Helping a visitor resolve their question
Helping a visitor resolve their questionAsk a question: natural conversation. Find an answer: verified company sources. Add context: goal · details · contact. Move forward: instructions · request · link. Exception: Answer absent from sources → acknowledge and investigate.01Ask a questionnatural conversation02Find an answerverified company sources03Add contextgoal · details · contact04Move forwardinstructions · request · linkNo sourced answer → acknowledge and investigate
The chat should help visitors in their situation. Ask for contact details only when they serve a useful next step.

What should the chatbot achieve on a first visit?

Understand the reason for visiting, answer from the right material and offer a specific next step. For a trade business, that may mean clarifying a job; for a manufacturer, finding instructions; for a shop, explaining product variants.

It need not collect every contact detail at the start. Someone asking about delivery does not need to complete a sales form. If they want a tailored quote, the agent gradually collects scope and contact details. It remembers information already provided during the conversation and does not ask again.

Clearly identify the chat as an AI assistant. A natural tone can improve usability without implying that the customer is speaking to a particular employee. Trust rests on accuracy and usefulness, not a hidden identity.

Why is crawling the whole website not enough?

Because websites rarely contain every necessary detail and may repeat information in conflicting versions. An old article may show an earlier price list. A product page may omit compatibility information support staff know by heart.

Prepare the core offer, common questions, service limits, current terms and contact rules. Define the authoritative source for each important fact. Prices and availability should come from a current system when they affect customer decisions.

A telling example was described by a specialist shop owner on Reddit in September 2026: technical recommendations depend on genuine product knowledge. This is one person’s experience, not a test of all chatbots. For your website, it highlights the need to test difficult product questions as well as opening hours.

What do established conversational products show?

In its Tidio case study, Anthropic reports 71% of Tidio’s own support handled autonomously, experience from over two million conversations and a typical response time of 5–20 seconds. The figures have different denominators: Tidio’s own support, usage volume and speed respectively. They do not predict universal results for your website. Tidio case study.

“Claude is more conversational and provides more details.” — Yasser Elsaid, founder of Chatbase, in the Chatbase case study.

Natural wording is useful but does not prove truth. A good business conversation needs verifiable information and a way to handle unknown answers. A long reply that invents service availability can do more harm than a brief acknowledgement of uncertainty.

How is a chatbot installed on different websites?

The simplest option uses embedded JavaScript. The script loads the assistant window and connects it to the right business configuration. It need not gain access to the whole database or change the order system.

Website typeTypical embedding routeSeparate work required
ShoptetCustom HTML code in administrationProduct and order integration
WordPressScript manager or theme templatePermissions and, if needed, WooCommerce
Custom websiteShared template via developer or SSHDeployment and testing of the specific app
Basic FTP hostingEdit the appropriate template fileStructure checks and backup

For Shoptet, we follow its official HTML code guide. Existing code must be preserved. FTP storage or access to images does not imply permission to edit system templates.

After embedding, test mobile layouts, overlapping buttons, loading speed and chat opening on several page types. Seeing it on the administrator’s homepage is not enough. Automated installation should also produce a backup and verify the widget is not loaded twice.

How should a conversation capture an enquiry?

First, the assistant helps with the question. When the visitor wants an individual solution, it collects missing information, summarises the request and explains the next step. It checks the contact email by repeating it or asking for correction of an obvious error.

An enquiry submission needs a concrete status. If a message has not been delivered, the chat must not claim the business received it. Users should be able to continue after handover without creating a new, unrelated case.

Visitor text must not choose arbitrary recipients or change sending rules. The business configures its contact channel in advance. This creates a useful customer service, rather than a public tool for sending messages to strangers.

How can we spot answers that look right but cause harm?

Test ambiguous and incomplete questions: a nonexistent product, a price without a variant or service outside the company’s area. The response should ask an essential follow-up or acknowledge source limitations.

In the OpenAI Community in August 2026 users discuss how current and usable business context is. This is community experience, not an official product guarantee. Test the same issue directly: change a fact in the source and check when and how the answer updates.

Also test returning to an earlier question. Good conversation handling uses corrected context instead of clinging to the first assumption. If a visitor meant another variant, all subsequent conclusions must adapt.

What should we measure after installation?

Measure useful resolutions, captured-enquiry quality, repeated unanswered questions and operating costs. Opening a chat window is not a commercial result, and longer conversations do not necessarily mean better service.

Start with the questions the chatbot could not answer. They often expose missing website information. Filling these gaps helps people, the assistant and search visibility. Once the knowledge is verified, extend the chat with an agent capable of service actions.

Frequently asked questions

Do I have to change my website for a chatbot?

Usually, a short JavaScript snippet in a shared page section is enough. The location depends on your CMS. Where templates can be edited through SSH or FTP, our integration can handle installation.

Can the assistant be embedded in Shoptet?

Basic embedding can use custom HTML code in the administration interface if the account supports it. Adding the script does not require publishing a public extension. Access to orders or product data is a separate integration step.

Does the visitor have to provide an email immediately?

No. They can ask a normal question first. Request contact details when they want follow-up from the business, a specific proposal or an answer that cannot be resolved in chat.

Research and solution design: Tanduva with AI assistance. External case studies are identified; illustrative examples are not measured results from our clients. Editorial methodology and corrections.