How Often Should You Refresh Old Content to Improve AI Visibility?

How often should different types of content be refreshed?

A fixed interval is a backstop. The real schedule comes from triggers: a statistic has a newer edition, a product changed, or the questions people ask have shifted.

How quickly do GEO citations decay, and how often should you refresh them?

There is no universal decay rate, but the evidence points one way. An Ahrefs analysis of about 17 million citations, published in 2025, found that URLs cited by AI assistants were on average 25.7% "fresher" than those ranking in Google's organic results: 1,064 days old versus 1,432. ChatGPT cited the newest pages and Google's AI Overviews the oldest. You can read the Ahrefs freshness study for its method.

Seer Interactive looked at citations in ChatGPT, Gemini and Perplexity from March to June 2026. It found that 75% of cited pages had been updated within the past year and 88% within two years; older pages were rarely cited at all. Its 2026 content recency study also compares its earlier log-file findings.

Both studies show correlation, not a rule. Pages that are well maintained tend to be better in other ways too, and topic mix affects the result. Measure decay on your own pages instead of assuming it:

1. Track which of your pages are cited for a fixed question set each month.
2. Note the date each page was last substantially updated.
3. When a page's citation rate falls for two or three consecutive checks, look at what is being cited instead. Often it is a newer page with more current facts.

For most businesses, that leads to quarterly checks on the pages that matter commercially and an annual review of everything else. How long a refreshed page takes to show up in answers is covered in how long it takes for rewritten content to appear in AI answers.

Why do AI answers favour recently updated pages?

• Time-sensitive questions: many questions imply "now", such as current prices, tools or rules. Engines try to answer with current information.
• Explicit dates in queries: questions including a year, such as "best CRM for small business 2026", favour pages about that year.
• Competition: when a newer page covers the same ground with current figures, it offers the engine a safer source.
• Visible dates: engines can see published and updated dates, and an answer quoting a three-year-old statistic is a weak answer.

How do you refresh old content with current statistics for AI search?

1. List every figure on the page with its source and year.
2. Go back to the original source, not a summary, and find the most recent edition.
3. Replace the figure and put the year and source in the same sentence: "In 2026, X% of Y, according to Z."
4. Remove figures that have no current version, or label them clearly as historical if they are still useful.
5. Check that the conclusions still hold with the new numbers. Sometimes the story has changed.
6. Update examples, screenshots and product names at the same time.
7. Change the visible updated date and dateModified only after these substantive changes.

What counts as a real refresh?

A real refresh changes what a reader learns: new facts, corrected steps, a clearer answer, a missing question answered or outdated advice removed. Changing only the date does not count. Google's guidance on helpful content asks whether pages have had their dates changed to seem fresh when the content has not substantially changed, and treats that as a warning sign. It also misleads readers. Minor edits, such as typo fixes, do not need a new updated date.

How can a content team decide whether to update, merge, or retire an old page for AEO?

Assess each page on four questions: is the topic still relevant, does it earn traffic, links or citations, is it accurate, and does another page cover the same question better? Then use this table:

What old content should you keep, delete, or rewrite for AI search?

Keep pages that answer a question people still ask, even if old, and update them. Rewrite pages on important questions whose current version is weak. Delete, or merge and redirect, pages that answer no current question, duplicate better pages, or give advice you no longer stand behind. Wrong or outdated pages are the ones that do active harm, because engines can repeat them. Fixing information AI is already repeating is covered in how to correct false or outdated information that AI repeats from its sources.

Does pruning thin or outdated content improve LLM visibility?

It can help indirectly, but deleting pages is not a tactic in itself. Google has said publicly that removing content simply because it is old is not something its systems reward. The benefit of pruning comes from removing pages that are unhelpful, duplicated or wrong. That improves how search engines judge the site as a whole, concentrates links and signals on the pages that remain, and stops engines finding outdated answers on your domain. For AI engines that rely on search indexes, those improvements carry over.

Pruning cannot remove what models learned in training, and it does not by itself make any remaining page more citable. The remaining pages still need to be the best answer.

Can a content library recover AI visibility after removing 20,000 thin pages?

It can, but recovery depends on what remains and how the removal is done. A large cleanup should:

1. Redirect pages with links or traffic to the closest relevant page, not all to the home page.
2. Return 404 or 410 for pages with no equivalent, and remove them from sitemaps.
3. Keep and improve the pages that answer real questions, before or during the removal.
4. Watch crawl data: AI crawlers already spend a large share of fetches on missing URLs, so a clean redirect map matters.
5. Track a fixed question set monthly for the topics the site should still own.

Expect search engines to take weeks to months to recrawl and reassess a site of that size. Search-based AI visibility tends to follow that recovery. If the remaining content is thin as well, removing pages will not produce a recovery.

How do you set up a refresh routine?

Keep a content inventory with one row per page: topic, owner, last substantial update, key facts and their sources, and current citation status. Each quarter, sort by commercial importance and falling citations, and refresh the top of the list. Once a year, review the rest and apply the update, merge or retire table. Assign every important page an owner, so refreshes happen when facts change rather than when someone notices.

What do people ask about refreshing content for AI?

Should I change the published date when I update a page?

Keep the original published date and show an updated date, changing it only for substantive edits.

Is updating old content better than publishing new content?

Often, for questions you already cover. An existing page with links and history usually benefits more from an update than a new page would.

Does adding the current year to titles help?

Only when the content really is current for that year, such as a yearly comparison. Adding a year without updating the content misleads readers.

How many pages should we refresh each month?

As many as your team can refresh properly. A few substantial updates to important pages beat many cosmetic ones.