Refining a drafted answer

Every drafted answer can be refined: tell the AI what you’d change, and it rewrites that one section — grounded in your knowledge base — while the rest of your draft stays untouched.

Refine a section

1. Find the section you want to improve and click Refine. If the button is greyed out, the answer is marked Complete — untick that box and it comes back.
2. Describe the change the way you’d brief a colleague: “tighten this to 150 words,” “lead with the youth-employment outcome,” “make the tone less formal.”
3. Submit. The section shows Refining draft while the AI works — the box stays right where it is, so you can watch it finish without losing your place on the page.
4. The refined text replaces the draft in place. Your refinement request is saved above the section, so you have a record of what you asked for.

Compare and undo

• Click a past refinement request to see a before / after comparison of exactly what changed.
• Not better? Undo restores the previous version — PowerGrant keeps the last few versions of every section.

Marking an answer Complete

Each answer has a Complete checkbox. Tick it when you’re satisfied and want to stop thinking about that answer: it locks the text, greys out Refine and Comment, and the progress count in the top bar goes up. Untick it any time you want to work on the answer again.

It is a way of keeping your place, not a gate. On a twenty-question application spread over three weeks, “which of these have I actually read?” is a real question, and the count in the bar answers it at a glance. But nothing waits on it — you can generate the second draft, run the final review, and file the application to your Submission Library with answers still unmarked. The bar tells you how many are unmarked; it doesn’t stop you.

Two things worth knowing:

• Generating a second draft clears every Complete mark, because the second draft is new text and worth re-reading.
• If an answer has an unresolved accuracy flag, the checkbox stays unavailable until you clear it — that one is a real gate, and it’s about a fact you haven’t confirmed rather than about progress. Editing the highlighted text also clears the flag: once you’ve changed the wording, the question the AI raised is about a sentence that no longer exists.

Refining vs. commenting

Refine asks the AI to rewrite. Comment leaves a note for your teammates (and yourself) without changing the text — useful for “board should review this number” or “waiting on Q3 data.” Comments never trigger a rewrite, and line breaks you type are kept, so a longer note stays readable.

Both buttons are available on a Unique Answer too. Comment works exactly as anywhere else. Refine asks for confirmation first — a Unique Answer is your own wording, and letting the AI rewrite it is the one thing that marker exists to prevent, so it never happens without an explicit yes.

Tips for better refinements

• One instruction per refinement beats five at once.
• Things you find yourself asking for on every section — “never say clients,” “always spell out program names” — belong in Writing Guidance instead. Set them once and every refinement follows them. A Writing Guidance applied note next to the refine box tells you they are in play.
• Name the fact you want featured — the AI pulls it from your knowledge base.
• If the AI keeps missing a number, it’s usually because that number isn’t in your knowledge base yet. Add it there and refine again.