Answer Quality
Denser grades every answer your chatbot gives, flags the ones that fall short, and tells you why: missing content, incomplete, contradicting your own sources, or off topic. Then it groups the gaps by subject, so you can see what your knowledge base is missing instead of reading transcripts.
Included on Standard and above. Card required, cancel anytime.
Answer Quality
Questions your chatbot could not answer
“I don't have information about Salesforce integrations. Please contact the team for details.”
Verdict: the answer supplies no substantive information and redirects the visitor elsewhere.
Same gap, other visitors
Upload your integrations page to close this gap.
Trusted by 2,000+ businesses worldwide
City of San José
Onondaga County
EOS Australia
Portfoliobox
Upsiide
Valk WeldingCity of San JoséOnondaga CountyEOS AustraliaPortfolioboxUpsiideValk WeldingSource-grounded retrieval, citations, contradiction checks, lead capture, analytics, and deployment controls in one product system.
Every response links to the exact source passage, so your team and your customers can verify it in one click.
Surface conflicting facts before your users do.
Every question your bot missed, grouped by topic.
Script, iFrame, React, or REST deploy.
Calendly, forms, Zendesk, and actions.
Lead forms and CRM-ready exports.
Queries, latency, satisfaction, topics.
Why an answer gets flagged
“The chatbot got it wrong” is not something you can act on. Every flag names its cause, and the cause picks the fix, so the work is obvious before you open the conversation.
The answer carried no real information. The bot admitted the knowledge base does not cover the subject, or redirected the visitor elsewhere.
Upload the missing page, or add a Q&A pair for it.
The answer gave some genuine information but explicitly left another material part of the question open.
Revise the source it cited so the rest of the answer is there.
The answer makes claims that conflict with each other or with the passages it cited, such as two different values for the same fact.
Fix the source of truth, then re-run KB Health across your documents.
The question was not this chatbot's job and the answer handled it badly: rambling, pretending to knowledge it lacks, or leaving the visitor stranded.
Dismiss it, or tighten the bot's instructions so it declines cleanly.
A visitor marked the answer unhelpful. A human signal always overrides a passing grade from the judge.
Open the conversation in context and edit the pair.
Topic clustering
Every question gets a short topic label, and the flagged ones are grouped by it with a count. One missed question is an anecdote. Fourteen of them on the same subject is a content decision, and it tells you which page to write next.
This is the part most teams never get to see: not what your website says, but what your visitors keep asking it and leaving without.
Flagged questions by topic
From flag to fix
Insight you cannot act on is just another dashboard. Every flagged question opens onto the thing that closes it, and nothing changes in your knowledge base until you say so.
Add the page, PDF, or doc the bot did not have, straight from the flag.
Prefilled with the question that failed, so the next visitor gets a real answer.
Jump to the passage the answer leaned on and correct it at the source.
Open the full conversation to see how the visitor got there, or dismiss the flag.
Two halves of one audit
KB Health
Reads the documents you already have and finds the facts that contradict each other, before anyone asks.
Explore KB HealthAnswer Quality
Reads the answers you actually gave and finds the questions your content never covered, after they were asked.
You are here
Plans
Grading runs on every plan, including Free, so no history is lost while you get set up. Browsing the flagged questions, their causes, and their topic clusters is included on Standard and above, and every paid plan starts with a 7-day free trial.
Questions
Five things, and each one names a different problem. Missing content means your knowledge base has nothing on the subject. Incomplete means the answer covered part of the question and left the rest open. Contradicts sources means the answer disagrees with itself or with the passages it cited. Out of scope means the question was not this bot's job and it handled it badly. Thumbs-down means a visitor told you the answer was unhelpful. The cause is what tells you which fix to make.
An LLM judge reads each question and answer pair against a fixed rubric and returns a verdict, a cause, and a one-line rationale. A grounded factual negative such as 'we do not offer refunds after 30 days' passes. An 'I don't know' does not. An off-topic question that the bot declines cleanly also passes, because no knowledge gap is revealed and you have nothing to fix.
No. Grading runs on a schedule after conversations have gone idle, never while a visitor is waiting, so it is never in the chat request path. Visitors see no difference.
The judge labels every question with a short topic, such as pricing, SSO login, or shipping. Flagged questions are then grouped by that label with a count, so instead of reading two hundred transcripts you see that fourteen people asked about pricing and got nothing. Click a topic to filter the list down to it.
Each flag comes with the fix attached. Upload the content the bot was missing, add or edit a Q&A pair prefilled with the question, jump to the cited source to revise it, or open the full conversation to see how the visitor got there. If a flag is not worth acting on, dismiss it and it stays out of your way. Denser never edits your knowledge base on its own.
Every answer is graded on every plan, including Free, so nothing is lost while you are getting started. Browsing the flagged questions, the causes, and the topic clusters is included on the Standard plan and above. Every paid plan starts with a 7-day free trial, so you can see your own flagged questions before you pay.
Point Denser at your website, docs, and PDFs. It answers in minutes, every reply cited to its exact source.
No code. Free to start. Cancel anytime.
