Set writing guidance for your organization

Writing Guidance is a place to tell the AI how your organization writes. You set it once and it applies to everything the AI generates for you, so you stop making the same correction in answer after answer.

It is different from Canonical Facts, and the two work together:

• Canonical Facts are what is true about your organization: your budget, your founding year, how many people you served last year.
• Writing Guidance is how you want it written: the words you use, the tone you want, the things you never say.

Where to find it

Open the Knowledge Base tab. At the top of the page is a panel called How the AI writes for you, which shows a preview of your canonical facts and your current guidance. Click Manage on the right of that panel to open it. Writing Guidance is the right-hand half. Type your instructions there and click Save Guidance.

The panel starts collapsed so your document library is the first thing you see, and it shows your guidance in the summary line, so you can check what is set without opening anything.

What to put in it

Write plainly, one instruction per line, the way you would tell a new grant writer on your team. Some examples:

We refer to the people we serve as "guests," never "clients" or "beneficiaries."
Write in first person plural ("we"), not third person ("the organization").
Always write our program names out in full. Do not abbreviate them.
Do not use the word "impactful."
Lead with outcomes before describing our process.

Specific instructions work better than general ones. “Use a warm tone” gives the AI very little to act on. “Do not use the word ‘impactful’” is something it can follow exactly.

There is a limit of 2,000 characters. That is deliberate: this is meant to be a short list of standing rules, not a style guide. If you find yourself running out of room, the longer material usually belongs in your Knowledge Base as a document instead.

How it interacts with our defaults

Your guidance wins. PowerGrant has its own built-in instructions about how a grant should read, and where yours conflicts with ours, the AI follows yours.

Two things your guidance cannot change:

1. Which organization it is writing for. Guidance cannot point the AI at another organization’s information.
2. The format of what comes back. The drafting steps return their answers in a specific structure that the app reads to display them. An instruction like “reply in a single block of plain text” applies to your wording, not to that structure.

Where it applies

Everything the AI writes for you: first drafts, refined answers, second drafts, and answers in the Query tab.

Once you have set guidance, you will see a small Writing Guidance applied note next to the buttons that generate writing: the Generate First Draft and Generate Second Draft buttons, the refine box on an answer, and the Query ask box. It is there so you can tell at a glance that your standing instructions are in play, without having to come back to this page to check. If you have not set any guidance, the note does not appear at all.

The note refers to the guidance you have set right now. It does not appear on drafts you generated earlier, which may have been written before you set your guidance or under an earlier version of it.

It does not apply to the behind-the-scenes steps that read your documents, such as pulling text out of an upload or suggesting canonical facts. Those produce data for the app rather than writing you read, so house style does not apply.

The Form 990 tab in Query is also not covered. That tab looks up foundation filings rather than writing prose for you, so your guidance is not used there and the note does not appear.

Expect to reword it once or twice

Getting an instruction to do exactly what you meant usually takes a try or two, and that is normal. A common example: writing “when the legal name of the organization is requested, use X” will often apply that name everywhere rather than only where the legal name was asked for. Naming the situation more tightly fixes it, and so does being specific about where the instruction should not apply.

The loop is quick, because guidance is not baked into what you have already written:

1. Edit the guidance and save.
2. Regenerate or refine one section.
3. Read that section and adjust the wording again if it is not what you wanted.

Guidance applies the next time the AI writes something. Answers you already generated keep the wording they have until you regenerate or refine them, so changing your guidance never rewrites work you have already approved. Once it is doing what you want, run the second draft to bring the whole application in line.

Changing or removing it

Edit the box and save again to change it. To remove your guidance entirely, clear the box and save. Every version is kept, so if someone asks later why a draft was written a certain way, the history of what the guidance said is still there.

Anyone on your team can edit it, and it applies to everyone in your organization. It is worth agreeing on the wording together before you set it.