Skip to main content

Generate answers

With a questionnaire Ready, SolveGRC can draft an answer for every question. Each draft is grounded in your evidence, quotes its sources, and arrives marked for review. It is never auto-approved.

Draft answers for the whole questionnaire​

On a Ready card, click Process AI. SolveGRC queues one drafting job per unanswered question and moves the questionnaire to In Progress. Answers fill in as the jobs run, and the card's progress bar climbs. You don't have to wait on the page; come back later and the drafts will be there.

Process AI runs once per questionnaire. Any question it leaves without a draft is handled one at a time in the Review Center.

A Ready questionnaire card showing the Process AI button
A Ready questionnaire card with the Process AI action, which queues a drafting job for every question and moves the questionnaire to In Progress.

You can also draft a single answer at a time from the Review Center, which is handy for regenerating one question after adding new evidence. That's covered on the Review and approve page.

What happens when the AI drafts an answer​

For each question, SolveGRC runs a retrieval-augmented pipeline:

  1. Retrieve your evidence. It searches your document library for the passages most relevant to the question, blending meaning-based and keyword search, and pulls the strongest matches. If your plan includes the Knowledge Graph, it also pulls related concept cards from it.
  2. Reuse what you've approved. It looks for similar questions you've approved answers to in the past year and feeds those in as examples, so the wording stays consistent with your house style. (More on this below.)
  3. Draft with citations. The model writes an answer grounded in those passages, and every claim quotes the exact text it came from.
  4. Score and check itself. It rates its own certainty, verifies each citation actually appears in its source, runs its other checks, and flags the answer if anything doesn't hold up.

The result is saved as a pending answer, ready for a person to review, edit, and approve.

Grounded in evidence

SolveGRC won't invent an answer. If nothing relevant is found, no draft is created. Generating a single question shows a message asking you to upload supporting documents; a Process AI run simply leaves the question unanswered. If evidence was found but doesn't answer the question, the draft is left blank and filed under Needs authoring. Some questions, such as legal positions, are reserved for a person: SolveGRC drafts nothing and asks you to answer them yourself.

The Answer Library grows as you approve​

The "Answer Library" isn't a separate place you manage. It's every answer your team has approved in the past year. When the AI drafts a new answer, it pulls your most similar approved answers in as reference examples.

That has a practical consequence worth planning around:

Your first questionnaire seeds the library

Until you've approved some answers, there's nothing for the AI to reuse, so early drafts lean entirely on your document library. As you review and approve, the answers get more consistent and sound more like your team. Approving good answers is what makes the next questionnaire faster.

Reading the trust signals​

Every draft carries four signals, shown together in the Review Center's trust panel:

  • Assurance. SolveGRC's estimate of whether the answer can be sent, as a percentage, or Gated when one of its checks is open.
  • Citations. Each claim quotes the exact passage it came from, and SolveGRC checks whether the quote actually appears in the cited source. The panel shows how many citations verified out of the total, and a citation that contradicts its source is flagged in red.
  • Integrity. Clean or Flagged: a hallucination check that flags an answer whose citations do not hold up.
  • Model self-rating. How certain the model reports being, as a percentage. It is the model grading its own work, so treat it as the weakest of the four.
Review Center trust panel
The Review Center detail pane: a drafted answer and its trust panel.

You act on all three during review, which is next: Review and approve.

A note on how answers are authored​

SolveGRC's platform model does the final grounded drafting and scoring, so every answer is checked the same way. If your organization has connected its own model (bring-your-own), that model can produce a first draft, but the platform still does the final grounded, cited version. The record notes which model did what, so authorship is never overstated.