Knowledge Assistant — Evaluation Release Guide

Modified on Thu, 18 Jun at 12:37 AM

Overview

Knowledge Assistant is Schema App's conversational interface for your Content Knowledge Graph. It lets you ask natural-language questions and receive grounded answers synthesized from your own structured schema.org data — with source citations on every response.

Knowledge Assistant is powered by NLWeb, Schema App's underlying retrieval and synthesis technology. In the app and in this article, we use Knowledge Assistant as the product name you interact with.

The Evaluation Release runs inside Schema App so you can:

Explore how your structured content performs in a conversational interface
Identify strengths and gaps in your schema markup coverage
Assess readiness before deploying a hosted experience on your website or connecting agent endpoints

In-app access: Log in to Schema App and open https://app.schemaapp.com/knowledge-assistant

Access to Knowledge Assistant is enabled for selected evaluation accounts. If you do not see the page, contact your Customer Success Manager.

Related articles:

Knowledge Assistant — Product Walkthrough (Evaluation Release) — ~7-minute video demo
Knowledge Assistant — Evaluation Release Guide — this article (written guide)

What This Release Is (and Is Not)

Designed for evaluation

This release helps you test how your Knowledge Graph performs before any public deployment. It is a preview and quality-assurance surface — not a fully hardened, customer-facing chatbot on your public website.

Use it to validate coverage, citation quality, and cross-site behavior. When you are satisfied, work with your CSM on hosted deployment, branding, security review, and agent endpoint configuration.

Not included in the evaluation release

The in-app evaluation does not currently include:

Hosted widget on your public website (separate scoping process)
Usage analytics or admin dashboards
Booking, registration, or other action workflows
CRM integrations
OAuth sign-in for end users
Advanced guardrails for regulated PII workflows
Full public-site hardening (SLA, production security review)

If you need any of these capabilities, speak with your Customer Success Manager to discuss roadmap and deployment planning.

How Knowledge Assistant Works

Grounded answers with citations

Knowledge Assistant operates in Summarize Mode by default. Each response:

Synthesizes information from your Content Knowledge Graph
Draws only from your structured content — not the open web
Includes supporting citations you can click to review source pages

This approach reduces hallucination risk. When the graph does not contain enough information to answer a question, the assistant may return a thin response or indicate that coverage is limited. That is expected during evaluation — it shows where to improve markup.

Typed rich results (FAQ and ContactPoint)

When the best sources are typed schema.org entities, Knowledge Assistant can render typed UI instead of flat paragraphs:

Source type	What you see
FAQPage	An FAQ accordion reusing questions and answers from your existing FAQ markup
ContactPoint	Contact routing cards (sales, support, general inquiries, department lines) with cited sources
Other entity types	Overview, Key Facts, and citations — synthesized across relevant pages

These patterns reuse markup you may already deploy for search rich results. They are a strong signal that your graph is ready for conversational and agent use cases.

Cross-site synthesis

If your Schema App account includes multiple websites (for example, a marketing site and a support portal), Knowledge Assistant can synthesize answers across them in a single response. Citations may reference more than one domain when All sites (or multiple sites) are in scope.

This is especially valuable for enterprises where product positioning lives on one subdomain and procedures live on another — one governed graph, multiple surfaces.

Using Knowledge Assistant in the App

Selecting data scope (Site filter)

Use the site filter at the top of the Knowledge Assistant page to control which content is queried:

Scope	Behavior
Single site	Query one website at a time
All sites	Query across all sites enabled for your Knowledge Assistant tenant

Which sites are available depends on your tenant configuration. Account administrators can review enabled sites on the Settings page (see below).

Relevance threshold

Knowledge Assistant scores retrieved sources by relevance. You can adjust the minimum relevance threshold (typically 50–100) using the slider in the UI:

Higher values return fewer, more precise sources
Lower values may surface more sources, including ContactPoint cards on contact-related questions
Your preference is saved in your browser for future sessions

Conversation memory

Knowledge Assistant keeps conversation history in your browser so you can continue or refine questions over time:

History persists between visits (stored in localStorage)
Memory is not stored server-side in the evaluation release
Use Clear History to reset the conversation
Clearing your browser cache also removes conversation memory

Memory is scoped to your browser session and does not carry across users or devices.

Settings (account administrators)

If you have account administrator permissions, open Knowledge Assistant → Settings to view tenant configuration, including:

Tenant ID and enabled sites
Site scope (which properties are queryable)
Branding defaults (where configured for your tenant)

Self-service settings reduce the need for engineering requests when adjusting scope during evaluation. Contact your CSM if you need sites added to your tenant or changes you cannot make in Settings.

What to Test During Evaluation

We recommend working through the scenarios below. They mirror what Schema App uses in live demos and what enterprise buyers care about most.

1. Cross-site synthesis (multi-site accounts)

Set the site filter to All sites. Ask a question that spans marketing and support content.

Example: "What is [your core product] and how do I [common support task]?"

Review:

Does the overview accurately reflect your offering?
Do citations include more than one domain or subdomain where appropriate?
Are Key Facts complete and correct?

2. FAQ rich result

Ask about a topic where you have deployed FAQPage markup.

Example: "How does [feature documented in your FAQ] work?"

Review:

Does the response include an FAQ accordion (not only a paragraph)?
Do the questions and answers match your published FAQ content?
Are sources cited to the correct FAQ pages?

If you see only a short text answer with no accordion, check that FAQPage markup is deployed and indexed for your tenant, and try lowering the relevance threshold.

3. Contact routing

Ask a contact-intent question where you have ContactPoint markup.

Example: "What is the contact for sales inquiries?"

Review:

Do contact cards show the right email, phone, or URL for each intent (sales, support, general)?
Are citations linked to the pages where ContactPoint entities are defined?

4. Core product and service topics

Ask Knowledge Assistant to summarize your main offerings.

Example: "Summarize your [primary product or service line]."

Review accuracy, completeness, and citation quality for each major business area you care about.

5. Comparison questions

Test cross-entity synthesis.

Example: "Compare [Product A] and [Product B]."

Look for structured comparison, multiple citations, and clear differentiation.

6. Coverage gap discovery

Ask about a topic you suspect is not fully marked up.

Example: "What can you tell me about [niche topic]?"

Thin or incomplete responses usually indicate:

Schema coverage gaps
Outdated or unpublished markup
Content not yet modeled in your Knowledge Graph

Use these findings to prioritize markup improvements before public deployment. Gaps discovered during evaluation are a normal and valuable outcome.

One Graph, Three Surfaces

Knowledge Assistant is one way to consume your Content Knowledge Graph. The same governed data can also power:

Surface	Who uses it	Evaluation release
Conversational UI	Your team (and eventually site visitors)	Available in-app today
MCP endpoint	AI agents, copilots, and MCP-compatible tools	Available per tenant; contact CSM for endpoint details and security review
WebMCP	Browser-based agents (emerging standard)	Early preview; contact CSM for availability

The in-app evaluation focuses on the conversational UI. MCP and WebMCP follow separate enablement and security steps. Do not confuse your tenant's Knowledge Assistant MCP endpoint with other Schema App MCP integrations (such as project-level connectors used in Copilot Studio).

Understanding Response Quality

Knowledge Assistant reflects your structured data layer. Improving markup quality directly improves conversational output.

If a response appears incomplete or inaccurate, common causes include:

Incomplete schema coverage for that topic or entity
Outdated structured data not yet republished or re-ingested
Content not modeled in the Knowledge Graph (no JSON-LD on relevant pages)
Site scope excluding the property where the answer lives

If you believe a response is wrong:

Click the citations and verify the source pages
Confirm markup on those pages in Schema App Editor or Highlighter
Check that the correct site(s) are enabled in Settings
Contact support if the issue persists after markup is correct and published

For client-side rendered (JavaScript or Tag Manager) deployments, schema freshness in Knowledge Assistant may depend on crawl schedules or real-time push configuration. Ask your CSM which ingestion path applies to your account.

Troubleshooting

Symptom	Likely cause	What to do
Knowledge Assistant page shows no data or errors	Tenant not provisioned or data not ingested	Contact your CSM; confirm evaluation access and data load
Page not available in navigation	Account not enabled for evaluation	Ask your CSM about Knowledge Assistant evaluation access
Thin or empty answers	Markup coverage gap	Review schema on cited (or missing) pages; expand markup
Incorrect citations	Outdated or misattributed markup	Audit and correct markup on cited pages; republish
No FAQ accordion	No FAQPage source in scope, or relevance threshold too high	Confirm FAQPage markup; lower relevance slider; retry
No contact cards	Missing ContactPoint markup or low relevance	Confirm ContactPoint entities; lower relevance slider
Only one domain in citations (multi-site account)	Single-site filter selected	Set site filter to All sites and retry
Conversation history missing	Browser cache cleared or different browser	Expected — history is stored locally in the browser
Slow responses or timeouts	Complex multi-site queries or synthesis load	Retry with a narrower site scope; report persistent issues to your CSM

Security and Data Handling

Knowledge Assistant runs in Schema App's secure, tenant-isolated infrastructure
Responses are synthesized strictly from your structured Knowledge Graph
Conversation memory in the evaluation release is client-side only (browser storage)
Operational logs are retained per account for support and troubleshooting
Customer data does not cross tenant boundaries

If you are evaluating Knowledge Assistant for use in regulated environments (healthcare, financial services, and similar), coordinate with your Customer Success Manager to review security and compliance requirements before public deployment.

When You Are Ready to Go Beyond Evaluation

When core topics pass your internal review and stakeholders are confident in response quality, your CSM can help you scope:

Hosted conversational UI on your website (branding, placement, guardrails)
MCP or WebMCP endpoints for agent integrations
Security review and hardening for production traffic
SLA and support alignment for customer-facing use

Public deployment is a separate process from the in-app evaluation. Your CSM will walk through use case, content scope, brand guidelines, and compliance needs before go-live.

Knowledge Assistant — Product Walkthrough (Evaluation Release) — ~7-minute video: cross-site synthesis, FAQ rich results, and contact routing

Need Help?

Contact your Customer Success Manager if you have questions about:

Evaluation access or tenant provisioning
Coverage gaps and markup priorities
Multi-site or Settings configuration
MCP, WebMCP, or public deployment options

You can also submit a support request through the Schema App platform.

Knowledge Assistant turns your Content Knowledge Graph into a conversational interface. This evaluation release helps you assess readiness — with citations — before you go live.