a nonprofit school network, 200 schools across three countries."
The change you're working toward, in one sentence. If it takes two, that's worth noticing.
Who you fund or run programs for, and who they ultimately reach. "We fund 40 early-stage orgs; together they reach two million learners in East Africa."
Why a grantee or partner chooses to work with you. Unrestricted funding, patient timelines, networks, hands-on support. For operating orgs: why your model over the alternatives.
The three funders or organisations you're most often compared with, co-fund alongside, or benchmark against. This calibrates what "normal" looks like in your corner of the sector.
Current numbers. Active grants or programs, annual budget, geographies, staff size.
The two or three outcomes your board actually asks about this year. Cost per outcome, learning gains, coverage, whatever it truly is.
How you sound. Warm or formal, plain language or comfortable with sector jargon, and anything you never say.
Build this document only from information you'd be comfortable seeing quoted back in an output. No grantee-identifying details, no unpublished partner financials, no personnel matters. The context document is the public-shaped version of who you are.
On a Claude Team or Enterprise account, set it once in a Project's instructions or your profile preferences, and every conversation inherits it. Pasting it into each prompt works on any tool, and costs more. Refresh it each board cycle; a stale context document produces stale output.