AI RECEPTIONIST & CHAT / GUIDED PRACTICE

Test difficult questions

Verify boundaries, ambiguity and handoff under pressure.

Written by Ray / SETLYUpdated 1 October 2026

THE RESULT YOU'RE WORKING TOWARD

Done means it works.

The assistant handles uncertainty without inventing facts, leaking private data or claiming actions that did not happen.

Practice project: Harbor enquiry helper

01 / GET READY

Before you start.

Platform / editor
HighLevel Conversation AI and supported knowledge-base tools
Access & plan
Demo account with the relevant AI feature enabled and an approved test budget. Voice AI and chat may have different configuration and usage requirements.
Starting point
Complete the previous lesson or use an equivalent prepared demo project.

Keep a note of the exact account, project and version you use. Work with demo data and record the expected result before making changes. Never place passwords, verification codes or private customer records in your evidence.

02 / LEARN THE TOOL

Open the official walkthrough.

Use the publisher's instructions and available screenshots to locate the current controls. Then return here for the SETLY exercise and its result checks. These links open on the publisher's website.

Source credit belongs to the named publisher. These are external references, not SETLY recordings or a claim of affiliation. Screen names and plan availability can change.

03 / PUT IT INTO PRACTICE

Your task, step by step.

Use one labelled demo record or project so you can find your work again. Finish each action before continuing; if the screen differs, consult the matching official source rather than guessing.

  1. 01

    Step 1

    Prepare scenarios with conflicting dates, unavailable services, unsupported discounts, unclear intent and requests for private information.

  2. 02

    Step 2

    Specify the safe expected response or human handoff for each case before testing.

  3. 03

    Step 3

    Run each case and inspect the response plus action logs. Check that quoted customer text does not override the assistant's configured role.

  4. 04

    Step 4

    Correct one failure at a time and rerun both the difficult case and a routine question to catch regressions.

04 / GET UNSTUCK

If something goes wrong.

Start with the likely cause

If a prompt change fixes one case but breaks normal answers, narrow the instruction and verify the knowledge source remains correct.

My screen is different from the guide

Check the product, editor, account role and selected project first. Open the credited official instructions for your version. Record the difference and pause any action whose effect you cannot verify.

The screen looks correct, but the result is missing

Inspect the saved record, published page or execution log that should prove the action. A preview, success message or draft is not always evidence that data was stored or a message was delivered. Retest with one unique demo value.

05 / PROVE THE RESULT

Check it before you call it done.

The assistant handles uncertainty without inventing facts, leaking private data or claiming actions that did not happen.

Keep useful evidence

Record the test date, project/version, expected result, observed result and a non-sensitive screenshot or record reference. Mark anything you did not test as untested. Your checkmarks do not certify an untested result.

Download the editable verification checklist ↓

Guide status: original written practice exercise with credited external references. SETLY screen recording and independent learner reproduction are not yet complete.

KEEP BUILDING

Next / 10Review and improve assistant responses ↗