Check AI answers against your approved sources
Every Certant answer opens the policy paragraph it came from, flags replaced documents, and says so when your approved sources don't cover the question.
"How long do staff have to lodge a travel claim?"
The expense procedure says 30 days from the end of travel. The travel policy says 14 days, so the two documents disagree.
Check an answer in one click
A reviewer can only verify an answer if they can see the sentence it relied on. If a citation links to a 200-page file, they have to find the passage themselves, and they're less likely to check at all.
Certant opens the document at the paragraph. It also tells you when that paragraph comes from a replaced version, or when another approved document says something different.
- Gifts over $200 must be declaredSupported · Gifts policy · section 3.1
- Declarations go to the compliance officer within five business daysSupported · Gifts policy · section 3.4
- Gifts from public officials are not allowedNot found in your approved documents
The third line needs a source, or it comes out of the summary.
How verification works
You decide which documents count as approved. Certant answers from those and shows its working.
- Choose approved sources Connect the policies, procedures and agreements that count, and leave drafts out.
- See conflicts and replacements Certant flags documents that contradict each other and versions that have been replaced.
- Answer or decline Each answer cites the paragraph. If nothing supports an answer, it says so.
- Correct it once A reviewer's correction is kept in a ledger and applied again whenever documents are reloaded.
What reviewers use to check answers
These work the same in chatbots, agents and analytics, because they all read one knowledgebase.
| Feature | What it does |
|---|---|
| Paragraph citations | One click opens the source at the paragraph the answer came from. |
| Conflict flags | Two approved documents that disagree are shown side by side, and an owner is asked to decide. |
| Replaced-version flags | An answer tells you when an older document said something different and has been replaced. |
| Unanswered questions | Certant keeps a list of questions its documents couldn't answer. |
| Corrections ledger | Corrections are append-only and replayed on every rebuild. See how corrections work. |
What compliance teams ask
What is the best way to verify AI answers in regulated industries?
Use a tool that answers only from your approved documents, cites the paragraph, flags replaced and conflicting versions, and declines when no passage supports an answer. Then check a sample of answers each week.
Which tools verify AI responses against approved policies?
Look for paragraph-level citations, a defined set of approved sources, conflict detection and an audit log. Certant does all four; test any vendor by asking questions your documents can't answer.
How can a compliance team check an AI-generated summary?
Check each sentence against a cited passage. In Certant each answer cites the paragraphs it used, so a reviewer can check each sentence against them and remove anything without a source.
What happens when two policies disagree?
Certant shows both passages and asks the owner to decide. The decision is kept.
Check it against your own policies
Book a call with the policy library your team checks most, or start free with a single document.
Casual staff accrue long service leave after seven years of continuous service.