How to choose a website chatbot for your business

A useful comparison starts with your customer questions and the work your team needs to finish. Use this checklist to evaluate a website chatbot with your own business information, rather than judging only a polished demo.

For business owners comparing AI chatbots for retail, online stores and service enquiries. The example questions below are a test plan, not customer results.

1. Pick one customer job to test

Choose a concrete starting point: helping a visitor compare products, explaining a delivery policy or gathering the details needed for a quote. Write down what a successful next step looks like and who in your team will handle it. Broad promises to automate everything are difficult to evaluate against real work.

For a product comparison, provide the current catalogue, prices and important differences. For a service enquiry, provide eligibility, service areas, required details and the route to a human. Ask the provider to demonstrate the same job with your supplied information.

2. Check both known and missing information

Prepare a question with a clear answer in your knowledge, one with an exception and one your source does not answer. For example: “What does this service include?”, “Does that apply outside your usual service area?” and “Can you guarantee availability tomorrow?” The assistant should distinguish a published fact from an unknown detail.

For website imports, ask which pages are included, how updates are refreshed and what happens when a page conflicts with a managed policy. Public website retrieval does not prove access to private orders, live inventory or customer accounts. In V7, imported public pages supplement managed business knowledge and need review and refresh.

3. Test follow-up questions in the same conversation

Try this sequence: “I need an option for a small team”; “My budget is lower than I first thought”; “Which of those fits that budget?” Check whether the assistant uses the earlier context and asks a useful question when a detail is missing. Also test a greeting and an unrelated question so ordinary conversation does not produce a random sales script.

Check the boundary as well: start a new conversation and verify that another visitor does not receive the previous visitor’s details. Ask how history is retained, who can access it and whether returning-customer memory needs a verified identity. Remembering one session and identifying a returning customer are separate capabilities.

4. Separate a suggested action from a completed action

Ask the chatbot to open a quote request, find an order or change a booking. Establish whether it offers a website button, opens a form or actually updates an authorised business system. The wording should match the result: opening a form is not submitting it, and a request is not a confirmed booking.

For an action that changes data, test permissions, customer identity, confirmation, repeated clicks and provider failure. Get the integration requirements in writing before depending on order changes, refunds or other private account operations. V7 provides configured website destinations and enabled enquiry flows; further business-system access depends on the integration and permissions available.

5. Compare the complete cost

Compare the subscription and any optional assistance separately from usage. Ask whether usage is billed by messages, conversations or model tokens, whether context counts, which model is used and whether one customer turn can trigger several paid calls. Ask for the exact behaviour when the budget runs out.

For V7, AI processing uses separately purchased prepaid credit, with no fixed message cap; automatic recharge requires explicit authorisation and platform availability. Website and WhatsApp share the credit balance. Phone numbers and WhatsApp provider fees are separate. Compare real conversations of different lengths rather than assuming that every message costs the same.

6. Check the customer and team experience

Test the live widget on a phone and with a keyboard. Check that text remains readable, links have clear labels, buttons open the intended destination and error messages explain the next step. Verify that your team can update an answer without changing application code.

Before launch, submit a test enquiry, find it in the workspace and have a team member complete the follow-up. Keep a short record of failed questions, fix the underlying knowledge or workflow and rerun the same sequence after a change. A successful demo is a starting point for this process, not proof of reliability in every conversation.

What to check before launch

This guide is published by Vertex Seven, the provider of V7 Agents. It describes evaluation questions and V7’s public capabilities, not an independent ranking of providers or a guarantee of customer outcomes.

Make an informed decision.