Keep your chatbot up to date
Use a sitemap, set Auto-retrain, clear out stale files and text, and check the answers after each retrain so your chatbot stays accurate as your website changes.
Updated 8 Oct 2026
Your chatbot answers from the knowledge sources you add under Knowledge. It doesn't browse your website live, so when a price, service or policy changes, the chatbot only knows once the source that holds it has been trained again. This article covers how to set that up so it mostly looks after itself.
Prefer a Sitemap source over single pages
On Knowledge → Add source, the Sitemap tab takes your site's sitemap.xml (for example https://example.com/sitemap.xml) and reads the pages it lists as one source. A URL source reads only the one page you enter.
A sitemap is the better choice for most websites. New pages you publish are listed in your sitemap, so they are picked up the next time the sitemap source is trained, without you adding them one by one.
How much of a sitemap is read depends on your plan:
| Plan | Pages read from one sitemap | Child sitemaps read from a sitemap index |
|---|---|---|
| Free | 10 | 3 |
| Starter and Founding | 50 | 5 |
| Pro | 100 | 8 |
| Business | 200 | 12 |
| Agency | 300 | 15 |
Pages are read in the order the sitemap lists them, and reading also stops once your plan's knowledge-base character allowance is reached. If your sitemap is long, put your most important pages first, or add a smaller sitemap that lists just the pages customers ask about.
Turn on Auto-retrain
URL, Sitemap and YouTube sources can retrain themselves. Open the source on the Knowledge page and choose Auto-retrain:
- No auto-retrain (the default)
- Daily
- Weekly
- Monthly
Pivra checks scheduled sources once a day at 2 AM UTC. A source is retrained at that check once 1 (Daily), 7 (Weekly) or 30 (Monthly) calendar days (UTC) have passed since the day its last retrain finished, so a Daily source retrains at every daily check even if its last retrain finished a little after 2 AM. A source that is already training is skipped until the next check. A source that failed last time is tried again at its next scheduled check.
Weekly suits most business websites. Use Daily only for information that really changes every day. Files and text sources have no schedule, because nothing outside Pivra changes them: update them yourself (see below).
What the status column means
| Status | Meaning |
|---|---|
| Trained | The source is in the chatbot's knowledge and is used for answers. If some pages or sections couldn't be read, the source still trains with the rest, and a note under its name lists what was skipped and why. |
| Training… | Pivra is reading and processing the source. If that fails for a reason that may pass, such as your website being briefly unavailable, Pivra tries again after 1 minute and then after 5 minutes, and the source shows Training… meanwhile. |
| Failed | The source couldn't be read after 3 attempts, or for a reason a retry can't fix (for example a page that doesn't exist, or a video without captions). Hover over the badge for the reason. The workspace owner also gets one email. |
| Pending | The source is waiting to be processed. |
Last trained shows when a trained source was last updated. That is normally when its last training finished, but saving a change to the source's title or schedule updates it too.
Remove or update stale files and text
Files and text sources stay exactly as you added them. If an old price list, brochure or policy is still in your knowledge, the chatbot can find it alongside the new one and give a wrong or mixed answer. Conflicting sources are one of the most common causes of wrong answers.
When something changes:
- Delete the old file or text source on the Knowledge page, then add the new version.
- Or open a text source, edit it, and click Save & Retrain.
Keep one source for each fact where you can. If you have a single page that sums up your business (services, areas, prices, hours, policies), add it as a URL source, or paste it as a Text source. Then remove older documents that say the same things differently.
When to press Retrain yourself
Every source has a Retrain button, and Retrain selected retrains several at once. Retrain straight away, rather than waiting for the schedule, when:
- a price, fee or package has changed on your website
- you've added or removed a service, or changed your service area
- your opening hours, policies or contact details have changed
- you've fixed a page that previously failed to train
What happens during a retrain
A retrain clears the source's current content first, then reads it again from the start. While a source shows Training…:
- That source's content isn't available to the chatbot, so questions that rely only on it may get "I don't know" or a referral to your team until training finishes. Your other trained sources keep working.
- Strict mode stays on. A source being retrained still counts as knowledge, so the chatbot keeps answering only from your trained material.
The new content is used as soon as the retrain finishes. If a retrain fails, that source has no content until you fix the problem and retrain it, so check the Knowledge page after a big change. For a large sitemap, retrain at a quiet time of day.
Check the answers after retraining
When the source shows Trained again, open Test in the dashboard and ask the five questions your customers ask most, for example:
- What does it cost?
- Do you service my area?
- When are you open?
- How do I book?
- What isn't included?
If an answer is wrong, find the source it came from and fix or remove it. For answers that must always be worded exactly, such as prices or legal wording, add a refinement. See also Training your chatbot and Improving chatbot responses.
Was this helpful?
If you have questions or suggestions, email us at support@pivra.ai .