For businesses
Your AI answers customers about refunds, pricing and policies in words that are each real, and the claim can still be wrong: a dropped condition, an old number, a sentence that no longer matches the policy. These two products check what it says against what your documents say, now and after the next change.
Your policies change. Does your AI know?
When your prices, policies or procedures change, Help Center Check flags help articles and AI answers that may now disagree, with the evidence.
$99/month founding rate, 14-day trial.
1. Check
Each claim is checked against the exact line in your source. Code confirms the quote exists; a model can’t approve it on its own.
2. Watch
When a policy or document changes, Xyntherium flags the articles and AI answers that may now disagree, and shows the evidence for review.
3. Prove
Each check leaves a record of what was checked, against which version, with what result.
When “Clearance items are excluded.” is added next to an unchanged 30-day sentence, an answer that says “you can return items within 30 days” without the exclusion is now incomplete. Help Center Check flags it, because it reads the new sentence against the passage the answer relied on.
Reflowed paragraphs, new headings, “thirty (30) days” for “30 days”: recorded with the diff, visible in your dashboard, no alert.
When the new version applies from a date, or only to new orders or new customers, the finding says “support changed; applicability needs review”, not that the answer is wrong. A scenario with an order date settles it.
Saved-answer findings say “this saved answer lost support under the new version”. Optional live tests send each question to your bot’s test endpoint and say “your bot currently answers X”.
Every version, change, check, finding and reviewer decision, as JSON or PDF, linked to a hash-chained evidence ledger: exportable records of what was checked and what was found.
For businesses running an AI chatbot or support bot.
Paste your policy or product info, paste an answer your chatbot gave, and Verify shows what it got wrong or made up, with the line from your document that decides it. If you don’t run a chatbot, you don’t need Verify.
Your document, for example:
What your chatbot said, from:
A verdict for each statement in the answer (supported, contradicted, unsupported or unverifiable), the line from your document that decides it, and the conditions the answer dropped, each quoted. A model proposes the line; code confirms it exists in your document before anything is shown.
Verify checks answers against the documents you supply. Not legal advice.
Help Center Check: $99/month founding rate, 14-day trial. Verify: free for 1,000 checks a month, then Builder and Team plans.
Pricing →