Training from your website

#Create the training

1. Open the website page

   In Chat agentTraining, press New training, then Select Website. “Website Training” opens.

2. Choose the method

   In “Import Method”, choose a tab, paste the address and press the button. The three methods are explained below.

3. Wait for the page search

   “Scanning in progress” appears, then the number of pages found, for example “24 pages found”. The pages found go into the “Extracted sources” column, with no duplicates: you can combine several methods.

4. Choose the pages

   In “Extracted sources”, pages are grouped by domain. Remove the ones you don't need with the Remove from the list trash can. For a missing page, use Add Link, in the same column: it adds an address to the list you already have. It is different from Add link, the button on the Manual tab.

5. Train

   Press Train Agent. “The training will run in the background” appears, meaning it carries on by itself while you do other things, and you go back to the list.

#How to check

From the website's ⋮ menu, choose Edit training, select a few pages and read the text on the right: it is what the agent knows. Then, in the Test Chat, ask the agent a question whose answer is on one of those pages.

#The three methods

The sitemap is the list of pages that many websites publish, often at the address /sitemap.xml. With Whole Site, the system uses it if it finds one. Otherwise it follows the links on the page you entered and on the pages they lead to, only on the same domain. The farthest pages are left out: for a large website, use the Sitemap XML tab.

With Whole Site, both when it uses the sitemap and when it follows links, and with Sitemap XML, the system discards links to PDFs and images, and privacy policy, cookie policy and terms and conditions pages.

#Messages during the search

If you need to redo the list, Remove all empties “Extracted sources” after a confirmation.

#Reading modes

In “Content Extraction Mode”, you decide which part of each page to read, with Reading mode.

Automatic Default

The system finds the main content of the page. It works for most websites.

CSS selector

An option for whoever built your website: leave it to them.

Whole page

Reads everything, including menus and footers. Use it when automatic mode leaves out useful parts.

#Pages with problems

When some pages can't be read, a yellow triangle with “N pages with problems” appears in the list next to the website name. To see them, open Edit training and press the N pages with problems filter. When you select a page, the reason appears on the right. The most common cases:

For other reasons, try again later; if the reason concerns the website, for example its security certificate, have whoever manages it fix it. Then retry the pages with Re-run training, or remove them: in Edit training, with the filter on, press Remove pages with problems and then Train Agent.

#Automatic sync

Sync rereads the website at regular intervals and updates the knowledge when the pages change. You open it from the calendar icon on the website's row, when the training isn't in progress.

Last Run

How many pages the last run found “Changed”, “New” and “Removed”, with the date of the “Last check”. With Run now you start a run right away, without waiting for the frequency: “Update started: the training is queued” appears. After a run started by hand, you have to wait about two hours before starting another one for the same training. If you try earlier, “You started an update a short while ago. Try again later.” appears.

Enable automatic sync Yes / No

Turns on automatic runs. If your plan doesn't include it, “Your plan does not include automatic website sync.” appears.

Source Sitemap · Site scan (crawl)

Where the system gets the list of pages at each run: the sitemap, or the website's links. “Crawl” is the name for the reading that follows links from one page to the next, as Whole Site does. Choose Sitemap when the website has one.

Frequency Daily (24h) · Every 36 hours · Weekly · Monthly

How often to reread the website. Frequencies your plan doesn't include show a padlock.

Sync rules

Automatically add newly found pages: new pages go into the knowledge base, at most 50 per run and only if there is room in your plan.

Remove pages no longer present: removes the pages that have disappeared from the website. It works only with the Sitemap source. Both rules start out on.

Press Save settings: “Settings saved” appears and the date of the next run appears on the website's row.

#Which pages to exclude

The general rules are in Good-quality sources. For a website, also remove from the list:

• search results, tag or category archives, numbered pages of lists;
• the same pages in other languages, if the content is identical: the agent answers in the customer's language anyway.

#Frequently asked questions

#See also