Create an AI Assistant From a Preset Without Skipping Review

Treat the preset as a starting draft

A preset can reduce the blank-page problem, but it cannot know the details of a particular business. Review the draft as if it were written by a capable stranger: useful structure may be present, while the audience, services, policies, and exceptions still need confirmation. Do not treat a plausible sample fact as approved company information.

Resonoon’s creation wizard collects business details and role setup. The draft can be edited before the final “Create AI Agent” action. Use the business step to enter verified context and the role step to define the audience, tasks, purpose, and communication style. Keep policy-sensitive claims out until an authorized owner confirms them. A preset should be judged by the review it requires, not by how little text it asks the team to change.

Before beginning, gather the business description, customer audience, services, and a small set of authoritative references. Decide which questions the first assistant should address and which requests should remain with a person. This preparation makes it easier to distinguish harmless wording suggestions from a substantive assumption that could mislead a customer.

Review each part of the draft

Read the business details for accuracy and relevance. Remove generic claims that do not describe the organization, add only services the business actually provides, and check locations or opening information against the current source. Then inspect the role. Does it define who the assistant serves? Are the allowed tasks specific enough to test? Does it state what happens when a fact is missing or an exception needs human judgment?

Next, compare the purpose to the customer outcome. A purpose such as helping visitors understand a published service is more bounded than a promise to solve every support need. Check the tone choice as guidance, not a factual safeguard. Friendly language cannot compensate for a missing policy or an unclear exclusion.

Finally, review knowledge separately. The editor supports PDF and Markdown uploads, while Google Docs knowledge uses a Google Drive Picker. Connecting an account alone imports nothing; a non-empty selection must be confirmed to start import and save it. Verify the resulting source selection and remove obsolete copies where needed. Do not assume that a preset supplies authoritative source material or that a connection automatically imports everything.

Test before creating customer expectations

Prepare one routine question and one tempting request outside the intended scope. For a service assistant, the routine question might ask what a published service includes; the second might ask for a guaranteed exception or a promise based on personal circumstances. Write the expected response before testing. If the draft cannot produce a source-backed answer or a safe stop, edit the business facts, role, or knowledge before launch.

The assistant test page is available in the authenticated editor for testing assistant responses. Use the same question after each meaningful change and keep a short record of what changed. If the answer is inaccurate, locate the source that should support it. If it oversteps, revise the role boundary. Avoid changing the prompt, source, and role together, because the team will not know which correction mattered.

Only after review should the wizard’s final “Create AI Agent” action create the assistant. That action is distinct from saving changes to an existing bot. When editing an existing assistant, use the relevant save control shown by the editor and wait for its success state; do not describe creation and editing as interchangeable workflows.

Make ownership explicit

Name an owner for business facts, one for policy decisions, and one for the test set. A preset does not transfer responsibility for customer-facing claims. Ask the business owner to approve every claim about services, coverage, eligibility, and commitments. Record unresolved items as exclusions until they are verified, rather than leaving sample content in place.

Keep a review copy of the draft before and after edits, with brief notes explaining material changes. This gives the approver a concrete way to inspect what came from the preset and what the team supplied. It also helps a second reviewer find unresolved assumptions without repeating the entire setup review. Store these review copies in the team's usual documentation location, identify the reviewer and date, and limit access to people who need them.

Separate unresolved questions from approved facts in that record. For each open item, name the person who can decide it and keep the claim out of customer-facing content until it is resolved. This makes the review useful as a decision log, not just a before-and-after archive of wording.

After creation, review the assistant in its intended channel and confirm the fallback process. Resonoon’s AI customer support use case offers context for a bounded support assistant. The reliable value of a preset comes from giving the team a draft to inspect; customer trust still depends on approved facts, clear boundaries, and repeatable tests.