Knowledge freshness

• Docs, pricing, or policies change weekly and answers must reflect it.
• Seasonal content (promos, feature flags) should expire automatically.
• You see regressions tied to outdated training or indexed data.

Freshness drives trust and conversion. PMs set SLAs for content recrawls, expiry, and alerts. Freshness tradeoffs: more updates cost compute and risk index instability; fewer updates risk wrong answers and lost deals. It also impacts SEO if public content is involved.

Tag content with last-reviewed and expiry dates; set recrawl schedules per collection importance. Use canaries and evals focused on recent changes. In 2026, stream change events from CMS to trigger targeted re-indexing and auto-archive expired chunks to keep latency low.

After adding weekly recrawl plus expiry on promo FAQs, a commerce assistant removed outdated offers within 24 hours. Refund tickets dropped 18% and conversion on chat-assisted checkouts rose 6% while keeping p95 latency under 1.2 s.

• Treating all content equally, wasting cycles on low-value or static pages.
• Leaving expired promotions indexed, causing user-facing mistakes.
• Not measuring freshness-specific KPIs, so regressions go unnoticed.