> ## Knowledge Base Index > Fetch the complete knowledge base index at: https://help.chrisp.chat/sitemap.xml > Use this file to discover available pages before exploring further. > Pure-Markdown content can be obtained by appending a '.md' suffix to the content URLs listed in the sitemap (without the trailing slash). # How to prepare your AI Data Hub (and why it matters) As Crisp AI becomes smarter and more capable, the quality of its responses depends on a single thing: ****the data you give it****. This article explains how to prepare your [AI for Customer Support](https://crisp.chat/en/ai/), not just technically, but strategically, so you get better, faster answers and stay ready for what’s coming next. --- ## Why this matters A well-structured AI Data Hub is more than just content, it’s the foundation for measurable, scalable impact. Here’s what a clean, complete AI Data Hub enables: * ****Faster time to resolution:**** AI surfaces relevant answers instantly, reducing response time for customers and agents. * ****Lower escalation rates:**** More issues are solved at first contact, without needing handovers to senior reps. * ****Higher self-service deflection:**** Crisp AI can deflect up to 30-50% of repetitive questions when trained with the right data. * ****Better CSAT:**** Fast, accurate answers = happier users. Crisp AI provides consistent answers that match your brand and tone. * ****More efficient teams:**** Support, product, and ops can align on a shared, always up-to-date source of truth. || ⚠️ Crisp AI doesn’t rely on generic web content or guesses, it learns from your real product knowledge. If your Data Hub is outdated, incomplete, or unclear, your AI will fail to deliver. This isn’t about “connecting features”. It’s about ****building an evolving system of truth**** that powers your entire customer experience, from Chat to [AI Chatbot](https://crisp.chat/en/chatbot/) to Copilot and beyond. [Ticket deflection: How AI Chatbots reduce support backlogs](https://crisp.chat/en/blog/reduce-support-backlogs-ai-chatbot/) [From FAQs to AI: How Chatbots Deliver AI Self-Service Support at Scale](https://crisp.chat/en/blog/ai-self-service-support-at-scale/) [First Response Time: The #1 metric that shapes long waiting times](https://crisp.chat/en/blog/first-response-time/) --- ## What is the AI Data Hub? The ****AI Data Hub**** is the knowledge base that powers [Crisp AI](https://crisp.chat/en/ai/#train). It’s made up of 5 types of data sources you control: ###### Answer Snippets Short internal Q&A pairs created manually. They’re not visible to customers, but heavily used by AI to generate answers. ![](https://storage.crisp.chat/users/helpdesk/website/-/7/1/8/6/71861673493c2400/51cfc9bd-4b25-4d1e-911f-1ee022_1ni4pox.png) → Ideal for covering edge cases, internal rules, or product limitations not documented elsewhere. ###### Web Content Crisp crawls public content on your websites. This includes your landing page, blog, documentation... ![](https://storage.crisp.chat/users/helpdesk/website/-/e/b/0/3/eb030bcc7f81c800/04ecec0c-be84-4d70-827f-98e079_1s3jot0.png) → Every indexed page feeds the AI. You can include or exclude any page manually. ###### Data Importer Upload internal files: PDFs, CSVs, TXT… ![](https://storage.crisp.chat/users/helpdesk/website/-/1/9/e/5/19e55fd3284c0000/70ec50c0-8989-4a6a-b2aa-4fbeab_11ricdj.png) → Useful for specs, policies, legal docs, price sheets, or other internal content from outside Crisp. ###### Inbox Messages AI learns from your past support replies. ![](https://storage.crisp.chat/users/helpdesk/website/-/5/7/3/e/573ea09a8f967c00/d79cfd83-236e-47d2-9109-fc7810_1sze2l7.png) → It replicates the tone, content and style your team uses every day. ###### Knowledge Base Articles Your official help articles, created inside Crisp. ![](https://storage.crisp.chat/users/helpdesk/website/-/8/6/c/a/86ca4695d5cfd800/86c2d061-e8ac-4082-b78a-50832e_wmt6fh.png) → The most structured, editable source. Updated in real time. Each source brings value. Together, they form your AI’s real brain. [How can I add data sources to Crisp AI Hub and why?](https://help.crisp.chat/en/article/how-can-i-add-data-sources-to-crisp-ai-hub-and-why-ijkdrl/) [How to use the Knowledge Base](https://help.crisp.chat/en/article/how-to-use-the-knowledge-base-i2tsm1/) --- ## What your AI must be able to answer Your AI shouldn’t just know what your product does. It should respond confidently to the questions your users actually ask and that depends a lot on your industry. ****Here’s a breakdown of essential topics by type of business:**** | | | Product Overview | What does it do / Who is it for | What do you sell | What is the service | What do you help clients with | What’s the training about | | Getting Started / Onboarding | How to sign up, invite users | How to order, track shipment | How to open an account | How to start a project | How to enroll / access course | | Pricing & Plans | Subscription tiers, billing | Pricing, discounts, delivery fees | Plan options, interest rates | Pricing model, retainers, scope | Course pricing, free trial | | Feature / Service | Using key features | Applying codes, managing orders | Making payments, using dashboard | Booking calls, submitting briefs | Watching lessons, passing quizzes | | Troubleshooting | Login issues, error messages | Failed payments, missing orders | Card issues, security alerts | Login to portal, missing deliverables | Course not loading, access denied | | Refunds / Cancellations | Cancel plan, get refund | Return item, refund timing | Cancel policy, refund criteria | Cancelling a retainer/project | Withdraw from course, refund process | | Contact Support | Support hours, escalation rules | Email/chat/contact page | Claims hotline, secure chat | Email, Slack, dedicated PM | How to contact instructor/support | | Account Settings | Change email, close account | Update shipping info | Change beneficiary, address updates | Change billing details, permissions | Change password, update profile | | Security / Privacy | GDPR, 2FA, hosting info | Data use, cookie policies | Encryption, compliance | NDAs, secure file sharing | Data privacy, certificates | | Integrations / Compatibility | Slack, Zapier, API, browsers | Shipping providers, payment gateways | Bank sync, app integration | Project management tools (Notion, Asana) | Browser support, mobile access | If your AI can’t answer at least one question per row for your industry, it’s time to fill the gap with an article, a Snippet, a data import or just link your website. || 💡 Tip: Use this table to review your own help content. --- ## How to build and maintain a solid AI Data Hub ### 🧼 Clean ****Start with an audit:**** * Are your Snippets still relevant? * Any outdated articles? * Are you indexing the right web pages? * Have you removed old PDFs? ****Checklist:**** * Remove or update anything older than 6-12 months * Check for duplicate answers or conflicting info * Unify tone and terminology (e.g. don’t say “workspace” in one source and “account” in another) * Disable web pages that should no longer be indexed The AI will use everything you give it, so what’s inaccurate is just as harmful as what’s missing. ### ✅ Complete Now plug the gaps. ****Use the 10-topics checklist above. Ask yourself:**** * Are these topics all covered? * Are they in the right place? * Are they recent enough? ****Actions:**** * Add Answer Snippets for topics not covered elsewhere * Upload key internal documents (e.g. pricing sheets, onboarding checklists) * Refresh your Web Content crawl if you’ve updated your site * Create 1-2 core articles if your KB is still empty Don’t aim for completeness. Aim for ****clarity on what matters most****. ### 🔄 Maintain Once it’s clean and complete, your goal is to keep it that way. ****Make these part of your routine:**** * After every major release → update related articles + snippets * Once a month → re-run Inbox training and Web crawl * Every quarter → do a content audit * Optional → assign an “AI owner” in your support or ops team If your AI performs worse over time, it’s rarely a model issue. It’s almost always a data decay problem. --- ## What to do based on your team size ### 👶 Starter * Add your website as Web Content * Create 5-10 Answer Snippets (copy/paste from real chats) * Write one article (e.g. “Getting started”) ### 🚀 Growth * Use all 5 source types * Cover the 10 priority topics * Refresh content monthly ### 🏢 Scale * Assign ownership * Connect all internal sources (API docs, policies, security docs…) * Set quarterly review processes * Use segmentation like to handle volume [What is a segment and how can it help your team?](https://help.crisp.chat/en/article/what-is-a-segment-and-how-can-it-help-your-team-88hhzw/) [How can I create sub-inboxes in my Crisp Workspace?](https://help.crisp.chat/en/article/how-can-i-create-sub-inboxes-in-my-crisp-workspace-ke3xk7/) --- Good AI isn’t magic. It’s structured content, maintained over time. Start small. Improve gradually. And let the AI do more for your team. || Want to go further? Check out these helpful resources: [How to measure AI Chatbot ROI with Crisp?](https://help.crisp.chat/en/article/how-to-measure-ai-chatbot-roi-with-crisp-1wfzwn/) [Set up Crisp Analytics for Customer Support: 4 dashboards templates](https://help.crisp.chat/en/article/set-up-crisp-analytics-for-customer-support-4-dashboards-templates-1f02z4o/) [11 AI chatbot best practices for always-on customer service](https://crisp.chat/en/blog/ai-chatbot-best-practices/) [Overnight support for SaaS companies: AI as the night shift](https://crisp.chat/en/blog/overnight-support-for-saas-companies-ai-as-the-night-shift/) [Nail 80% AI-resolved conversations through human-grade support with AI](https://crisp.chat/en/blog/business-nail-80-ai-resolved-conversation/)