Citation Trends and When We Refresh a Page

Most content calendars are built on a hunch. "That article is a year old, let's update it." Sometimes the hunch is right. Often it burns a week on a page that was fine and ignores one that was quietly fading.

Citation data gives us a better trigger. This is how we use it.

A page has a life in AI answers

Pages do not earn citations at a constant rate. Looking at citation history, we see a rough shape. A page is published, gets crawled, starts being cited, ramps up to a peak, and then often declines as fresher or better pages take its place. Where a page sits on that curve decides what we should do with it.

Promptwatch's Citation Trends gives us the numbers that describe the curve: when the page was first cited, how many days it took to ramp up, its peak citation rate, and its decay half-life. It also classifies each citation by content type and source type, so we know whether we are being cited as a guide, a comparison, a documentation page, or something else.

You do not need the maths. You need the question each figure answers.

First cited: how long does a new page take to get noticed on this topic? This sets expectations for new content.
Ramp-up days: is this an engine that rewards new pages quickly or slowly? It tells us how long to wait before judging a launch.
Peak rate: how good did this page get? A high peak that has since fallen is a refresh candidate. A page that never peaked is a rewrite candidate.
Decay half-life: how fast do pages like this lose their edge? Short half-lives mean a topic that needs regular upkeep. Long ones mean evergreen material we can leave.

Four page types and what we do with each

Rising. The page is young and its citation rate is climbing. We do not touch it. Editing a rising page is how you reset the clock. We watch it and make sure crawler access stays clean.

Peaked and decaying. This is the main refresh target. The page proved it can win, and the rate is slipping. Before editing, we look at what replaced it: which competitor pages are being cited now for the same prompts. We use the Page Tracker to watch those URLs and read them. We usually find a fresher date, a clearer first paragraph, or a data point the old page lacks.

Never cited despite being crawled. The page is retrieved by AI crawlers and still loses. Leave refreshing aside and consider a rewrite or a merge. Often several thin pages on one topic compete with each other, and consolidating them helps. We covered the logic of pruning in a separate post.

Cited but declining only on one engine. Engine-specific drops usually mean a change in what that engine prefers, or a competitor winning there. We compare across engines before editing, because a refresh that helps one can be neutral for another.

Refreshing without breaking what works

A refresh is a surgical edit, not a rewrite. Our rules:

Keep the URL. Changing it throws away history and invites redirect problems.
Keep the opening answer unless it is wrong. The first paragraph is often what gets quoted.
Update facts and dates where they have actually changed. Do not touch the date if you did not touch the content. That is a bad habit.
Add what the replacing pages have that yours lacks, but only if it is true and useful to the reader.
Recheck structure: headings, a clear summary, and a table where a table helps.

Promptwatch's Content Refresh produces diff-style suggestions for existing pages, which we find easier to review than a rewritten draft. We accept what is correct and reject what is generic. Nothing gets published without a human reading it.

Timing the check

After a refresh we resist the urge to check the next morning. We wait a couple of crawl cycles and then compare the same prompt panel over the same length of period before and after. The A/B period comparison in the dashboards is built for this. If the page was crawled after the edit and the citation rate is up, good. If it was not crawled, that is an access question and Agent Analytics will show it.

A note on honesty

We cannot promise that a refresh will lift citations. Engines make choices we do not see. What we can promise is a reason for each edit and a way to check the effect. If a refresh does nothing, we note it and learn from it. Some pages are simply outranked by better sources, and the right answer is to publish something different.

Putting it in a monthly rhythm

Once a month we pull the list of pages with decaying citation rates and sort it by how much the page matters to the client. Pages tied to high-intent prompts go first. Informational pages with low conversion value go later or never. The result is a short list of five or six pages, not forty. Brief, edit, publish, recheck.

Our recommendation

Refresh by evidence, not by calendar. Use first-cited, ramp-up, peak, and decay figures to decide which pages are worth the time, and keep your hands off the ones that are climbing. We run this on Promptwatch because Citation Trends, the Page Tracker, Content Refresh, and Agent Analytics all feed one decision. The platform is at promptwatch.com.