# A chatbot that answers with YOUR data: how a knowledge base works

> A generic chatbot invents answers because it has no information about your company. A knowledge-base chatbot answers only with what you give it — FAQs, catalogues, policies — and resolves 90%+ of enquiries without hallucinating. S/ 349 setup (Peruvian soles, IGV included), plans from S/ 39/month, in production in 48 hours.

- Canonical: https://fluxr.pro/blog/chatbot-that-answers-with-your-data-how-a-knowledge-base-works
- Publicado: 2026-07-14 · AI for Business
- Fonte: Fluxr Pro (https://fluxr.pro)

---

# A chatbot that answers with YOUR data: how a knowledge base works

**Direct answer: a generic chatbot invents answers because it uses internet data, not your company's. A knowledge-base chatbot only answers with what you give it — FAQs, catalogues, policies, manuals — which eliminates hallucinations and lets it resolve 90%+ of enquiries correctly.** At Fluxr, we load your information during the S/ 349 setup and the agent is in production within 48 hours, with private AI that never shares your data with third parties.

## Why does the "smart" chatbot you use today still get things wrong?

There's a fundamental difference between a generic chatbot and one with a private knowledge base. The first uses a general language model trained on billions of internet pages. When it doesn't know something, it invents it with disconcerting confidence — that's called hallucination, and in customer service it has real consequences.

A customer asks the price of a specific plan and receives a figure that doesn't exist. They ask about the returns policy and the bot describes the industry's standard policy, not yours. They ask about a product and the agent quotes specifications from an earlier version. The bot doesn't "know" it's inventing — it simply generates the most probable answer from internet patterns, with no reference to your company at all.

A knowledge-base chatbot operates on the opposite principle: **the agent can only answer with what's in your knowledge base.** If the information isn't there, it says so openly and escalates to the human team. No inventing. No hallucinating. No damage to the customer relationship.

## What information goes into a knowledge base?

It isn't about uploading a PDF and hoping the AI "understands" it. A well-structured base includes:

- **Answered FAQs** — not just the questions, but the exact answers you want the bot to give, in your brand's tone, at the level of detail that matters
- **Product or service catalogue** — current pricing, specifications, available variants, purchase terms
- **Company policies** — returns, warranties, delivery times, accepted payment methods
- **Operational documents** — user manuals, installation guides, onboarding protocols
- **Escalation flows** — which situations always go to a human, even if the bot knows part of the answer

The quality of the chatbot's answers is directly proportional to the quality of the information going in. If your catalogue has stale prices, the bot will quote them with total confidence. If your FAQs are vague, the bot will be too.

A good exercise before setup: write down the 20 questions your support team repeats most. Those are exactly the ones that should be in the knowledge base, answered at the level of detail your best agent normally gives.

## How does the knowledge base stay current?

This is the point most often ignored by teams deploying chatbots on their own: the knowledge base isn't static. Every time you change a price, launch a product or amend a policy, the base has to be updated — otherwise the bot starts giving incorrect information with its usual confidence.

Companies managing this alone typically fall into three common traps: they update the website but not the chatbot; the support team knows about the changes but nobody transfers them to the knowledge base; or months of stale information accumulate until a customer exposes it on social media.

At Fluxr, knowledge base maintenance is included in the monthly plan. You send the changes — an email, an updated spreadsheet, the new catalogue as a PDF — and the team integrates them before they reach your customers.

## Generic bot vs knowledge-base bot: a real comparison

| Criterion | Generic chatbot | Knowledge-base chatbot |
|---|---|---|
| Source of information | The internet (general public data) | Only your documents and FAQs |
| Hallucination risk | High — invents confidently | Minimal — answers or escalates |
| Answers about your company | Generic or incorrect | Precise and current |
| Tone customisation | Limited | Total — writes in your brand voice |
| Data privacy | May be processed by third parties | Self-hosted AI — data on your infrastructure |
| Cost of an error | High (misinformed customer) | Low (escalates to a human) |
| Deployment time | Minutes (generic template) | 48 hours (complete, tested agent) |

The privacy row deserves particular attention. Chatbots from popular platforms process conversations on third-party servers and, on some plans, use them to improve their models. At Fluxr, the AI is **self-hosted on our own infrastructure** — your customers' data never goes to OpenAI, Google or any external provider, and it isn't used to train any model.

## Fluxr's process: from your documents to an agent in production

The S/ 349 setup (Peruvian soles, IGV included) includes five steps before the agent takes its first real enquiry:

1. **Discovery session** — 30 minutes to understand which enquiries you receive and what information the agent needs to answer them correctly
2. **Document collection** — you share your FAQs, catalogues, policies and any relevant material
3. **Knowledge base structuring** — we don't just upload the files: we organise, clean and structure them so the agent consults them efficiently and accurately
4. **Agent configuration** — a custom prompt in your brand tone, clear human-escalation rules, active channels
5. **Integration and testing** — we connect to the channel you use (web, WhatsApp Business or API) and test with real cases before launch

The result: an agent that answers in **under 1 second**, handles unlimited simultaneous conversations and knows when to route to your team. You can see the real impact in [Fluxr's case studies](https://fluxr.pro/case-studies) or use the [ROI calculator](https://fluxr.pro/calculator) with your company's enquiry volume.

## When a knowledge base isn't the right solution

A knowledge-base chatbot handles informational and standard transactional enquiries well. It isn't the right tool for:

- **Negotiations or non-standard terms** — a customer wanting a special price or bespoke conditions needs someone with authority to decide
- **Deep technical support** — if the diagnosis requires step-by-step testing or expert judgement on variable symptoms, the agent has clear limits
- **Situations with legal implications** — contracts, disputes, claims where every word carries weight
- **Customers in intense emotional states** — serious complaints or urgent situations respond better to genuine human empathy

The chatbot's proper role is to resolve 70-90% of routine enquiries so your team can focus on the cases that genuinely require human judgement. You can compare the available automation options on the [plans page](https://fluxr.pro/pricing).

## Next steps

- **[Try the text agent live](https://fluxr.pro/text-agents)** — there's a real-time chat demo on the landing page showing exactly how Fluxr's private AI responds
- **[Compare the full plans](https://fluxr.pro/pricing)** across chatbots, voice agents and automations
- Want to know which documents to include or how to structure the knowledge base for your business? [Message us on WhatsApp](https://wa.me/51981095704) — we reply the same day with a free assessment

---

*Fluxr Pro builds private knowledge-base chatbots for businesses worldwide — native support in English, español and português.*