KB Health scans every document in your knowledge base, flags conflicting facts across sources, and groups your content into topics — so your chatbot answers from a single source of truth.
Included in every Denser plan
Source A states the Starter plan costs $29/mo, while Source B states it costs $47/mo for the same plan tier.
One policy doc states 14-day refunds, another states 30 days.
KB Health is built into Denser. No extra setup, no separate tooling — just click Run and review the findings.
Upload PDFs, sync from a website, or connect a Google Drive folder. KB Health works with whatever you already feed your chatbot.
Click Run from the chatbot dashboard. KB Health reads every passage, clusters them into topics, and looks for cross-document conflicts.
See a two-paragraph summary of what your KB covers, a topic breakdown, and a list of conflicting passages with confidence scores and source links.
Open both passages side-by-side, fix the source of truth, or dismiss the finding if it's a false positive. Re-run anytime to verify.
Two pricing PDFs were uploaded to the same chatbot. They disagree on the Starter plan price. Here’s how KB Health surfaces it.
Step 1 — Two documents in the chatbot’s knowledge base
Suited for personal · 2 DenserBots · 1,500 queries/mo
Same plan tier — different number
Marked "RANDOM PRICING — FOR DEMO ONLY" but still indexed
Step 2 — KB Health flags the conflict
Source A states the Starter plan costs $29/mo, while Source B states it costs $47/mo for the same plan tier.
Why this matters: without KB Health, your chatbot might quote $47 to a prospect — confidently and with citations. One click resolves it.
Findings are LLM-graded, source-cited, and tracked across runs — so fixing your knowledge base feels like working through a reviewable punch list, not an audit.
An LLM compares every passage against every other passage in your KB and surfaces conflicting facts — pricing, policies, dates, specs, anything.
Every finding ships with a 0–100% confidence score so you can triage. Start with the 90%+ items; treat lower-confidence ones as leads.
Each contradiction shows the exact two passages — Source A vs Source B — with document name and an inline preview, so you can verify in seconds.
KB Health groups your documents into auto-generated topics with counts and example docs, revealing what your KB actually covers — and the gaps.
A plain-English overview of your knowledge base, generated by an LLM. Great for onboarding teammates or auditing what's in scope.
Spot mixed-language content that may confuse your retriever. Flags the primary language and any outliers for translation or removal.
False positive? Dismiss it once and KB Health remembers. The dashboard tracks resolved vs open findings across runs.
Add new docs, edit existing ones, then re-run. KB Health is incremental — you'll see whether your fix actually resolved the conflict.
Open both conflicting passages with the original document highlighted, so you can decide which version to keep without leaving the dashboard.
Anywhere the cost of a wrong answer is real, KB Health gives you a reviewable safety net before users see the bad answer.
Catch conflicting refund windows, SLAs, or policy answers across your help center before a customer reads them on chat.
Spot stale spec sheets and version-specific quirks that disagree with current docs — so the chatbot quotes the latest release.
Make sure quoted prices, discounts, and seat caps match across pricing pages, sales decks, and onboarding PDFs.
Surface contradictions in security policies, data retention rules, or terms of service that could expose the company to risk.
Triage conflicting dosage guidance, eligibility rules, or clinical guidelines before they reach a patient-facing assistant.
Detect contradictions across contracts, jurisdictional carve-outs, and policy revisions to keep your assistant defensible.
Everything you need to know about auditing your chatbot’s knowledge base for contradictions and topic coverage.
