Your AI citations expire. Most teams publish like they do not. | Leveret Drive

Your AI citations expire. Most teams publish like they do not.

A page that was getting quoted in March can be invisible by July without anything going wrong. That is not a bug you fix once.

Jon Usborne
· 7 min read
A studio wall of printed work in progress

Here is a pattern we keep running into. A client publishes something good. It gets picked up in AI answers, they see their name showing up when people ask the obvious question in their category, everyone is pleased. Four months later it stops. Nothing broke. Rankings are the same. The page still loads. It just quietly stopped being the source.

Citations decay. Rankings mostly do not.

About half of the content cited in AI answers is less than 13 weeks old (Frase, 2026). Sit with that for a second, because it upends how most content calendars are built.

A good SEO page is an asset. You write it once, it climbs, it holds, it pays you back for years. Plenty of businesses have a page from 2021 that still brings in work. That model is real and it still functions in classic search.

It does not function the same way in AI answers. Recency is doing heavy lifting in what gets pulled, which means the asset model breaks. You are not building a library. You are keeping a signal alive.

Why this hits small teams hardest

Big publishers were already built for cadence. They have people whose whole job is to ship. A ten-person company that publishes when someone has time is now competing on a dimension it never had to think about.

And the instinct when this gets pointed out is to go volume. Spin up a content engine, publish five times a week, flood the zone. That fails for a boring reason: mass-produced content does not earn citations either. Flooding a site with generated articles hurts search performance and does not help AI visibility (LLMrefs, 2026). You end up with more pages and the same invisibility, plus a site that is harder to maintain.

What actually works

Two things, and they are less exciting than a content engine.

1. A cadence you can hold for a year

Two solid pieces a month beats twelve pieces in January and nothing until June. Not because two is a magic number. Because the systems pulling sources are sampling continuously, and a gap is a gap. Pick the rhythm your team can actually sustain in a busy quarter, not the one that sounds impressive in a planning meeting.

2. A refresh schedule for the pages that matter

Most businesses have somewhere between five and fifteen pages that carry the whole operation. The main service pages, the two or three articles that answer the question everybody asks, the pricing explainer. Those get a real review on a calendar, not when someone notices they look old.

A real review means updating the numbers and their dates, adding what changed since you wrote it, cutting the parts that are no longer true, and updating the modified date honestly. Changing a comma and republishing is not a refresh, and treating it like one is the kind of thing that catches up with you.

The uncomfortable math

If half of what gets cited is under 13 weeks old, then a page you wrote a year ago is competing against a much fresher field every time someone asks the question. You do not need to rewrite it. You need it to have been touched, meaningfully, inside a window that keeps it in the running.

For most of the businesses we work with that lands at roughly a quarterly review on the top pages and a steady two pieces a month of new work. That is not a big content operation. It is a small one that does not stop.

The teams winning this are not publishing more than you. They are publishing without gaps.

What to stop doing

Stop treating the blog as a project with an end date. Stop the big quarterly push followed by nothing. Stop measuring the content program on the month it launched.

And stop writing pieces you would not be willing to update. If a topic is not worth revisiting in six months, it is probably not worth publishing now.

What to take from this
  • About half of content cited in AI answers is under 13 weeks old, so citations decay in a way rankings do not (Frase, 2026).
  • A page can hold its ranking and still lose its place in AI answers. Track both separately.
  • Volume is not the answer. Mass-produced content fails in search and in AI answers alike.
  • Pick a cadence you can hold through a busy quarter. Consistency beats bursts.
  • Put your top five to fifteen pages on a quarterly refresh with real updates and honest modified dates.

Questions people ask

How often should we publish to stay visible in AI answers?

A steady rhythm matters more than volume. Two substantial pieces a month sustained across a year outperforms a large burst followed by a gap, because AI systems sample sources continuously and favor recent material.

Does updating an old post help it get cited again?

Yes, when the update is substantive. Refreshing statistics with current dates, adding what has changed, and removing outdated claims can return a page to the recent pool. Cosmetic edits and republishing without real changes do not.

Why do AI answers favor recent content so heavily?

Generative systems are answering questions where being current is part of being correct, so recency acts as a quality proxy. Analysis in 2026 found roughly half of cited content was under 13 weeks old (Frase, 2026).

Should we delete old blog posts that no longer get traffic?

Usually consolidate rather than delete. Merging thin overlapping posts into one strong page that you commit to maintaining gives you a page worth refreshing instead of a back catalogue nobody updates.

Jon Usborne, founder of Leveret Drive
Written by

Jon Usborne, Founder

Jon runs Leveret Drive, where the team builds strategy, brand, websites, video, and demand systems for businesses that want to be the obvious choice.

Keep reading
Follow in Google

Want this kind of thinking pointed at your business?

Tell us what you are working on. We will tell you what we would do first.

Chat with us →