Brilo AI's Widget Knowledge Base Training can be configured to use your existing knowledge base or help centre articles as a primary source for widget answers. With proper ingestion and content mapping, the Brilo AI voice widget will surface and cite relevant articles, use semantic search over your documents, and fall back to scripted prompts when confidence is low. Typical setup includes content ingestion, embedding generation, and intent mapping so the widget answers reliably from your documentation.

Can I train the widget using my help centre?

Yes. Brilo AI can ingest and index your help centre content so the widget answers from it with semantic search and answer retrieval.

Can the widget learn from my knowledge base updates?

Yes. When configured with regular syncs or a webhook, Brilo AI can refresh embeddings and update the widget’s knowledge store.

Can the widget reference specific articles?

Yes. Brilo AI’s Widget Knowledge Base Training can return article excerpts or article links as part of its answer when configured to do so.

Enterprises ask this when they already invested in a large help centre or regulatory knowledge base and want to reuse that content without manual reauthoring. Buyers in healthcare, banking, financial services, and insurance need to balance answer accuracy, auditability, and content governance. They want to know whether Brilo AI will respect content structure, versioning, and sensitive data controls during training and live operation.

Brilo AI’s Widget Knowledge Base Training ingests documents (HTML, Markdown, PDFs, or exported help centre content), generates semantic embeddings, and stores retrievable vectors for the voice widget to query at runtime. At call time, the Brilo AI voice widget issues a semantic query (conversational retrieval) against that indexed content and returns the best-match passage, plus confidence metadata and answer provenance. Administrators can configure sync frequency, which sections to include, and how the widget should format or cite article text in responses.

In Brilo AI, knowledge ingestion is the process of extracting, normalizing, and embedding document content so the voice widget can search it efficiently.

In Brilo AI, widget knowledge base training is the configuration and indexing pipeline that prepares your help centre content for real-time voice and text queries.

In Brilo AI, confidence scoring is the numeric assessment the system uses to decide whether to return the retrieved article, ask a clarifying question, or escalate to a script or human.

Related technical terms: knowledge ingestion, embeddings, semantic search, vector store, conversational retrieval, intent mapping, webhook, answer provenance.

Brilo AI applies safety boundaries and workflow limits to prevent the widget from over‑relying on unvetted content or exposing sensitive fields. For example, you can restrict which folders or article tags are indexed, exclude pages with PII, and set minimum confidence thresholds that force a fallback script or human handoff when the match is weak. Brilo AI also supports configurable answer length limits and citation policies so the widget does not read full legal or clinical paragraphs by default.

In Brilo AI, a fallback intent is the configured behavior (scripted prompt, clarification question, or human escalation) used when confidence falls below your threshold.

What the widget will not do without explicit configuration:

Healthcare example:

Banking / Financial services example:

Insurance example:

Note: Configure privacy and retention controls when indexing sensitive content. Do not treat indexed content as a substitute for regulated advice or formal legal/medical consultation.

Brilo AI supports deterministic handoffs from the widget to a human agent or secondary workflow. Handoff triggers can include low confidence scores, detection of regulated-intent phrases (for example policy exclusions or clinical symptoms), or explicit caller requests to speak to a person. When a handoff occurs, Brilo AI can pass the retrieved article excerpt, confidence score, and call summary to the agent or to your webhook endpoint so agents receive full context before picking up the call.

Typical handoff patterns:

If you need help with any step, contact your Brilo AI implementation team or support for assistance.

Do I need to rewrite my help-centre articles for the voice widget?

No. Many articles can be ingested as-is, but lightweight editing (short headings, clear Q&A sections, and removing navigation chrome) improves retrieval quality and the brevity of spoken answers.

How often should my knowledge base sync with Brilo AI?

Sync frequency depends on how often content changes; many teams start with daily syncs and move to near-real-time webhooks for frequent updates or regulatory changes.

Can the widget attach a link or article ID to the spoken answer?

Yes. Brilo AI can return article provenance and an article ID or link in the agent UI or call summary; spoken prompts should be configured to offer sending a follow-up link via SMS or email when appropriate.

What happens if the widget finds conflicting articles with similar relevance?

Brilo AI returns the top-ranked passage and a confidence score; you can configure tie-break rules, prefer canonical documents, or force a clarification question when multiple matches are close.

Will the widget expose private or personal data from my knowledge base?

No—only indexed content that you have authorized will be available. You should apply content tagging and exclusion rules to prevent private or PII-containing pages from being ingested.