Documentation & Knowledge Base

Keep AI answers in sync with your documentation

A documentation release flowing through recapture to a verified answer
A documentation release flowing through recapture to a verified answer

Publishing a documentation change and refreshing an assistant are two separate events. A recurring capture job connects them by revisiting the selected docs and processing the current text into the dataset.

The schedule should follow the documentation release process, not an arbitrary SEO target.

Match jobs to documentation versions

If current and legacy docs live under different paths, capture them into separate datasets or exclude the versions that should not answer public questions.

For example, include the current docs path and exclude archive or versioned paths that no longer apply. Review allowed hostnames when docs use subdomains.

Choose the schedule

Publishing patternStarting approach
Continuous small editsDaily capture
Several releases each dayMulti-hour capture if freshness requires it
Weekly release trainCapture after release or every seven days
Controlled compliance docsManual capture after approval

Seekdown provides intervals from every hour through every seven days, plus Never.

Seekdown schedule selector for documentation capture with sample data
Seekdown schedule selector for documentation capture with sample data

Add freshness checks to the release

For each material documentation change:

  1. name the changed page;
  2. record one question affected by the change;
  3. confirm the next capture finishes;
  4. find the new wording in the dataset;
  5. ask the recorded question; and
  6. open the citation.

This is more reliable than checking only the capture timestamp. A finished job may still have missed a URL because of path rules, depth, or a page error.

Handle removals and contradictions

When a workaround or old option is removed, search the remaining dataset for the old wording. Release notes, translated pages, and cached copies can continue to support the obsolete answer.

Do not ask the assistant to choose silently between conflicting docs. Designate one current source, remove the obsolete page, or state the version boundary explicitly.

Recurring crawling keeps the source copy moving. Regression questions show whether the answer moved with it.