The AI Assistant’s system prompt defines the role, scope, and response rules that guide replies to visitors. In Dragwyb Click to Chat, you edit this text in the AI Engine & Prompt settings. Use it to describe what the assistant should help with, how it should respond, what information it can rely on, and when it should explain that it does not know an answer.
Open the system prompt editor
- In WordPress admin, open Click to Chat → AI Assistant.
- Choose AI Engine & Prompt from the AI Assistant navigation.
- Find the System Persona & Instructions card.
- Edit the Custom ChatBot System Prompt field.
- Save the AI Engine settings using the screen’s save action, then test the assistant in Chat Preview or on a controlled test page.
Saving this field updates the configured prompt through the plugin’s administrator-protected settings endpoint. A user who does not have the required WordPress administrator capability should not be able to save these settings.
Write useful, specific instructions
A good prompt gives the assistant clear directions without trying to replace the site’s actual knowledge sources or access controls. Include:
- Role and purpose: explain whether the assistant answers product questions, helps visitors find documentation, or routes support requests.
- Scope: state which site topics it should answer and when it should redirect visitors to an appropriate support resource.
- Response style: specify the preferred tone, answer length, language, and formatting.
- Accuracy rules: tell the assistant not to invent product details, links, policies, prices, or capabilities when those details are not available.
- Escalation: describe when it should suggest the site’s support channel rather than guessing.
For example, a site-specific instruction might say: “Answer questions about the products and documentation covered by the available site sources. Keep answers concise, link to the most relevant guide when its URL is known, and say when the sources do not contain enough information. Do not invent prices, policies, or features.” Adapt the example to the content and support process of your own site.
Understand the prompt’s limits
The system prompt guides the model’s behavior; it does not guarantee that every answer will follow the instructions perfectly. Results also depend on the selected provider and model, conversation context, and any knowledge-base content retrieved for the question. Do not place API keys, passwords, private customer data, or other secrets in the prompt.
The inspected interface does not have a separate dedicated tone selector in this card. Tone is primarily expressed in the instructions. For related response controls, see Configure AI Assistant Tone.
Test changes before publishing
- Ask a common question that the assistant should answer from known site information.
- Ask a question outside the intended scope and check that the assistant does not fabricate an answer.
- Ask for a link and verify that the returned destination is the correct published page.
- Check that the tone and answer length match your instructions.
- If the response remains inaccurate, review the selected model and the knowledge-base sources before continually expanding the prompt.
For provider setup and response parameters, see AI Assistant and Response Behavior. These are recommended checks; no live response test is claimed here.