Content Refresh: A Practical Process for Keeping Old Posts Ranking and Cited in AI Answers

A content refresh is a material update to a page you already published. New facts, a sharper answer, repaired links, and an honest date all count. The URL and the search intent stay the same. The goal is to win back clicks the page used to earn while keeping it worth citing in AI answers.

The archive is where the decay hides. Search Console shows it page by page once you run the right comparison.

This guide covers the decision first, then the work. It sticks to what Google documents. Everything else is labeled as research or as this guide's own recommendation.

Key Takeaways

• A refresh keeps the URL and the intent: if the intent has moved, you are rewriting, and if two pages answer one question, you are merging.
• Recency bias is measurable in the lab: a 2025 study found that seven LLM rerankers promoted passages carrying newer dates, shifting the top 10's mean publication year forward by up to 4.78 years.
• The practical point: most old posts should be left alone. Google advises against radical changes to a page that already performs well. The audit's main job is to say no.
• A date is a claim, not a lever: Google lists changing dates without substantial change as a warning sign, and it uses sitemap <lastmod> only when that value is consistently and verifiably accurate.
• Cadence follows triggers: a new edition of a cited report, a product change, or a falling clicks line should start a review. Google publishes no refresh interval, so a fixed one is a habit.

What Is a Content Refresh (and What It Is Not)

A content refresh is maintenance on a page that still has the right job. People still search the question it answers, and you still want that URL ranked. What changed is the world around it: a newer statistic, a retired interface, or a rival's clearer answer.

The trouble is that four other jobs get filed under the same word. Each one has a different owner, a different cost, and a different risk to the URL.

Refresh, Rewrite, Republish, Merge, or Delete

• Refresh: same URL, same intent, updated substance. Facts, examples, links, and structure change. The page keeps its history.
• Rewrite: same URL, new intent. The results page now shows a different kind of page. You rebuild the piece around what searchers want today.
• Republish: the same content moved to a new date with little or no change. Google names this pattern as a warning sign, covered in the dates section below.
• Merge: two or more pages answer one question. Keep the stronger URL, fold in what the weaker one adds, and redirect it.
• Delete: the page has no audience, no links, and no topic you still cover. Remove it, or redirect it to the closest page that still serves the reader.

A content refresh is the default for a page that still ranks for its query and has slipped. The other four are exceptions you should be able to justify in one sentence.

Why Freshness Matters More in AI Search

Content freshness is a working assumption in AI search, and part of it now has evidence. A results page shows ten links and lets the reader judge their age. An AI answer picks a few sources and builds an answer from them, so its preferences get amplified.

OpenAI's guide to researching with ChatGPT says its search "Prioritizes the latest available information." No engine publishes how much weight a date carries. What exists beyond that is lab evidence that models react to dates, plus one large observational study of what AI assistants cite.

What the Lab Shows

Fang, Tao, Chen, Chang, and Sakai tested recency bias in LLM-based reranking in a study posted to arXiv on 14 September 2025 and revised on 21 September 2026. They added artificial publication dates to passages from the TREC Deep Learning 2021 and 2022 collections. Across seven models, including GPT-4o and LLaMA-3 70B, "fresh" passages were consistently promoted.

The effect was large: dates shifted the top 10's mean publication year forward by up to 4.78 years. Single passages moved by as many as 95 ranks. In two-passage tests, the preference between equally relevant passages reversed by up to 25% on average.

"Although larger models attenuate the effect, none eliminate it," the authors write. Read it for what it is: a controlled experiment on research collections. It shows a date can sway a model's ranking, and it did not test how live engines weight dates.

What Live Citations Show

Ahrefs analyzed 17 million citations from its Brand Radar data in a study published 28 July 2025. Pages cited by ChatGPT, Perplexity, Gemini, and Copilot averaged 1,064 days since publication. Google's organic results averaged 1,432, so AI-cited pages ran 25.7% "fresher."

This finding is correlational, and Ahrefs reports caveats against itself. Google's AI Overviews cited slightly older pages than its organic results. The average AI-cited URL was also close to three years old, so long-lived pages still get cited.

Ahrefs' own conclusion is guarded: "there may be some benefit to regularly updating content." Perplexity publishes nothing about what a fresh page is worth. Treat freshness as a reason to keep facts current, since touching a page raises no documented score.

How to Spot Content Decay

Content decay is the slow loss of clicks, positions, or citations on a page that once performed. New posts mask old ones in a site-wide chart. You find it page by page.

Google's own guide to debugging drops in Search traffic gives you the method. Most of it runs inside the Search Console Performance report.

Run the Comparison Google Describes

1. Set the date range to the last 16 months. Google recommends it so a seasonal dip does not read as decay.
2. Use the Compare tab. Google's options include "Compare last 3 months to previous period" and the same window year over year.
3. Open the Pages table and sort by Clicks Difference. That orders pages by traffic lost. It also shows whether the drop sits on one page, a group, or the whole site.
4. Check the top queries that lost clicks in Google Trends, one at a time. Google suggests this to tell a drop on your site from a drop in demand across the web.

Read the Pattern Before You Act

Google's guide separates the shapes worth telling apart. Impressions holding while clicks fall points at the title and snippet. In Google's words, "you might not be generating the best page title and snippet that you could."

A small drop in position, such as 2 to 4, can recover on its own. Google recommends "avoiding making radical changes if your page is already performing well." A large drop across many terms calls for a self-assessment of the whole site.

Decay Signals Search Console Cannot Show

• Outdated facts: a statistic with a newer edition, a retired feature, a price that changed.
• Lost citations: the page stopped appearing among the sources for prompts you track. Only a dated prompt log shows this.
• Broken or redirected outbound links: each one is a small trust cost and an easy fix.
• Engagement changes: compare organic entrances and engagement for the URL across the same windows in your analytics platform. Use them as supporting evidence, not as a substitute for Search Console clicks and impressions.

Content Audit Template: Refresh, Merge or Leave Alone

A content audit template is a list of your own pages with one decision per row. An AI visibility audit records what engines say about your brand across buyer prompts. This one records what you will do to each URL you own.

Download the content audit template, or copy the columns below into a sheet. The six columns are the minimum; add one only when you can name the decision it changes. The example rows are hypothetical.

Decision Rules for Each Row

These rules are this guide's own working defaults, and no published standard exists. Write your versions into the template's Decision rules tab. Two people should then fill a row the same way.

The merge rule is the same one behind topical authority.

Keep the sheet shared and hold one row per URL. Teams skip the Owner column. It decides whether anything happens.

How to Update Old Blog Posts, Step by Step

When you update old blog posts, work in the order a reader meets the page. Intent decides everything after it. Check it first and fix links last.

A Worked Example From This Blog

Take our complete guide to ranking on AI search, as read on 15 October 2026. It shows a date of November 22, 2025 and carries "for 2025" in its title. Its four H2s each hold one short paragraph, with no cited sources, Key Takeaways, or FAQ.

Run it through the template and the year stamp alone makes it a candidate. Newer posts on this blog cover parts of the same question, so the template could return "merge" instead. If the URL now redirects, you are looking at that option in practice.

The Seven Steps

1. Re-check the intent. Search the main query and read the current results. If they show a different kind of page, stop and rewrite instead.
2. Replace dated statistics. Find each number's primary source, check for a newer edition, and update both the figure and the date it was checked.
3. Add an answer-first summary. The first two sentences under the H1 should answer the query on their own.
4. Rework headings toward the questions people ask. Chen and Koudas studied 158,099 AI answers and 468,037 citations in an October 2026 arXiv paper. Within one website, pages whose title and paragraphs match the questions people ask were cited more, while FAQ blocks and summary boxes "appear effective only across websites." Page properties still explained little within a site: what ChatGPT's own search retrieved accounted for 32% of the variation, against 2.8% for page properties. This finding is observational, so it shows a pattern without proving a cause.
5. Fix links. Repair broken outbound links, swap secondary sources for primary ones, and add internal links to newer posts that go deeper.
6. Add or prune an FAQ. Keep the questions readers ask after finishing the body and cut any that restate it. Given the finding above, write it for readers first.
7. Update the dates and the markup last, and only if the substance changed. The next section covers what that means.

Last Updated Dates: What to Show and What Not to Fake

Most last updated date SEO advice reduces to one rule: the date has to describe the page. Google estimates a byline date from several signals. Its guidance on byline dates explains: "Google doesn't depend on a single date factor because all factors can be prone to issues."

That gives you three places to state a date and one reason to keep them aligned. A mismatch makes every one of them less useful.

The Three Dates and What Google Says About Each

The same sitemap page says Google ignores <priority> and <changefreq>. Implementation details for the markup itself live in schema markup for AI search.

What Not to Fake

Google asks the question in its own self-assessment. "Are you changing the date of pages to make them seem fresh when the content has not substantially changed?" A yes is on its list of warning signs.

The lab result above explains the temptation, since a date can sway a model's ranking. That is also the reason not to fake one: a misdated page tells readers and machines something false. Inflated <lastmod> values also fail the accuracy test Google sets for using them.

Refresh Cadence by Content Type

Google publishes no refresh interval, and neither does any AI engine. Any number of months you read elsewhere is someone's habit. Plan around the events that make a page wrong.

The table below names the trigger for each content type. The last column is this guide's own suggested check and carries no research behind it. Change it once your own data says otherwise.

Teams running AEO in-house can fold the refresh review into the iteration phase of a 90-day AEO operating plan. The audit sheet becomes that phase's input, and the owner column becomes its to-do list.

How to Tell Whether a Refresh Worked

Decide the test before you ship the change. Write down the page, the date it went live, the queries that matter, and the prompts you will rerun. Without that note, any later number tells a story you chose afterward.

Before-and-After Windows

Compare equal windows on either side of the change date in the Performance report, filtered to the refreshed URL. Give it time. Google says "some changes can take effect in a few days, while others could take several months."

Its advice is that "you'll likely want to wait a few weeks to analyze your site in Search Console again." Check seasonality with the year-over-year view. Note any core update inside the window, too, since Google lists them on its ranking updates page.

Citation Checks Across Engines

Search Console says nothing about AI answers outside Google. For those, rerun a frozen prompt set on the same engines, logged out, before and after the refresh. Log the runs the way AI citation tracking sets out: one row per cited domain, per prompt, per run.

Report the result in the LLM visibility metrics that matter, as a range across runs. One new citation after one refresh is an anecdote. A citation that holds across repeated runs on the page's prompts is worth reporting.

What to Expect, and When

No engine publishes how long a refreshed page takes to show up in its answers. The Google side is easier to read first, because you can request a recrawl and watch the URL in Search Console. Expect the AI side to be uneven, and judge it on the same prompts over at least two runs.

A refresh fixes the page you own. The answers buyers read draw on more than your page, and the engines' own source lists show which other domains shape them. That is the half of the measurement Search Console cannot give you.

Klarivo Monitor tracks five answer engines: ChatGPT, Claude, Perplexity, Gemini, and Grok, each tracked independently. It reports query-level mention rates and top citation sources, the domains the engines drew on, updated on a schedule you control. Because it works at domain level, it shows whether your domain joins a tracked query's sources after a refresh, without naming the page.

Sometimes the list keeps naming review sites, communities, and publications that never mention you. That fix sits outside your site, and it is the work of Klarivo's managed AEO and Reddit service: building the Reddit, review, and publication footprint models retrieve when buyers ask who to buy. Start with Klarivo Monitor to see the domains engines draw on for your category, or book a Klarivo discovery call.