Training Sources Overview

Everything your bot knows comes from its training sources — the pages, posts, files, and facts you feed it. This article explains what sources are, how to manage them from the Sources tab, and how they translate into the answers your customers see.

What a training source is

A training source is one piece of content your bot has learned: a webpage, a WordPress post, a PDF, or a fact you taught it in chat. When you add a source, HelpJet fetches the content, strips the boilerplate, splits it into chunks, and stores each chunk as a searchable embedding in your bot's knowledge base.

At chat time, the bot searches that knowledge base for the chunks most relevant to the visitor's question and answers from what it finds. If nothing relevant turns up, it says it doesn't have that information instead of guessing.

Two practical consequences:

• Your bot is only as good as its sources. If an answer is wrong or missing, the fix is almost always adding or updating a source.
• Sources take effect as soon as they finish processing. There's no separate "publish" step — a processed source is live immediately.

The Sources tab

Open any bot and click Sources in the sidebar (under Knowledge). The page has two parts:

1. An add area at the top — a single smart input labeled Training URL, plus an Upload menu for files. Enter a WordPress site, a sitemap, or any webpage and HelpJet detects the type automatically. New sources collect here until you click Start training.
2. The live source list below — every source the bot already has, with its type, status, priority, and when it was added. Filter tabs (All, plus one per type present) and a sort control (Priority, Recent, Status) keep long lists manageable.

Source types at a glance

Sitemap and WordPress sources are parents: the row you add expands into child sources, one per discovered page or post. Refreshing the parent re-discovers its children.

Source statuses

Every source shows a status badge in the list:

Processing happens in the background — you can leave the page and statuses keep updating. Failed fetches are retried automatically (up to 3 attempts with increasing delays) before a source is marked Failed.

Keeping sources fresh

Automatic refresh

HelpJet re-fetches your web-based sources on a rolling schedule: a processed source older than 30 days is marked stale and re-fetched automatically. You don't have to do anything for routine freshness.

When a parent (sitemap or WordPress site) is re-discovered:

• New pages are added as new child sources.
• Changed pages are refreshed in place — no duplicates.
• Missing pages are flagged as potentially removed after being absent from two consecutive successful re-discoveries. They're never deleted automatically — flagged sources keep answering until you remove them.

Manual refresh

• Click Refresh on any source row to re-fetch it now. Refreshing a parent re-discovers all of its children.
• Re-adding a URL the bot already has doesn't create a duplicate — HelpJet refreshes the existing source in place instead.
• Select multiple sources with the checkboxes and click Refresh Selected to refresh in bulk.
• Rows with failed children show a Retry failed action that re-queues only the failures.

Removing sources

Click Delete on a source row. A confirmation dialog explains what's removed — deleting a parent also removes its crawled child pages. After confirming, the rows disappear immediately and a toast gives you a short undo window before the deletion is final.

Once a source is deleted, its content is removed from the knowledge base and the bot no longer answers from it.

Source priority

Each source has a priority that influences how strongly its content is weighted. Click the priority value on a row to adjust it with a slider. Tiers:

You can also select several sources and set their priority in bulk. Facts you teach in Train & Test Mode are saved at high priority automatically, so corrections tend to win over older page content.

How training affects answers

When a visitor asks a question, the bot:

1. Runs a semantic search over your knowledge base for the most relevant content chunks.
2. Composes an answer from what it found, following your Brand & Voice instructions.
3. If nothing relevant was found, says it doesn't have information on that topic.

This is why source curation matters more than volume: five accurate, current pages beat fifty stale ones. Review the Activity tab regularly to see where the bot comes up empty, then add sources to fill those gaps — see Reviewing Conversations.

Common questions

How many sources can I add?

There's no hard cap on total sources, but each training batch processes up to 300 items at a time. If a sitemap or WordPress site has more, train the most important content first and add the rest in another batch.

How long does processing take?

Single pages and files usually finish in seconds. Large sitemaps and WordPress sites are fetched page by page with deliberate spacing to be polite to your server, so hundreds of pages can take a while. Statuses update live in the list.

Does adding sources cost interactions?

No. Interactions are billed on bot responses to users, not on training. See Understanding Interactions.

The bot answers from a page I deleted from my website — why?

The source is still in your knowledge base. HelpJet flags children that disappear from a sitemap or WordPress site, but never deletes them automatically. Remove the source from the Sources tab and the content stops being used.

Where to go next

• Training from a Website or Sitemap — the fastest way to bulk-train on your existing site.
• Training from WordPress Content — post-type selection and private-site authentication.
• Files, YouTube & Text Sources — PDFs, text files, and other content.
• Train & Test Mode — teach the bot individual facts and corrections by chatting with it.