Refresh or remove a source

An update must remove the old truth

Contradictory answers often come from two active sources describing different versions. Before retraining, identify every occurrence of the old information: manual page, crawled page, document, pasted text or ecommerce catalogue.

Deactivate, retrain and delete serve different needs. Deactivation is reversible and useful for comparison. Retraining rebuilds source knowledge. Deletion permanently removes the record.

Change rule
Do not leave a replaced policy and its new version active at the same time. A transition period should be explicitly described in one source.

Choose the right action

Start with the safest reversible option, then delete only after validation.

• Temporarily deactivate

  Use this to isolate a contradiction or wait for a publication date. The source stays listed but should no longer influence new answers.

• Retrain after modification

  For a URL, publish the new page first and then rerun the source. For a file, replace it through the available form or create a clearly named version and deactivate the old one.

• Delete permanently

  Delete duplicates, removed content or unauthorized documents. Bulk action helps cleanup, but review the selection because restoration is not guaranteed.

Refresh a full site without reintroducing noise

Persistent exclusions are essential during a new crawl.

1. Keep exclusions

   A persistently excluded page should not return merely because the source is rerun. Review intentionally restored URLs before training.

2. Monitor monthly allowance

   A refresh may rediscover many pages. Clean up parameters and filters, then check remaining URL allowance to avoid partial synchronization.

3. Wait for background jobs

   Counters change during discovery and processing. Test the new policy only after relevant pages are processed.

Prove the old information is gone

One positive test is not enough: use a small regression test.

1. Old wording question

   Ask using the wording customers used before the update. The answer must no longer present the old rule as current.

2. New wording question

   Verify the new value, effective date and exceptions exactly as published.

3. Duplicate search

   Search titles, URLs and crawled pages for copies of the old content. A store connection may also supply newer data than an old PDF.

Recommended maintenance log

Keep a simple record explaining why a source changed.

• Date and owner — Who approved the new content and when.
• Affected sources — Title, URL and action: deactivation, retraining or deletion.
• Control questions — Expected answers before and after publication.