You want the chatbot to answer about your business, your services, your help pages, and your FAQs. It can only do that if it has that information. For example, in our case: we maintain documentation for our AI Chatbot plugin. Most people still do not open it until something breaks, or until there is no live person to ask.
If the only option is a contact form and a reply in an hour, a lot of people just leave. So the win is not “we have docs somewhere.” The win is: people get instant answers from the same text you already wrote.
What “train” means for a WordPress chatbot
Training your content on AI works by storing it in a numerical form the system can search quickly. Each piece of text gets turned into a list of numbers called a vector. You can think of it as a fingerprint for meaning: two sentences about the same topic end up with similar fingerprints, even when the wording is different. You never see those numbers. The plugin handles that part.
When you run Learn in the Knowledge Base, Limb reads your articles, splits them into smaller chunks, converts each chunk into vectors, and saves them in a searchable store on your site. Later, when a visitor asks a question, the chatbot does the same conversion on their message, finds the closest matching chunks, and uses those to build an answer grounded in your content.
Content structure that trains well
In theory, a good chatbot does not need perfect pages. You can throw a long wall of text at the Knowledge Base and it will still learn something. In practice, though, well-structured content almost always beats a messy dump of text. The chatbot finds the right facts faster, and the answers it gives back are clearer.
By well-structured, we mean a normal, accurate help article: a clear title, headings that name the topic, short paragraphs, and lists where steps belong. One section, one idea. If someone asks “How do I reset my password?” the answer should live under a heading that says something like that—not buried in paragraph twelve of a generic About page.
A few habits that help: put the most important fact in the first sentence under each heading; use H2 and H3 for real sections, not for styling; break long guides into separate pages instead of one giant post; add a short summary or FAQ block when the same question comes up often. You are not writing for the algorithm. You are writing so a human—and a chatbot retrieving that page—can spot the answer without hunting.
Set up Limb in three steps
A chatbot on your site is one way to deliver that. Below we use Limb AI Chatbot as the example. Install the plugin from Plugins in WordPress. Open Limb. You will see three steps: AI Settings, Appearance, and Knowledge Base.
Step 1 — AI Settings
Pick an AI company and paste an API key. Limb works with OpenAI, Google Gemini, Anthropic (Claude), xAI (Grok), DeepSeek, and OpenRouter (one key that can reach many models). Not sure how to get a key? Use our API key guides—each provider has its own page. Put the key in the field. Click Save.
Step 2 — Appearance
Change colors, icons, where the avatar sits, how the window opens, and the rest. Here is the full list: Appearance.
Step 3 — Knowledge Base
This is where you give the chatbot your real material. You can add different kinds of sources. A simple start: use pages and posts that are already on your WordPress site. On our site, help articles live in a type called Documentation. In Knowledge Base you pick WordPress, then that type, then the exact articles. When you are ready, press Learn.
When that runs, Limb reads the text from those articles and saves it in a way the chatbot can search when someone types a question. (If you want the deep technical version, read Knowledge Base.)
For each WordPress group (pages, posts, custom types), you can turn on Auto in Knowledge Base. With Auto enabled, saving an edited page relearns that content automatically—you do not need to press Learn again after every update.
What you can add to the Knowledge Base
In Limb, training starts in Knowledge Base. You pick a source type, choose the exact items, then press Learn. You can mix several sources in one chatbot—use whatever matches where your answers already live.
WordPress pages and posts are the usual starting point. Help articles, service pages, FAQs, blog posts that explain how things work—if it is already on your site, you can index it without copying text elsewhere. Custom post types work the same way (on our site, documentation is a custom type). Enable Auto on that group if you want edits to relearn on save without running Learn each time.
Other options when the answer is not a full page:
- Q&A entries — short question-and-answer pairs for pricing, policies, or anything visitors ask in one line.
- Manual text — paste or type content that does not need its own WordPress page.
- Files from Media — PDFs and plain text files already in your WordPress Media library (upload the manual or spec sheet first, then select it in Knowledge Base).
- URLs and sitemaps — pull in a single page from the web, or crawl a whole section via sitemap when docs live outside WordPress.
A practical order: start with your core help pages and FAQs, add Q&A for gaps, then add files from Media or external URLs only where you need them. Full reference: Knowledge Base.
Test and polish in the playground
On the Knowledge Base screen you have Playground. It is usually already on. If it is off, turn it on. You can talk to the chatbot right inside the WordPress admin. You do not have to open your public website. More here: Knowledge Base.
Now type the same kind of questions a real visitor would type—about the pages you just added. Read the answer. Does it make sense? If the chatbot shows you which page it used, check that it picked the right page.
Start with easy questions: “How do I…?” “Where is…?” Then ask harder follow-up questions. Then type short, messy questions, like someone is in a hurry and not thinking.
If the answer is wrong, open the page you expected it to use. Is the fact written there in plain language, easy to find? If your page is clear, the chatbot has a better shot. If the page is confusing, do not be surprised if the answer is confusing too.
After a few back-and-forth messages, some chatbots start to slip. That is not always your fault. It can be the tool. Your job is to give it good pages. The tool’s job is to keep using those pages when the chat gets long and messy.
If the information is right but you do not like the style—too long, too formal, too much formatting—go to Chatbot → AI Settings and change the instructions. That changes how it talks, not which pages you connected.
When the chatbot cannot answer
Sometimes the chatbot will not find a safe answer in your knowledge. You still need a fallback so the visitor is not stuck. In Limb, two common paths are Forms and Live Agent.
Conversational forms
Limb’s forms are meant to feel like part of the chat, not a separate old-style form page. You can tell the assistant when to open a form based on what the visitor is trying to do, or when it is clearly stuck. For example: if it cannot answer, you can have it offer a Contact us form or a Support ticket flow. The visitor fills it out step by step in the conversation, and your site admin gets notified with what they submitted. Learn more: Conversational forms with an AI Chatbot for WordPress.
Live Agent handoff
You can also teach the chatbot when to hand off to a human. That is built into Limb: real live chat for your team, not only an AI-only widget. In your instructions, spell out the cases—for example the visitor sounds frustrated, or the topic is billing, orders, refunds, or security and you want a person involved. After you set those rules, Limb can escalate into the Live Agent flow.
For email alerts (for example when someone is waiting on an agent) to arrive quickly, your WordPress site should have working SMTP so mail is not stuck in the default PHP mail path. Full setup and behavior: Live Agent in the AI Chatbot for WordPress.
When to retrain
For WordPress groups with Auto turned on (pages, posts, custom types), you usually do not need to retrain manually. When you edit and save a page, Limb relearns that content in the background.
For everything else—Q&A entries, manual text, files from Media, URLs—run Learn again when the source changes. Same rule if Auto is off for a WordPress group: after you add, remove, or rewrite pages in that group, press Learn so the index matches the site.
Also retrain (or wait for Auto to finish) when the Playground still gives stale answers about topics you already fixed. You do not need to retrain for cosmetic changes in Appearance. Focus on facts: policies, features, shipping, support steps, anything a visitor might ask verbatim.
FAQ
Do I need an API key to train a WordPress chatbot?
Yes. Limb needs an API key from an AI provider for chat replies, and training (embeddings) uses the same or a related key depending on your setup. Add it under AI Settings before you run Learn. Step-by-step for each provider: API key guides.
What does Learn do in the Knowledge Base?
Learn reads the sources you selected, splits them into chunks, turns each chunk into vectors, and saves them in your knowledge store. After that, when someone asks a question, the chatbot can search those vectors and pull answers from your content instead of guessing. More detail: Knowledge Base.
How often should I retrain the chatbot?
When Auto is on for pages, posts, or a custom type, edits relearn on save—no fixed schedule. For other sources, run Learn when content changes. If nothing factual changed anywhere, you can leave the index as is.
Why are answers still wrong after training?
Usually the fact is missing, buried, or unclear on the page the chatbot retrieved—or the wrong page was indexed. Open the source in Playground, check which chunk was used, and fix the wording on that page. If you edited the page after training, run Learn again. Style issues (too long, too formal) are separate: adjust instructions under AI Settings, not the Knowledge Base.