Review & approve

Drafting is the AI's job; deciding is yours. The Review Center is where your team works through each drafted answer: reading the evidence behind it, editing where needed, and approving what's ready to send.

Open the Review Center

From a questionnaire card, click Review (or View). The Review Center opens with that questionnaire loaded, and you can switch to another from the picker at the top.

If you arrive from the sidebar instead, without naming a questionnaire, the Review Center opens the one most likely to need you — the most recently updated questionnaire still awaiting review. The picker switches between them.

The screen is split in two: the question list on the left, and the answer detail on the right.

The Review Center: the filterable question list on the left (each row with a confidence meter and a flag tag where relevant), and the answer detail on the right with the trust panel, drafted answer, and citations.

Across the top, a few numbers tell you how far along the questionnaire is: answered versus total, how many answers are flagged, the average confidence, and the share already approved.

Work the queue

The left list is how you triage, and it splits by what each answer needs next rather than leaving everything in one flat pending pile:

• Needs review. Awaiting your decision, with nothing specific holding it up.
• Ready. The AI's own checks came back clean and well-evidenced. These are the ones you can move through quickly.
• Needs evidence. The answer outran what it could prove — a citation didn't verify, or the integrity check flagged it.
• Needs refresh. The answer stands, but the evidence behind it has aged.
• Needs authoring. The draft abstained: the evidence didn't answer the question, so a person has to write it.
• Approved and Rejected. What's already decided.

The middle five only ever match answers still awaiting a decision, so the queue empties as you work rather than re-showing settled ones.

"Ready" has to be earned

An answer drafted before the assurance checks existed was never evaluated against them, so it is never filed under Ready on the strength of its age. It stays in Needs review until something gives it a real verdict — your decision, or a regeneration. No answer inherits a clean bill of health it never sat for.

Search narrows the list further. Click any question to load its answer on the right.

Read the answer before you decide

The detail pane gives you everything you need to judge a draft without leaving the page.

• Trust panel. The confidence percentage, the count of verified versus problematic citations, and a clean-or-flagged integrity read, side by side.
• Inline citations. Numbered [1], [2] chips in the answer text. Click one to jump to the source it came from.
• Evidence and citations. The list of sources, each showing the exact passage quoted and a tag for whether that quote holds up. A citation that contradicts its source is called out in red.
• Assurance. The overall verdict on whether the answer is supported, along with any open gates — specific reasons it isn't clear to ship.
• Model self-rating. How certain the model reports being. Useful context, but it is the model grading its own work: treat it as the weakest of these signals, not the strongest.

The Explainability tab beneath the answer shows how the draft was put together — what was retrieved and what it was used for.

Read the source, not the score

Every number here is a claim about the answer rather than proof of it. The one move that settles it is opening a citation and reading the passage. That is how you tell a good answer from a merely confident one.

An answer that must not ship as-is: the Confidence / Citations / Integrity trust panel showing 0 of 4 citations verified, and the citation-conflict banner telling the reviewer why.

The Evidence & citations list: each source card shows the verbatim quoted text and its verification tag, so every claim traces to a document.

Look closely at flagged answers

When integrity flags an answer, it's telling you a citation didn't line up. The answer might still be right, but it needs a person to confirm or fix it before approval. Give flagged answers a proper look rather than approving them on autopilot.

Edit an answer

Click Edit, adjust the text, and Save version. SolveGRC keeps the previous wording as a version in the history and marks the answer as human-edited, so it's always clear which answers a person shaped. (Approving or rejecting without changing the text doesn't create a new version; only edits do.)

Approve, reject, or request a revision

Use the action bar at the bottom of the detail pane:

• Approve. The answer is good to send. SolveGRC stamps who approved it and when, then moves you to the next question.
• Reject. The answer shouldn't be used.
• Request revision. Send it back to be redrafted or rewritten.

If an answer needs a fresh draft, say you've just added a relevant policy, Regenerate re-runs the AI with your current evidence and returns a new pending draft.

When Approve is unavailable

Sometimes Approve is disabled and hovering it explains why. This is deliberate: the same rule is enforced in the database, so the button is telling you in advance what the save would refuse. There are five reasons:

• The draft abstained. The evidence didn't answer the question. Write the answer yourself, or regenerate once the evidence exists.
• The sources disagree and nothing resolved it. Edit the answer to state which source governs, or regenerate once they agree.
• It answers less than was asked. Part of the question isn't covered. Submitting it would leave the rest unanswered — add the missing part.
• It claims more than its citations support. Check the claim against the cited text and edit it down to what the evidence actually says.
• Citations couldn't be verified against their sources. Edit the answer to take ownership of the wording, or regenerate it.

Each has the same two ways out: edit the answer so it is true of your evidence, or change the evidence and regenerate. Note what is not on this list — staleness alone never blocks approval, because an aging document usually still says the same thing.

The hardest defect to catch by reading

"Claims more than its citations support" is the one to slow down for. The prose reads fluently and the citations beside it are real documents — they just don't say what the sentence says. That combination is very hard to spot at a glance, which is exactly why it's gated.

Approve in bulk

For long questionnaires, select answers with the row checkboxes (or Select all), then apply Approve, Reject, or Request revision to the whole selection at once. Keyboard shortcuts speed it up: Ctrl/Cmd+A to select all, Ctrl/Cmd+Enter to approve the selection.

Bulk review: row checkboxes selected and the inline bulk action bar offering Approve / Reject / Revision across the selection.

Quick-approve the clean ones

With the Ready filter selected, a Quick-approve button appears whenever any of those answers qualify, clearing them in one action. Qualifying is deliberately narrow — an answer must be all of:

• AI-drafted and still pending,
• triaged Ready,
• carrying an assurance score of 0.8 or better with no open gates,
• citation-validated, with no integrity flag,
• and attached to a question that isn't marked sensitive.

Anything short of that is left for you. Answers drafted before the assurance checks existed carry no measured score, so they never qualify either: no measurement, no shortcut.

Let confidence guide your effort

Sort or filter by confidence and citation health, approve the clean, high-confidence answers in bulk, and spend your attention on the flagged and low-confidence ones. That's the fastest way through a large questionnaire, and it doesn't cut corners.

Collaborate on an answer

Open the Discussion tab on any answer to talk it through with your team. Comments thread, so you can reply in place, and you can tag a teammate with @mention to pull them in. They get a notification that links straight back to the exact answer. Mark a thread resolved once it's settled to tuck it out of the way.

The Discussion tab: a threaded comment with an @mention and its type badge, plus reply / mark-resolved controls.

Track versions and diffs

The History tab shows every version of an answer, newest first, with who changed it. View Diff lays two versions side by side with additions in green and removals in red. Restore brings an older version back. Note that restoring returns the answer to pending review, since the approved wording has changed.

The History tab's version comparison: two versions side by side, with added text in green and removed text in red.

What approval sets in motion

Approving isn't only a status change. When you approve an answer:

• It's stamped with your name and the time, for the audit trail.
• It becomes eligible as an Answer Library example, so future drafts of similar questions start from your approved wording.
• Depending on your organization's settings, it can be registered as evidence automatically, or you can register it yourself. See Register evidence and export.

Once every question in the questionnaire has an answer, SolveGRC advances the whole questionnaire to In review on its own. That's the module's finish line. From here you register evidence and export the answered copy.