Training Context

Why training context matters

Your assistant's persona tells it how to behave. Training context tells it what to say.

Without training context, your assistant can only rely on general knowledge from its underlying AI model — which means it won't know anything specific about your products, policies, pricing, or processes. Adding training context is what transforms a generic AI into an assistant that actually knows your business.

Go to AI Assistant > Training Context to manage your knowledge base.

The six types of training context

The Training Context page is divided into six main tabs, each designed for a specific type of information:

1. Products

Import your product catalog so your assistant can answer questions about what you sell, pricing, features, and availability. You can import products via CSV upload, JSON upload, or by providing a URL to a CSV or JSON feed.

If your product feed encounters a sync error, you'll see a detailed error message and a Retry sync button to manually trigger a fresh pull.

2. Documents

Upload PDF, DOCX, or plain text files containing your policies, product documentation, onboarding guides, or any other reference material. Autoch.at will extract the text and make it available to your assistant.

3. UI Screens

Upload structured UI Screens context files. This helps the AI answer questions about specific screens in your application without mixing them into general documents.

4. Website

Provide a URL and Autoch.at will crawl the page and pull in its content. This is the fastest way to get your existing documentation, help center, or product pages into the system.

Website crawls run as background jobs. You'll see a live progress banner showing how many pages have been crawled and the current queue size. After the initial crawl, you can re-crawl a link at any time to pick up changes.

5. Instructions

Custom instructions are for specific scenarios or edge cases. These are like sticky notes for your AI — "always mention our 30-day guarantee" or "if someone asks about X, explain Y". They're great for business rules that don't fit neatly into documents or products.

6. FAQs

FAQs are the most targeted type of training context. Each FAQ is a specific question-and-answer pair. FAQs take priority over other context types when the customer's question closely matches the FAQ question.

You can add FAQs in three ways:

• Add manually — Write individual question-and-answer pairs directly in the app.
• Import CSV — Upload a spreadsheet of FAQs.
• Import from URL — Paste a link to your website's FAQ page (or paste the HTML directly). Autoch.at will scan the page, extract the questions and answers, let you preview them, and import them as drafts.

Managing your knowledge base

Keeping context up to date

Stale training context leads to wrong answers. Make it a habit to review your knowledge base after major product releases, policy changes, or pricing updates.

1. Go to AI Assistant > Training Context.
2. Find the outdated item across the various tabs.
3. Edit, re-sync, or re-crawl it to reflect the current information.
4. Delete anything that's no longer accurate.

Checking what's being used

You can see which pieces of training context your AI is using most frequently in Insights > Analytics under FAQ Analytics. If a document is rarely referenced, it might be outdated or too general. If a particular FAQ is being used constantly, consider expanding on that topic in your main documentation.

Training context vs. messaging directives

These two features are easy to confuse, but they serve different purposes:

If you want your AI to know your return policy, add it to Training Context. If you want your AI to always end every response with a follow-up question, use a Messaging Directive.