Content Refresh Strategy: A Triage Table and Test Log

Touch a page only when you can name its problem. Sort every page into the first row of the table below that fits it. Pages that fit no row stay untouched, and their dates stay as they are.

To fill the table, export clicks, impressions (how often your page appeared in results), click-through rate (CTR, clicks divided by impressions) and average position for each page from Google Search Console, the free tool that reports how your pages perform in Google results. Pull the last six months and the six months before them.

The position 6 to 15 band and the monthly check for comparison pages are our working rules, not published thresholds. We use them because pages just off the first results page have the most to gain from a better answer, and comparison pages go stale fastest.

A real refresh makes at least two substantive changes, one of them a data fix, a new section, or a corrected claim. Change the visible date and the structured date only after that. Google's own guidance warns against changing dates when the content has not substantially changed.

Our test, which you can run on every edit:

1. New or replaced data, with a named source.
2. A new section answering a question the top results or AI answers now cover.
3. A wrong or outdated claim corrected.
4. An original element: a worked example, screenshot or calculation.
5. Internal links fixed, in and out.

Pass: at least two items are true, and at least one of them is item 1, 2 or 3. Fail: the date stays where it was.

Google's guidance supports this. Its helpful content guidance asks: "Are you changing the date of pages to make them seem fresh when the content has not substantially changed?" John Mueller of Google put it plainly: "Is there significantly new content? Then update the dates. Is there no significantly new content? Then don't update the dates." He added: "Just tweaking, and saying 'oh, still valid in 2023' is not a significant update." His original post is on X.

Two date fields matter. datePublished and dateModified are labels in a page's structured data (machine-readable markup) that tell search engines when it was published and last changed. Google's publication dates documentation says the user-visible date and the structured date should match, and that its systems "look at several factors" to estimate when a page was published or significantly updated.

Run this check on your own library: open a page's source, search for dateModified, and compare it with the date a reader sees. Pass: both match, and the modified date is later than the published date only on pages with a logged substantive change. Fail: they differ, or an edited page shows identical dates.

We ran the same check on 10 October 2026. We took nine pages that surfaced for "content refresh strategy": seven from a web search and two from the sources ChatGPT cited when asked about the topic. We opened each page's source and searched for datePublished and dateModified. PageLens.ai had no page among the nine, and one page blocked our request, so eight were readable. Three declared a modified date later than the published date, one declared identical dates, two declared only a published date, and two had no date fields in the markup. In this small sample, only three of eight readable pages declared a modified date later than the published date.

Updating helps AI citation only when the update makes the page more relevant. Freshness is a modest, uneven factor. Ahrefs' July 2025 study of 16.975 million cited URLs found AI assistants cite fresher pages than Google's organic results, yet the cited pages are still years old on average.

Source: Ahrefs, July 28, 2025. Ahrefs measures the two columns from different data sources, so compare rows within a column, not across columns. Google AI Overviews are the AI-written summaries at the top of some Google results.

Against the table's organic figure, the 1,064-day average for AI assistants is 24.9% younger by publication age (352 days of 1,416), by our calculation from Ahrefs' figures. Ahrefs' own summary uses 1,432 days for organic and reports 25.7%; its table lists 1,416. Ahrefs also states that freshness is one factor among many. In its data, Google AI Overviews and organic results show the weakest preference for fresh pages.

A second Ahrefs study analyzed 1.4 million ChatGPT prompts from February 2025 and was published in April 2026. The median cited page was about 500 days old. Within a single prompt's retrieved pages (the candidates ChatGPT pulls in before choosing what to cite), the cited pages tended to be older and the freshest pages tended to be discarded. Ahrefs concludes that relevance to the sub-questions ChatGPT generates for a prompt (it calls them fan-out queries) matters more than recency: a new page that matches them well gets cited, and one that matches them poorly is retrieved but ignored. News is the exception, where recency decided between pages of similar relevance. For the mechanics of earning citations beyond freshness, see our guide on how to get AI search citations.

Google applies freshness selectively too. Its ranking systems guide describes "query deserves freshness" systems "designed to show fresher content for queries where it would be expected", such as recent events. For a question whose answer does not change, a refresh buys little. We recommend spending refresh effort where facts move: prices, product specifications, regulations, comparisons.

Delete a page only when it has no value, not because it is old. Google's guidance does not encourage removing content for age alone, as Search Engine Land reported after a large publisher deleted thousands of old articles. Google's helpful content guidance also says adding or removing large amounts of content to seem "fresh" will not help rankings.

Use the first two rows of the triage table. Delete or redirect only when a page earns nothing and can't be made useful. Merge when two pages compete for the same query, because one stronger page beats two weaker ones. Every other old page goes to the correct, rework or leave rows.

Set the review window before you edit, based on the type of change. Our defaults: four weeks for title and description changes, eight weeks for content rewrites, twelve weeks for merges and redirects. Merges take longest because search engines must recrawl (visit again) and consolidate the pages. These are working defaults, not published figures.

Write the log entry before the edit:

Keep a few comparable pages untouched as controls. If their clicks rise or fall by a similar amount over the same period, the cause is seasonality or a Google update, not your edit. Compare equal-length periods, and don't judge a page before its review date.

Run the same prompts before and after the edit, and compare the share of runs that cite your URL. Pick three to five questions the page should answer. Our recommendation: run each prompt 5 times per engine before the edit and 5 times at the review date, then count the share of runs in which your URL is cited. Answers vary from run to run, so a single answer proves nothing.

Pass: the share rises by at least 20 percentage points while the control pages' share, measured the same way, moves by less than 10 points. Fail: the share rises less, or the controls rise too. In that case, something other than your edit changed the answers.

We run this check for you at PageLens.ai. You choose the prompts, and we run them on a schedule, weekly on our Launch plan and daily on Growth and Enterprise. Citation reporting shows which pages each engine cites, so your before-and-after is a record, not a memory. Our free AI Visibility Audit takes about 30 seconds with no signup. It shows which pages ChatGPT, Claude, Gemini and Perplexity cite for your domain and which competitors win those citations. For the full method, see our guides on tracking AI citations over time and running an AI visibility audit.

As of October 2026, our plans cost $299 per month (Launch: 100 tracked prompts, weekly, covering ChatGPT, Google AI Mode, Google's conversational AI search mode, and Perplexity), $699 per month (Growth: 100 tracked prompts, daily, adding Gemini and Grok) and $1,499 per month (Enterprise: 200 tracked prompts, daily). Every plan includes automatic content refresh for the pages we publish on your own domain.

Book a demo