AI chatbot, the kind that does not embarrass you.
An AI chatbot reads a customer message and writes a reply each time from scratch, using a language model. That is the only thing it shares with the chatbots of 2018. Everything else, the keyword trees, the if-this-then-that flows, the canned dead-ends, is gone. The hard part now is keeping it grounded in your actual business.
An AI chatbot writes each reply on the fly using a language model, grounded in your own content. Unlike old rule-based bots it can handle paraphrasing, edge cases, and unknown questions gracefully. The trick is keeping it honest. The good ones cite their sources and admit when they do not know.
- ✓No keyword trees, no flow builders, no "sorry, I did not understand that" dead-ends.
- ✓Each reply is generated fresh from your content, so adding a new FAQ takes seconds, not weeks.
- ✓Handles paraphrasing and edge cases natively. Customers stop hitting the dead-end wall.
- ✓Cites the source page it used. You and your customers can verify the claim.
AI chatbot versus the chatbot you remember
The old chatbot was a decision tree. It matched a keyword to a branch and read out a canned line. Customers hated it because rephrasing a question dropped them into nowhere. An AI chatbot reads the meaning of the question and writes a new sentence each time. It can still get things wrong, but the failure mode is different. With the right safeguards, an AI chatbot admits the gap and hands the conversation to a person. The old kind silently led you in circles.
What an AI chatbot can actually do
Three jobs, well. First, answer factual product and policy questions from your own content. Second, hold a multi-turn conversation that remembers what was said two messages ago. Third, take a structured action when allowed: look up an order, attach a return label, qualify a sales lead. The fourth thing, taking unsupervised destructive actions, is the trap most people walk into. Good AI chatbots stay read-only by default and ask before they change anything.
Where it fails (and how to spot it)
AI chatbots fail in two ways. They invent facts when retrieval is broken, and they leak personal data when redaction is missing. Both are solvable. Demand citations on every reply you read in the demo. Ask the vendor how personal data is handled before any model call. If a chatbot replies confidently to an out-of-policy question without citing anything, walk away. The line between a useful AI chatbot and a liability is whether the safeguards are real or marketing.
What an AI chatbot needs from you
Less than you think. A working public website, a clear list of policies (shipping, returns, refunds), and a couple of FAQs is usually enough to start. The biggest lift is not technical, it is editorial. If your site is silent on returns, the chatbot will be silent too. If your shipping policy contradicts itself, the chatbot will pick one of the two and answer confidently. Cleaning up your own content makes the chatbot better than any tuning trick.
Where Keloa fits
The Keloa AI chatbot is on every plan, including the free Starter. Paste your URL, the AI reads your content and starts answering in a sandbox in minutes. Drop the widget snippet on your site when you are ready. Citations are on by default, the AI escalates when it is unsure, and personal data is redacted before any model call. Hosted in the EU, no shared training pool, free Verwerkersovereenkomst (DPA).
Frequently asked questions about AI chatbots.
What is the difference between an AI chatbot and a normal chatbot?
A normal chatbot matches keywords to scripted answers. It breaks when a customer rephrases a question. An AI chatbot uses a language model to write a fresh reply each time, grounded in your content. It can handle paraphrasing, edge cases, and unknowns without dropping the customer into a dead-end. The chatbot you remember from 2018 is not the same product category.
How do I add an AI chatbot to my website?
Sign up, paste your site URL, and copy the script tag the product gives you into your site head. That is it. For Keloa the path is the same on every plan. The chatbot indexes your public pages, you tweak the tone in the sandbox, then ship the snippet. Most teams are live in under ten minutes.
Does an AI chatbot need training?
Not in the old sense. You do not train a model from scratch. You point the AI chatbot at your existing content (website, FAQs, PDFs, policies) and it builds a retrieval index. When a question comes in, it looks up the relevant facts and writes an answer. Updates are instant: change your shipping policy on your site, the chatbot reflects it next time someone asks.
Can an AI chatbot do more than answer questions?
Yes, when you give it tools. The Keloa chatbot can look up an order or a product on a connected Shopify store, query a custom HTTP endpoint of yours, and qualify a sales lead with structured questions. Destructive actions (changing inventory, issuing refunds) are deliberately off by default. We prefer the AI proposes the action and a human confirms.
Will an AI chatbot annoy my customers?
It depends on whether it is honest. An AI chatbot that admits uncertainty and hands cleanly to a human is a relief for most customers. One that pretends to know and gives a wrong answer is the chatbot that ruins your reputation. We tune Keloa toward the first. Confidence below seventy percent? It does not send. It escalates with a clean summary your team can pick up.
What does an AI chatbot cost?
Less than you think. Keloa starts free, fifty replies a month, no card. Paid plans start at €49 a month for fifteen hundred replies and go up by reply volume, not by seats. Top up at €30 per thousand replies on any plan. If your chatbot handles thirty conversations a day, you are usually well inside Growth or Business.
Drop an AI chatbot on your site in ten minutes.
Free Starter plan, fifty replies, no credit card. Paste your URL and watch the chatbot answer real questions before you commit to anything.