Propivix drafts proposal sections with AI, and every draft needs checking before it goes to a customer. There is no single AI quality score. Instead, Propivix runs two automated checks on drafted text, the claim review and the quality gate, and shows separate figures while you pick assets. This article explains what each one measures, so you know what is left for you to check.
Propivix does not give a section an AI quality rating, and it does not keep a report of AI performance over time.
Claim review
After it drafts a prose section in English, Propivix runs a second AI pass over the draft. The reviewer compares specific claims, such as numbers, names and past projects, with excerpts from the assets used to write the section. A claim it cannot trace to those excerpts is recorded as unverified.
You see the results in two places:
- At the end of the proposal wizard, "Sections with unverified claims" gives the count, when any section reported a review.
- In the editor's Review & export dialog, "Claims needing evidence" lists the sections and claims.
The review runs only on text of 100 to 24,000 characters, and only when time allows. A section it could not review is marked as not reviewed. Editing a reviewed section and saving it with its own Save button makes the review out of date until the section is regenerated. See AI Proposal Generation.
What it means: an unverified claim is one the reviewer could not match to your assets. It may still be true, but you need to add its proof to your library or remove it. A clean review does not prove the text is accurate. It means the reviewer found no untraceable claims in the excerpts it was given.
The quality gate
The Pre-Submission Quality Gate panel in Review & export scores the whole proposal out of 100 and shows Pass or Needs Attention. It combines four figures: compliance with the RFP's requirements, clarity, evidence and claim integrity. Select Re-check to run it again after changes.
The gate must pass before Propivix prepares submission files. It blocks on missing mandatory requirements, unverified or out-of-date claim reviews, and an RFP that was not fully read. See The quality gate for each figure and each blocking message.
What it means: the score is a set of automated text signals. It is not an evaluator's score, and it does not predict whether you win.
Figures you see while picking assets
On the wizard's Assets & Team step, asset cards carry figures that are about the assets, not about drafted text:
- The evidence chip counts the RFP's requirements an asset supports, by word matching. See Requirement evidence on assets.
- "AI relevance" ranks how closely an asset reads against the RFP. It is a picking aid, not coverage.
- After Run AI Match, CV cards and staffing scenarios show team matching scores. See Team Matching Explained.
None of these tells you whether a drafted section is good. They help you choose what the AI draws on.
What you still check yourself
Whatever the checks say, read every generated section before submission:
- Confirm every number, name, date and past project against your records.
- Make sure each requirement is answered in the RFP's own terms.
- Remove anything you cannot support. Evaluators can treat an unsupported claim about past performance as disqualifying.
- Regenerate a section after substantial edits, so a new claim review runs.
Getting better drafts
Drafts improve with better inputs. Each draft draws on the RFP, your requirement responses, your assets, your business line's profile and your style references:
- Settle requirement responses before drafting. Responses still at TBD or Needs Input are left out of the draft.
- Keep your asset library current, so claims can be traced to real documents. See Asset Types Explained.
- Fill in the business line's profile. See Organizing by Business Line.
- Pick past proposals under Style Reference (Optional) to guide the writing style.