Does Updating Old Content Help AI Citations?

Fresh content correlates with AI citations, but the evidence that refreshing a page causes more of them is weak. Pages updated within a year are cited more often, assistants lean toward newer information when relevance is equal, and yet the pages they cite are often old. New pages are cited less in their first month. Changing only a date has not been shown to help. Update a page because the information changed, and measure it against pages you left alone.

Key Takeaways

• A Seer Interactive study (July 2026) found 75 percent of cited pages had been updated within a year. That is a correlation.
• Among pages ChatGPT retrieved for one prompt, the median cited page was about 500 days old, and the newest were most often left uncited.
• In lab tests, a stated recent date and a stated price changed which source a model cited first, and an old date was penalised.
• No refresh cadence is supported by data. Claims like "refresh every 8 to 14 weeks" have no dataset behind them.
• Changing a visible date on its own has no demonstrated effect, and the tests that exist have no valid control.

What the Correlations Say

Updated pages are cited more. Seer Interactive reported on 24 July 2026 that 75 percent of cited pages had been updated within the last year. It does not say that updating caused citation, and pages that are maintained are usually better pages.

But cited pages are not young. AirOps's April 2026 analysis found citation rates of 32.8 percent for pages 30 to 89 days old, 32.0 percent for one to two years and 27.5 percent for two to five years. That is flat from one month to two years, with a dip for brand-new pages that how long does AEO take puts a number on. In another dataset, from 2025, the URLs cited by assistants averaged 1,064 days since publication against 1,432 for Google's organic results, and Google AI Overviews' cited URLs were about as old as organic results.

News is the exception. Muck Rack's December 2025 analysis found about 4 percent of citations came from the last week and 8 percent from the last month, with half from the last 11 months. For news-like questions, recency matters. For evergreen buyer questions, much less.

What Lab Studies Add

A 252,000-trial lab study (Vishwakarma et al., arXiv 2605.25517, May 2026) put two anonymised product reviews in the context of six models, differing in one factor, and asked which was cited first. Four factors decided it across all six models: on-topic against off-topic, a stated price against none, a recent date against an old one, and first against second place. A missing date was not penalised. An old date was.

This is a lab result, with two documents and first citation only. It says that when relevance is equal, models favour a recent date, and it does not say how a page gets retrieved in the first place.

Does Changing the Date Work?

There is no good test. Otterly (September 2026) changed only the date on ten pages and saw citations rise 31.5 percent, while two unmatched controls fell 10.6 percent, but the window included a ChatGPT change in August, and the controls were not matched. In another Otterly test, adding "2026" to eleven titles saw edited groups rise 56 to 61 percent while an untouched page rose 63 to 64 percent, so the lift was not the edit. One page in a guest post on Sitebulb's site fell from an average Google position near 20 to near 79 after the year was added to its title, headline and subheadings.

The sensible conclusion is that a bare date change has not been shown to help, and pages untouched by anyone moved by more than 60 percent in 15 days in the title test, which is how much noise surrounds any before-and-after figure.

What a Refresh Should Be

Update what is out of date.

1. Prices and facts that change. A stated price and a recent date both mattered in the lab. Make sure the page states current numbers.
2. Claims assistants repeat. If an assistant quotes a wrong figure or a discontinued feature, correct it where the figure comes from. See AI brand misrepresentation.
3. Sections that answer buyer questions. Add the questions people now ask. Treat it as a content improvement, not a freshness trick.
4. A visible "last updated" date that matches a real change. Do not backdate or touch dates to look current.

Do not republish near-identical pages to look new. Google's spam policies cover scaled content created to manipulate Search, including its AI responses.

How to Test It

1. Pick a set of pages with similar traffic and similar citation history.
2. Refresh half, leave half alone.
3. Fix the panel of questions that these pages answer, and run it at least weekly.
4. Wait at least a month, since new and edited pages are cited less at first. How long does AEO take covers the timing.
5. Compare the changes in the two halves, not the refreshed half against its past.

The Short Version

Keep pages accurate and current because buyers and assistants both need correct facts. Do not expect a refresh calendar to produce citations, and do not trust a vendor who quotes a multiple for refreshing. If you want to know whether a refresh worked, measure it against pages you did not touch.

Lectern tracks citations to your pages and to competitors' pages across assistants, so a refresh can be read against a control group. The free Hobby plan checks 10 buyer questions weekly; see the pricing page.