AI receptionist knowledge base: approved answers, updates and fallback

Description: Compare how an AI receptionist uses business information, handles conflicting sources and receives updates before relying on it for customer answers.

An AI receptionist knowledge base is a buying decision about information sources and ongoing maintenance. Before choosing a service, ask where its answers come from, who can change those sources and how the office checks that a change reached the active receptionist.

The vendor examples below describe documented controls, not independently tested performance. They do not establish that Revo supports website ingestion, file uploads, a knowledge hub or any particular testing interface. Sources were reviewed October 23, 2026.

Separate information from instructions and actions

A greeting tells the receptionist how to introduce the business. A reference answer supplies information about a service. A business record may contain approved hours or locations. A booking action attempts to reserve an actual appointment. Those are different responsibilities.

For example, a page saying that the business offers weekday appointments does not prove that next Tuesday is available. A document describing a warranty does not authorize the receptionist to approve a particular claim. Keep those boundaries visible when discussing the setup with a provider.

Start with this proposed source register:

This is an editorial worksheet, not a built-in product feature. Do not add customer-specific records to a general reference collection merely because uploading them is technically possible.

Ask which sources the product actually uses

RingCentral’s hosted AIR skills guide distinguishes generated, editable FAQs from knowledge bases containing website links or documents. It also describes selecting knowledge bases for a receptionist after adding sources to the knowledge hub. That is a useful example of why storing material and attaching it to the active agent are separate checks. RingCentral-hosted AIR skills guide, pages 5–8

Jobber describes a different starting point: its receptionist uses business data already in the Jobber account and can be fine-tuned with operating details. That does not mean every field in an account is available for every response. Ask for the documented scope and a demonstration of your specific requirement. Jobber Receptionist

For each shortlisted service, record whether it uses manually approved answers, structured account settings, uploaded files, selected web pages or another documented source. Do not assume that adding a homepage imports every linked page, that a PDF is fully understood, or that a website edit is immediately reflected in answers.

Resolve contradictions before asking the AI to resolve them

Suppose a fictional contractor’s approved policy says that estimates need office review, while an older brochure says all estimates are automatically booked. The first task is to identify and correct the conflicting material—not to add a vague instruction to “always be accurate.”

Ask the provider what happens when sources disagree. If it cannot document a precedence rule, do not invent one in your operating instructions. Remove or exclude stale material through the supported process and keep one approved answer for the consequential policy.

Record the decision, source owner and effective date outside the system if the product does not provide those controls. The point is to maintain a usable evidence trail, not to pretend every receptionist includes version history.

Establish the update process before launch

An uploaded file can be correct today and wrong after a service change. A website source also needs an understood refresh process. Ask these questions before treating source maintenance as included:

• Who may edit answers or attach sources?
• Does a website change require a manual refresh or another documented action?
• How can the office tell that processing is complete and the intended source is active?
• Can an outdated source be disabled or replaced without disabling unrelated answers?
• What is the supported recovery route if a change produces the wrong response?
• Who pays for and performs ongoing maintenance?

RingCentral’s skills guide documents controls to edit, disable or remove individual FAQs and to activate or deactivate its knowledge-base skill. These are product-specific examples of controls to inspect, not a claim that all services offer the same options. RingCentral-hosted AIR skills guide

Do not describe a file upload as retraining the underlying AI model unless the provider establishes that this is actually what happens. A business owner generally needs to know which information is consulted and whether the resulting answer is correct—not an unsupported technical label.

Check changed answers without involving real customers

Use a provider-approved non-live test interface where available. RingCentral’s hosted setup guide describes testing an AIR response through chat as well as by calling an extension. It also describes a backup extension setting. These are documented options for that product, not verification that a particular account or Revo includes them. RingCentral-hosted AIR setup guide, pages 4–7

The following proposed checks use fictional business details. They are not tests performed for this article:

1. Known answer: Ask a straightforward question whose approved answer is present.
2. Different wording: Ask the same question conversationally without copying the FAQ.
3. Missing answer: Ask for information the business has not supplied. Confirm the approved fallback rather than an invented answer.
4. Conflicting material: In an authorized isolated test, inspect how the product handles a known conflict; do not deliberately expose live customers to contradictory policies.
5. Updated policy: Change an authorized test source through the supported process and check that the current answer is used.
6. Action boundary: Ask for a specific appointment or exception that the reference material cannot authorize.

Record the source version or dated copy, expected answer and observed response. A successful test demonstrates that scenario under that configuration; it is not a guarantee that future answers will always be correct.

Make uncertainty an acceptable outcome

Give the office a practical fallback for unanswered or consequential questions. Depending on the supported setup, that may be a request for review or a transfer to an authorized person. Confirm which action actually occurs and how the office knows it owns the next step.

A caller asking whether an unusual repair is covered should not receive an invented warranty decision. Likewise, intake should not turn a symptom into a diagnosis or a request into an arrival promise. A clear statement that the office needs to review the question can be more useful than a confident but unsupported answer.

Compare the maintenance obligation, not just setup speed

A purchasing demonstration should show how the service handles both an initial approved answer and a later change. Ask what your staff must maintain, which controls are included, and what requires provider support. A fast initial import is not the same as a maintainable process.

Is a knowledge base the same as a phone script? No. The script shapes the conversation; the knowledge source supplies reference information. Neither alone establishes permission to complete a transaction or booking.

Does connecting a website guarantee current answers? No. Confirm the supported refresh process and verify consequential changes.

Does this guide establish Revo knowledge-base features? No. Ask Revo which information sources, editing controls and fallback actions it currently supports.

Use the phone-answering script guide for conversation wording and call-quality monitoring for reviewing interactions. Compare Revo’s current plans only against capabilities confirmed for your setup.