LLM Best Practices

Content refresh strategy

5 min read

View markdown source

Overview

A content refresh is a substantive edit to an existing page that updates facts, intent, or depth, paired with a new dateModified. Done right, it recovers rankings on pages that have drifted from the SERP. Done wrong (a date bump without an edit) it triggers Google’s fake-freshness downweight under the helpful-content-update system. The rule: edit the body first, change the date second.

Refresh when one of four triggers fires

Refresh on signal, not on a calendar. Four triggers justify the work.

  • Facts changed. A version released, a price moved, an API deprecated, a statistic updated. If the page asserts something that is now false, fix it.
  • SERP intent shifted. The dominant result type for the query changed from tutorials to comparisons, or from definitions to news. Match the new format.
  • A competitor outranks the page with newer information. Open the top three competing URLs and identify the gap.
  • Google deprecated a feature the page references. Search Console reporting changes, retired schema types, and ranking factors being phased out all warrant an edit.

Refreshing because it has been twelve months since the last update is not a trigger. The classifier checks content diffs, not the calendar.

Find drift through GSC plus a manual SERP check

The detection loop has two halves.

  • Google Search Console: Performance > Pages, filtered to the page URL. A page that has lost more than 25 percent of clicks year-over-year, or whose average position dropped more than three spots in a quarter, is drifting.
  • Manual SERP check: query the page’s primary keyword in an incognito window. Read the top five results. Note the publish dates, the format, and any claim the page makes that the SERP contradicts.

Pages that show drift in GSC but match the SERP on intent and freshness do not need a refresh; the loss is from algorithm volatility, not content quality. See audit-checklist for the full diagnostic loop.

Rewrite at least 30 percent of the body

The threshold for Google’s content-diff fingerprint to register a real edit is not published, but the working rule across SEO practitioners is that a substantive refresh rewrites at least 30 percent of the page. Token-level changes (swapping a synonym, reordering a sentence) do not count.

A refresh that meets the threshold includes some combination of:

  • A new H2 section with original information (a benchmark, a screenshot, a workflow).
  • A rewritten introduction that addresses the current SERP intent.
  • Updated tables or code blocks with the current versions of the tools.
  • Deleted sections that no longer apply, with their wikilinks redirected to current pages.

Light touch-up edits to fix a typo or update one date are maintenance, not a refresh. Do not bump last_updated for maintenance.

Update freshness signals in lockstep after the edit

Once the body is rewritten, update all four freshness signals in the same commit. Stale dates erode trust faster than no date.

  • last_updated in the page frontmatter.
  • A visible “Last updated” line near the title.
  • dateModified in Article JSON-LD.
  • <lastmod> in the sitemap.xml entry for the URL.

Then notify the engines. Submit the URL through indexnow for Bing and Yandex; Google picks up the change on its next crawl, which the sitemap lastmod accelerates. Do not re-publish the page at a new URL; preserve every inbound link.

Preserve the URL across refreshes

The single most expensive refresh mistake is moving the content to a new URL. Every backlink, every internal link, every ranking signal accumulated over years lives on the existing URL.

  • Keep the slug. If the topic shifted enough that the old slug is misleading, you are writing a new page, not refreshing.
  • If the URL absolutely must change, 301 the old URL to the new one. Inbound link equity transfers but not perfectly; expect a 5 to 15 percent decay on the redirect.
  • Update internal links across the site to point at the new URL after the redirect is in place. See redirects.

Do not delete helpful content to “clean up”

Pruning low-traffic pages is a common audit recommendation and a frequent mistake. A page with no clicks but real information is still a citation target for AI search and a link destination for other pages on the site. Three rules.

  • Noindex before delete. If a page seems dead, noindex it and observe for four to six weeks. Many low-traffic pages are passing link equity invisibly.
  • Merge before delete. If two pages overlap, merge the weaker into the stronger and 301 the URL. Do not 404.
  • Delete only if the content is wrong, harmful, or genuinely orphaned with no inbound signals. See content on pruning.

Common errors

  • Redating without editing. The content-diff fingerprint catches it and the page receives a downweight under HCS.
  • Refresh plus new URL. Ranking signals do not fully transfer through a 301; preserve the URL.
  • Touch-up edits with a last_updated bump. The classifier treats this as fake freshness.
  • Deleting “low-traffic” pages without a noindex test first. Many pages have no clicks but real link-equity contribution.
  • Refreshing on a calendar instead of a signal. Pages that still match the SERP do not need edits; pages that have drifted do.

Graph View

Backlinks