How We Schedule Content Refreshes Using ChatGPT Half-Life Data

Seventy-eight percent of landing pages ChatGPT cited on day zero were still being cited 30 days later. For news pages the picture is different: almost a quarter were never cited again after day zero. Both numbers come from Promptwatch's ChatGPT content half-life report, published on October 5, 2026. The team took every classified page ChatGPT cited on July 6, 2026 and tracked it for 90 days, to October 4.

We use that report when we build a refresh calendar with a client, because most calendars run on the age of the page. Age is a weak signal. The report suggests that page type matters at least as much, and that gives us a better way to rank who gets attention first.

What the report found

After 30 days, ChatGPT still cited 78% of the landing pages and 76% of the docs that it had cited on day zero. After 60 days the figures were 69% and 67%. Comparison pages held 72% at 30 days and 63% at 60. Docs were cited again later 93% of the time, against 77% for news pages.

Product pages decay faster. 55% were still cited after 60 days, level with reviews and how-to guides, and all three reach their half-life after 74 to 77 days. The report notes that pricing pages are classified as landing or product pages. Listicles kept 60% after 60 days.

News is the short-lived type: half gone after 36 days, 41% still cited at day 60. One more detail matters for reading the chart correctly. Landing pages, docs, comparisons and listicles only cross the 50% line in the final week, because the window ends. The report's own wording is that their half-life is likely close to or beyond 90 days, so we do not quote a number for them.

What it does not tell us

One cohort, one starting day, one engine, one vendor's dataset. A page cited on July 6 is not the same as a page the client published last week, and the report cannot say what would have happened if any of those pages had been edited. It is correlation, not a recipe. A page that remains cited might be good, or might simply have no better competitor for that prompt.

We also keep in mind that "still cited" is a share of pages that had been cited, not a share of all pages. Pages that were never cited do not appear in the data at all.

How it changes the refresh calendar

Here is the planning logic we apply, using the report's suggestions as the frame.

1. Landing, docs and pricing pages are long-term assets. They are the types the report shows holding up longest, so they get scheduled, steady upkeep, not emergency attention. The task is to keep facts current. A stale price or a retired feature on a page that stays cited for months is the risk.
2. Product pages get a closer watch. The report cautions against assuming they stay cited by themselves. Specs, availability and prices are the things that go out of date, so we put a recurring check on those fields, owned by someone at the client.
3. Comparisons and listicles are worth investing in. Both keep more than 60% after two months in the data. For a client with none, that is a case for building one carefully. For a client with several, it is a case for keeping them accurate.
4. News and announcements are bursts. We do not schedule a refresh for a launch post three months later. We link to a durable page instead.

None of that is a promise. It is a way of deciding where a limited number of editing hours should go first.

The checks we run without any tool

Before touching a page, we read it as a stranger would. Is the headline claim still true? Is the date visible? Do figures link to a source? Are there outdated screenshots, dead links, or product names that changed? A refresh that updates the facts and leaves the structure alone is usually enough. We avoid cosmetic rewrites, because we cannot say what exactly makes a page keep being cited, and changing everything makes it impossible to learn from the result.

The measurement step

For the client's own pages, the equivalent view lives in Promptwatch, which is the platform we run client programs on. Citation Trends classifies each citation by content type and source type, and for every page it shows first cited, ramp-up days, peak rate and decay half-life. That is the same shape of analysis as the report, applied to the client's URLs and their tracked prompts.

We read it in three steps. First, which pages are past their peak and decaying? Those are refresh candidates. Second, which important pages have never been cited? Those need a content gap review, not a refresh. Third, which pages are still climbing? We leave those alone.

Page Tracker follows specific URLs, so a page we have just updated can be watched over the following weeks. Content Refresh produces diff-style update suggestions for existing pages, which we review with the client before anything is published. The plan tiers for these features are not spelled out in the sources we work from, so we confirm the client's plan before promising any of them. If the client wants a hands-off approach, the automated Content Agent is a Business plan feature at $579 a month, and it publishes through a review inbox or in a hands-off mode with a quality check.

What we tell clients

A page that was cited once is not guaranteed to be cited again, and a page that fades is not necessarily bad. We use the half-life figures as a prior, then replace them with the client's own curves as soon as there is data. The calendar stays short, the refreshes stay factual, and every edit goes into the log so we can see afterwards what moved and what did not.