# How to prepare documents for an AI chatbot

URL: https://instantknowledgebot.com/learn/preparing-documents-for-ai-chatbot  
Last updated: 2026-10-17

> To prepare documents for an AI chatbot, keep one current version of each fact, put the answer first in plain text, use clear headings and question-and-answer pairs, include specifics such as prices, timeframes and contact details, and remove outdated or contradictory files. Document quality affects a knowledge base chatbot's answers more than any other setting.

The best way to improve an AI chatbot's answers is to **prepare documents for an AI chatbot** carefully. A knowledge base chatbot answers from what you give it, so clear, current and specific documents produce clear, current and specific answers. This guide gives you a practical checklist, with examples, that works for any knowledge base chatbot, including InstantKnowledgeBot.

## Why do documents matter more than the AI model?

A knowledge base chatbot finds the passages relevant to a question and asks an AI model to answer from them. If the passage is missing, vague, outdated or contradicted by another file, even the best model will struggle: it either says it doesn't know or, worse, picks the wrong version. Improving the source is almost always the fastest fix. See [AI chatbot hallucinations](/learn/ai-chatbot-hallucinations) for how errors arise.

## What should you upload first?

Start with the documents that answer most questions:

1. **FAQ**: your existing one, or the top 20 questions from your inbox with answers.
2. **Policies**: returns, cancellations, shipping, privacy summary, terms that customers ask about.
3. **Products or services**: what you offer, who it's for, key details and prices or ranges.
4. **Practical information**: hours, locations, contact details, booking steps.
5. **How-to guides**: setup, troubleshooting and common tasks.

Leave out marketing fluff, old announcements and drafts.

## What are the rules for chatbot-friendly documents?

**1. One source of truth.** Keep a single, current version of each fact. If you upload two price lists, the bot may quote either. When something changes, replace the file rather than adding a new one.

**2. Answer first.** Put the key fact in the first sentence.

- Weak: "We know how important timely delivery is, and we work with several carriers to…"
- Strong: "Standard US shipping takes 3 to 5 business days and is free over $50."

**3. Be specific.** Numbers, timeframes, prices, names, links, phone numbers and conditions. "Usually quick" can't be quoted usefully; "within 24 hours on weekdays" can.

**4. Use clear structure.** Headings per topic, short paragraphs, bulleted lists and numbered steps. Structure helps the system find the right passage and helps the model quote it accurately.

**5. Name things consistently.** If your plan is "Premium" on one page and "Pro" on another, the bot may treat them as different things. Pick one name.

**6. Write the conditions and exceptions down.** "Except sale items", "outside the US", "for new patients only": edge cases are where wrong answers happen.

**7. Put facts in text, not images.** The chatbot reads text. Prices in a graphic, a scanned price list or a chart aren't readable. Add text versions.

**8. Include questions in your customers' words.** If customers ask "Do you take walk-ins?", include that phrase near the answer, even if your official term is "unscheduled appointments".

## How should you structure an FAQ for a chatbot?

Question-and-answer pairs work very well:

```text
Q: Do you offer same-day appointments?
A: Yes, we keep same-day slots for urgent problems. Call (555) 010-0000 before 10 a.m.

Q: Which insurance plans do you accept?
A: We accept Delta Dental, Cigna and MetLife. We file claims for you.
```

InstantKnowledgeBot's industry templates include a starter FAQ in this format, with placeholders to replace; browse the [templates](/templates).

## Which file formats work best?

| Format | Notes |
|---|---|
| **TXT / Markdown** | Clean and simple; headings in Markdown help structure. |
| **DOCX (Word)** | Good; headings and lists are preserved as text. |
| **PDF** | Good if text-based. Scanned PDFs need OCR first. |
| **HTML** | Good for saved web pages; scripts and styles are stripped. |
| **CSV** | Good for catalogues and listings: one row per item with clear column names. |
| **JSON** | Works for structured data exports. |
| **Web pages (import from URL)** | Imported straight from your site: a single page, the whole site or a sitemap. Menus, footers and cookie banners are left out; pages that need JavaScript to show their text can't be read. |

In InstantKnowledgeBot, each file can be up to 50 MB, with 25 files per upload, and you can paste text directly. Website imports take up to 10 pages per import on Free, 50 on Starter, 200 on Pro and 500 on Agency; the same advice applies, so remove outdated pages from the list after importing. See [how to build a chatbot from your website URL](/guides/build-chatbot-from-website-url). More on PDFs: [how to train a chatbot on PDFs](/guides/train-chatbot-on-pdf).

## How do you handle large or technical documentation?

- **Split by topic:** one file per product area or chapter.
- **Remove boilerplate** repeated on every page, such as legal footers and navigation.
- **Keep code examples** in code blocks with a sentence explaining each.
- **Label versions** clearly, or remove docs for retired versions.

When the knowledge base is large, the chatbot searches it for each question, as explained in [RAG chatbots explained](/learn/rag-chatbot-explained). Clear headings and focused files make that search more reliable.

## What shouldn't you upload?

- **Confidential or personal data:** anything in the knowledge base may be quoted to whoever uses the bot.
- **Internal notes and drafts.**
- **Outdated versions** of current documents.
- **Content you don't have the right to use.**

Use the bot's **extra instructions** for behavioural rules ("Keep answers under three sentences", "Never quote delivery dates") rather than putting them in documents.

## How do you keep the knowledge base current?

1. **Assign an owner** for each document.
2. **Update when facts change**: prices, hours, policies. In InstantKnowledgeBot, the live bot updates immediately.
3. **Review knowledge gaps** in the bot overview: questions your documents couldn't answer.
4. **Read thumbs-down answers** and fix the source.
5. **Re-test** with your standard questions after big changes.

## What does a before-and-after document look like?

**Before:**

> Our team is passionate about getting your order to you as quickly as possible. Delivery times may vary. Please contact us with any questions about shipping.

**After:**

> **How long does shipping take?** Standard US shipping takes 3 to 5 business days and is free on orders over $50. Express shipping takes 1 to 2 business days and costs $15. We ship to Canada (7 to 10 business days, $20). We don't ship to other countries yet. Questions: support@example.com.

The second version answers six different questions accurately. The first answers none.

## What's the quick checklist?

- [ ] One current version of each fact
- [ ] Answer first, specifics included
- [ ] Clear headings, lists and steps
- [ ] Consistent names
- [ ] Conditions and exceptions written down
- [ ] Facts in text, not images
- [ ] No confidential data or drafts
- [ ] Owner and update routine

Ready to build? [Sign up free](/signup) and [create a chatbot from your website FAQ](/guides/create-chatbot-from-website-faq), or compare plans on the [pricing page](/pricing).

## Frequently asked questions

**Which file format is best for a chatbot knowledge base?**

Any format with real, selectable text works. Plain text, Markdown, Word (DOCX) and text-based PDFs are all good. Avoid scanned image-only PDFs unless you add a text layer with OCR.

**Should I upload one big document or many small ones?**

Several focused documents (one per topic) are easier to maintain and help the chatbot find the right passage. One big document works too if it has clear headings.

**Do I need to rewrite everything as questions and answers?**

No. Normal well-structured documents work. Question-and-answer pairs help for FAQs, and adding a few common questions in your customers' own words can improve matching.

**How often should I update the chatbot's documents?**

Whenever the underlying facts change (prices, hours, policies) and after reviewing knowledge gaps. A monthly check is a good habit.

**Can I include internal notes for the bot?**

Use the bot's extra instructions for rules about tone and behaviour. Anything in the knowledge base may be quoted to visitors, so don't upload private notes.