How to Build a Content Refresh Calendar for AI Search

No AI search engine publishes a “freshness window” after which it stops citing a page, and earlier versions of this guide repeated per-engine numbers that we could not source. They are gone. What is documented is simpler: AI answers are generated from the pages a system retrieves, so a page with an outdated price or a retired feature can be quoted with that error, or passed over for a page that is current. The calendar below is built around that, not around guessed decay curves.

Why Freshness Matters for AI Answers

• Wrong facts get repeated. If your page says a plan costs $29 and it now costs $39, an assistant that cites the page can state the old price with your name attached. That costs more than a lost click.
• Time-sensitive questions favor current sources. “Best X in 2026”, “current pricing”, “does X still support Y”: a page that visibly covers the current state answers the question; one that doesn't, can't.
• Fake freshness is a known pattern. Google's helpful content guidance lists, as a warning sign, changing the date of pages to make them seem fresh when the content has not substantially changed.

For how a page gets picked and quoted in the first place, see how AI search retrieves, reranks and cites passages. A refresh can't rescue a page whose answer is buried; fix structure first, then keep it current.

10 Steps to Build the Calendar

1. Step 1: Pick the pages the calendar defends

   Start from the pages that matter for AI answers: the ones that already earn AI referrals or citations, the ones that answer your most important buyer questions, and the pages that hold facts people ask about (pricing, product specs, policies, comparisons). Use the same page list your AI-visibility tracking uses, so refresh work and measurement point at the same pages.

2. Step 2: Record each page's last real update

   For every page, note the date its content last changed in a meaningful way, not the last deploy. Pull it from your CMS revision history or git log. Many sites find their visible dates, structured data dates and sitemap lastmod all disagree; this step is where you find out.

3. Step 3: List the facts on each page that can go stale

   Write down what on the page can become wrong: prices, plan limits, feature lists, model or product names, competitor details, legal or platform rules, screenshots, statistics with a year attached. A page with none of these (a definition, a timeless how-to) needs far less attention than a pricing comparison.

4. Step 4: Tier pages by volatility, then by value

   Volatility sets the cadence: how fast do this page's facts change in your market? Value sets the order inside a tier: AI referrals, citations, conversions. A useful default is three review tiers (monthly, quarterly, yearly) plus a fourth bucket for pages to merge or retire. Adjust the intervals to your category instead of borrowing someone else's.

5. Step 5: Set triggers alongside the schedule

   Some updates can't wait for the calendar: a price change, a product launch or retirement, a competitor change you reference, a platform policy update, or an AI answer that misstates something on the page. List the triggers and who watches each one; a trigger moves the page to the top of this week's queue.

6. Step 6: Define what counts as a real update

   A refresh changes what the page says: corrected facts, new examples, a rewritten answer, an added section for a question people now ask, updated screenshots or product images. Changing the date, the copyright year or a few words is not a refresh, and the date should not move for it.

7. Step 7: Make every date come from one field

   The visible 'Last updated' date, the dateModified in structured data and the sitemap lastmod should all derive from the same stored value, set only when a real update ships. Google asks that visible and structured dates match, and uses lastmod only when it is consistently accurate.

8. Step 8: Run the review, then log it

   For each page due: check every volatile fact against its source, re-run the main questions the page answers in the AI assistants you track and compare the answers to your page, then update or confirm. Log the date, the reviewer, what changed (or 'checked, no change needed') and whether the date moved.

9. Step 9: Merge or retire pages that keep failing

   Pages that are never cited, duplicate a stronger page, or cover the same question as a sibling should be merged into the best page with a 301 redirect, not refreshed forever. Fewer, more complete pages are easier to keep current.

10. Step 10: Review the calendar itself every quarter

    Move pages between tiers based on what actually changed during the quarter and which pages earned citations or referrals. Promote pages that became important, demote pages whose facts proved stable, and remove pages you merged.

A Starting Tier Model

These intervals are a starting point to adjust, not a benchmark. The right interval for a page is roughly how often its facts changed over the last year.

Keep Your Dates Honest

Three dates describe a page to machines, and they should never disagree:

• The visible date. Google's byline date guidance recommends a prominent, clearly labeled user-visible date (“Published”, “Last updated”). Render it in the server HTML, not only with client-side script.
• Structured data. datePublished and dateModified on Article (or the relevant CreativeWork type). Google asks that they match the visible dates.
• Sitemap lastmod. Google's sitemap documentation says it uses lastmod only if it is consistently and verifiably accurate, and that it should reflect the last significant update: main content, structured data or links. A copyright-year change doesn't count.

The common failure is three teams updating three things: editors change the visible date, the structured data still carries the original date, and the sitemap stamps every URL with the build time. Store one “content modified” value per page, set it only when a real update ships, and derive all three from it.

What a Real Refresh Changes

• Facts corrected against their source (pricing page, docs, the regulator's page).
• The main answer rewritten if the best answer changed, kept in the first lines of its section.
• A new section for a question buyers now ask, found in support tickets, sales calls or the follow-ups AI assistants suggest.
• Examples, screenshots and product images that match the current product.
• Dead links fixed and links added to newer related pages.
• Sections that no longer apply removed. A shorter, correct page beats a longer one with an outdated half.

If the review finds nothing wrong, log “checked, no change” and leave the date alone. The log proves the page was reviewed; the date tells readers when it last changed.

Diagnose Before You Refresh

When a page starts losing AI citations or referrals, staleness is only one possible cause. A few checks separate it from the others:

• Losses across all the page's questions point to the page itself (staleness, a technical problem, a crawler blocked in robots.txt). A loss on one question usually means a competitor published a better answer to it.
• The answer cites a newer page that says the same thing: currency is the likely gap. Update and make the current state explicit.
• The answer misstates your facts: find where the wrong fact comes from. Sometimes it is your own old page, sometimes a third-party page; our guide to brand entity and wrong AI facts covers fixing it at the source.

Track the effect of refreshes with the same prompt panel and AI referral reports you use for everything else, described in our AI search attribution guide. Compare refreshed pages with similar pages you didn't touch before crediting the refresh.

Images and Video Go Stale Too

A refreshed page with last year's packaging in the hero image still describes the wrong product. Put product images, screenshots and demo clips on the same checklist as the text, and replace them when the product changes, not on a fixed timer. Update the alt text and any ImageObject or VideoObject markup with them. More on visual answers in our video and image answers guide. If you publish an llms.txt file, review its summary and links in the same pass.

Common Mistakes

• Date bumps without changes. Misleading to readers and listed by Google as a warning sign.
• One cadence for every page. Stable pages soak up review time while volatile ones go stale.
• Schedules with no triggers. A price change can't wait for the quarterly review.
• A frozen tier list. Re-tier quarterly based on what actually changed.
• Refreshing instead of merging. Three overlapping pages are three times the upkeep and compete with each other.