How to Use AI Assistants in UniLink (Build a Chatbot for Your Page in Minutes)
How to Use UniLink AI Assistants (Build Custom AI Bots for Your Page)
Create a personalized AI assistant trained on your content — attach it to the ChatAI block on your page so visitors can ask questions and get accurate, on-brand answers instantly.
- An Assistant is a configurable AI brain: system prompt, knowledge base, model settings, and a fallback message.
- The ChatAI block is the chat widget on your page — it connects to an Assistant and shows visitors a live chat interface.
- One assistant can power multiple ChatAI blocks; one page can have multiple ChatAI blocks pointing to different assistants.
- Fill the knowledge base with real content — FAQs, product info, pricing — or the bot will make things up.
A contact form on a link-in-bio page answers nothing. It just collects names and waits for you to reply, usually days later when the person has already moved on. An AI assistant does something different: it answers questions immediately, in your voice, using the information you give it. A coaching client asks about your program structure at 11pm — the assistant answers from your knowledge base. A potential customer asks whether a product works for their specific situation — the assistant explains based on the product details you've loaded. The conversation happens without you. UniLink's Assistants feature is how you build that. This article covers how to configure an assistant that actually helps people, rather than one that makes up answers or deadends every conversation.
What assistants do
An Assistant in UniLink is a configuration record — think of it as the brain behind the chat widget. It defines how the AI behaves: what personality it projects, what it knows, how creatively it responds, and what it says when it can't answer. The Assistant itself isn't visible to page visitors; they interact with it through the ChatAI block, which is the visible chat interface on your page. The separation matters: the Assistant is the logic, and the ChatAI block is the UI. You can update the logic (edit the assistant) without touching the page layout, and you can change which assistant a ChatAI block uses without rebuilding anything.
An assistant record contains several configurable fields. The system prompt is the set of instructions you give the AI about how to behave — what its role is, what it should and shouldn't say, what tone to use, what to do when it doesn't know something. The knowledge base is a block of text you provide — your FAQs, product descriptions, service details, pricing, bio, policies — that the AI uses to answer questions. The model setting determines which underlying language model powers the assistant. The temperature setting controls how creative versus focused the responses are: low temperature (closer to 0) means more consistent, factual answers; high temperature (closer to 1) means more varied and creative responses. Max tokens controls the maximum length of each reply. The fallback message is what the assistant says when it genuinely can't answer something — this should always be set.
Multiple assistants let you build purpose-specific bots without cramming everything into one configuration. A business selling coaching, digital products, and consulting might want three separate assistants: a "Course FAQ bot" trained on course curriculum and enrollment details, a "Product Support bot" trained on product specs and troubleshooting, and a "Coaching Intake bot" that asks qualifying questions and explains the coaching process. Each assistant stays focused and accurate because its knowledge base is narrow. Trying to pack all three into one assistant produces a bot that knows something about everything and answers precisely about nothing.
How to create an assistant
- Open Assistants in the Dashboard: Navigate to your UniLink Dashboard and find the Assistants section — usually under Tools or AI in the sidebar navigation.
- Click "New Assistant": A new assistant record opens with empty fields. Give it a name that describes its purpose — "Product FAQ Bot," "Coaching Intake," etc. The name is internal only; visitors don't see it.
- Write the system prompt: This is the most important field. Start with the assistant's role: "You are a helpful assistant for [your name/brand]. You help visitors understand [specific topic]." Then add behavioral rules: what tone to use, what to avoid, what to do when you don't know the answer. Three to five focused sentences is better than one vague sentence.
- Add a description (optional): An internal note for yourself about what this assistant is for. Useful if you manage multiple assistants.
- Fill the knowledge base: Paste in the text content you want the assistant to draw from — FAQs, product descriptions, service details, pricing, bio, terms, policies. Format it clearly: use headers and short paragraphs. The more organized the knowledge base, the more accurate the assistant's answers.
- Set the temperature: For factual Q&A (pricing, policies, product specs), use a lower temperature (0.3–0.5). For creative or conversational assistants (lead warm-up, brand voice), use a higher temperature (0.6–0.8). Avoid 1.0 for any assistant handling factual questions.
- Set max tokens: Controls the maximum length of each response. 150–300 tokens works well for concise Q&A. 400–600 tokens is appropriate for assistants giving detailed explanations.
- Write the fallback message: What the assistant says when it can't answer. Something like: "I don't have that information — please email [your address] or check [relevant page]. I'll make sure you get an answer." Always set this. An empty fallback leaves unanswered questions as dead ends.
- Save the assistant: The configuration is saved and the assistant is ready to attach to a ChatAI block.
How to attach an assistant to a ChatAI block
- Open the page editor: Open the page where you want to add the chat widget.
- Add the ChatAI block: Click "Add block" and select ChatAI from the block picker. The block is added to your page immediately.
- Select the assistant: Inside the ChatAI block settings, open the Assistant dropdown and select the assistant you just created.
- Configure the widget appearance: Set the chat window title, opening message (what the bot says first when a visitor opens the chat), and any other display settings.
- Save and test: Save the page. Open your public page URL and test the chat widget by asking questions your visitors would actually ask. Verify that answers come from your knowledge base, not fabricated details.
Key configuration settings
| Setting | What it controls | Recommended approach |
|---|---|---|
| System prompt | The AI's role, behavior rules, tone, and how it handles uncertainty | 3–5 specific sentences. Include the role, the topic scope, tone instructions, and what to do when it doesn't know something. |
| Knowledge base | The text the AI draws from to answer questions — FAQs, product info, pricing, bio, policies | Paste in all relevant content. Organize with headers. More organized input = more accurate output. Update whenever your info changes. |
| Model | The underlying AI model powering the assistant | Use the default (GPT-4o equivalent) unless you have a specific reason to change it. Higher-tier models produce more accurate answers. |
| Temperature | How creative vs. focused responses are (0 = deterministic, 1 = highly variable) | 0.3–0.5 for factual Q&A. 0.6–0.8 for conversational/warm assistants. Never use 1.0 for assistants handling facts. |
| Max tokens | Maximum length of each response from the assistant | 150–300 for concise answers. 400–600 for detailed explanations. Don't set too high — long answers lose readers. |
| Fallback message | What the assistant says when it can't answer | Always set this. Include a direct route to help: email, link, or "I'll have someone follow up with you." |
| Assistant name (internal) | Your label for this assistant in the Dashboard | Use a descriptive name: "Course FAQ Bot," "Lead Qualifier," "Support Bot." Not visible to visitors. |
| Description (internal) | Your own notes on what this assistant does and where it's used | Optional but useful when managing multiple assistants across multiple pages. |
How to get the most from assistants
Spend real time on the system prompt — it's not boilerplate. Most people write "You are a helpful assistant" and wonder why the bot sounds generic. The system prompt is your chance to give the AI a specific job description: what it's there to do, who it's talking to, what tone fits your brand, and how to handle edge cases. A coaching business might write: "You are a warm, encouraging intake assistant for [Name]'s 1:1 strength coaching program. You help visitors understand whether the program is a fit for their goals. Speak in a supportive tone, use plain language, avoid jargon. If someone asks about pricing, provide the investment details from the knowledge base. If you don't know something, say so and invite them to book a free call." That's a real system prompt. It produces a real assistant.
Update your knowledge base whenever your information changes. Prices change. Programs evolve. Policies get updated. An assistant trained on outdated content will give outdated answers — and unlike a static FAQ page, visitors trust the AI's real-time conversational responses, which makes stale information more damaging. Build a habit: whenever you change pricing, add a service, or update a policy anywhere on your site, update the assistant knowledge base at the same time.
Build multiple assistants for different use cases rather than one catch-all bot. A single assistant that tries to handle product support, lead qualification, coaching intake, and content discovery will do all of them poorly. Focused assistants with narrow, well-organized knowledge bases outperform general-purpose bots every time. If you have multiple pages serving different audiences, give each a dedicated assistant. A page for your creative services and a page for your digital products should have different bots, with different knowledge bases, different system prompts, and different fallback messages.
Test the bot the way a real visitor would — not with questions you already know the answers to, but with the messy, specific questions real people actually ask. Ask about edge cases. Ask something the bot doesn't know. Ask something slightly off-topic and see how it responds. If it makes something up, your knowledge base is missing that information or your system prompt doesn't include clear instructions on how to handle uncertainty. Adjust both based on what the test reveals, then test again before you consider it ready for live traffic.
Troubleshooting
| Problem | Likely cause | Fix |
|---|---|---|
| Bot gives inaccurate or made-up answers | Knowledge base is empty, too thin, or missing the relevant information | Fill the knowledge base with all relevant content. For every topic the bot should know, add explicit text about it. If the knowledge base is empty, the model will hallucinate plausible-sounding answers. |
| Bot sounds too generic or robotic | System prompt is too vague or missing tone instructions | Rewrite the system prompt with specific role, audience, tone, and behavioral rules. "You are a helpful assistant" is not enough. Describe the voice you want like you're briefing a new team member. |
| Bot gives inconsistent answers to the same question | Temperature is set too high | Lower the temperature to 0.3–0.5 for factual content. High temperature (0.8–1.0) produces creative variety, which is fine for conversational bots but unreliable for bots that need to answer the same question the same way every time. |
| Unanswered questions leave the conversation dead | Fallback message is empty or too vague | Set a specific fallback message that includes a real next step: an email address, a booking link, or a promise of follow-up. "I don't know" with no direction is a dead end. "I'll have someone reach out — drop your email here" is a conversion opportunity. |
| Bot answers questions outside its intended scope | System prompt doesn't define topic boundaries | Add a scope instruction to the system prompt: "Only answer questions about [specific topic]. For anything else, say you're not able to help with that and direct them to [contact method]." |
| Responses are too long and visitors don't read them | Max tokens is set too high, or system prompt doesn't specify concise answers | Lower max tokens (150–250 for Q&A bots). Add "Keep answers concise — two to three sentences unless more detail is explicitly requested" to the system prompt. |
When AI assistants add real value
- You get repetitive questions from visitors — pricing, availability, how your service works — and want them answered instantly without your involvement
- You have a product or service with enough complexity that a static FAQ doesn't cover the variation in visitor questions
- You're running a lead-generation page and want to pre-qualify visitors before they book a call or fill a form
- You're serving an audience in different time zones where real-time human response isn't practical
When assistants aren't the right tool
- Your page is simple and visitors don't have questions — adding a chat widget creates friction without adding value
- You haven't filled the knowledge base yet — launching an empty bot that makes up answers damages trust more than having no bot at all
- Your sales or intake process depends on nuanced human judgment that an AI can't reliably replicate
- You're in a regulated field where automated AI responses could create liability around advice or information
Frequently asked questions
What's the difference between an Assistant and a ChatAI block?
An Assistant is the configuration — the system prompt, knowledge base, model, temperature, and fallback message that define how the AI behaves. The ChatAI block is the visual chat widget on your page that visitors interact with. The Assistant is the brain; the ChatAI block is the interface. You can attach any assistant to any ChatAI block, and one assistant can power multiple ChatAI blocks on different pages.
Can I have different assistants on different pages?
Yes, and this is the recommended approach when your pages serve different audiences or purposes. Each ChatAI block can be connected to a different assistant. Build a dedicated assistant for each context — product support, lead intake, coaching Q&A — and attach the right one to the right page. Focused assistants with purpose-specific knowledge bases consistently outperform one general-purpose bot.
How do I know if the bot is giving accurate answers?
Test it before going live. Ask the same questions your visitors typically ask. Ask edge cases. Ask something the bot shouldn't know. If it answers something accurately that isn't in your knowledge base, it's drawing on its training data rather than your content — whether that's acceptable depends on whether the topic is factual and current. For pricing, policies, and product details, make sure those answers are explicitly in your knowledge base so you control the response.
What should I put in the knowledge base?
Everything a visitor might ask about: your FAQ content, service or product descriptions, pricing and packages, your bio and credentials, booking process, refund and cancellation policy, social proof like testimonials or case study summaries, and anything else that comes up in your typical pre-sale or support conversations. Organize it with headers and short paragraphs. The more structured and specific your knowledge base, the more accurate the assistant's answers.
What happens when a visitor asks something the assistant can't answer?
The assistant delivers the fallback message you configured. This is why the fallback message matters: a good fallback gives the visitor a real path forward — an email, a booking link, a form, or a clear promise of human follow-up. A poor fallback ("I don't know how to answer that") leaves the visitor with nothing. Treat the fallback message as a conversion tool, not an error state.
- An Assistant is the AI brain — system prompt, knowledge base, model settings, and fallback message — while the ChatAI block is the visible chat widget that connects to it.
- The knowledge base is the most important field: fill it with your FAQs, product details, pricing, and policies — an empty or thin knowledge base causes the bot to hallucinate answers.
- Use a specific, behavioral system prompt — not just "you are a helpful assistant." Define the role, tone, topic scope, and fallback behavior in three to five concrete sentences.
- Set temperature to 0.3–0.5 for factual Q&A bots; reserve higher temperature for conversational or creative assistants where answer consistency is less critical.
- Always set a fallback message that gives visitors a real next step — an email, booking link, or follow-up promise — so unanswered questions become opportunities, not dead ends.
Ready to build your first AI assistant? Open your UniLink Dashboard, go to Assistants, and create your first configuration — then add a ChatAI block to your page and connect them. Your visitors will have answers waiting for them the next time they visit.
Create Your Free Link-in-Bio Page
Join thousands of creators using UniLink. 40+ blocks, analytics, e-commerce, and AI tools — all free.
Get Started Free