Getting Cited by AI Is Easy. Staying Cited Is the Hard Part.

Most businesses treat AI citations as achievements to unlock. You get cited by Perplexity for your service category, note it down, and move on. The problem is that most citations are not permanent. They have an expiration date. And most AI visibility strategies are not built around that reality.

The data on this comes from a 2026 primary study with a sample size large enough to make the finding hard to dismiss.

The half-life finding

In our June 2026 investigation of content-level citation signals (session 43, 2026-06-10), we documented a joint study by Scrunch and Stacker that tracked 3.5 million citation events across ChatGPT, Perplexity, and Google AI Overviews. The study covered 120,000+ domains across eight industries from September 2025 through March 2026. The research question: once a page entered the AI citation rotation, how long did it stay there?

Median half-life: 4.5 weeks.

Half of all pages that earn an AI citation drop out of the rotation within 31 days. Not because the page disappeared. Not because the information became inaccurate. The content is still there, still indexed, still technically accessible. AI engines just stopped using it.

The platform-by-platform breakdown tells a more specific story:

**ChatGPT:** 3.4-week half-life, the fastest churn of any platform measured. ChatGPT had 76.4% of its most-cited pages updated within the last 30 days -- meaning the active citation set is almost entirely drawn from recently-refreshed content, not newly published content.

**Perplexity:** 5.7 weeks, the most durable. Perplexity's RAG-based retrieval pulls from live web sources, so pages that maintain external engagement -- directory listings with recent reviews, content that other sites link to -- stay in rotation longer than static pages.

**Google AI Overviews and Gemini:** 4.3-4.8 weeks, mid-range.

The one-time citation problem

A second study from the same June 2026 session adds a finding the Scrunch/Stacker data doesn't capture. Writesonic analyzed 23 million cited sources across the same four platforms between April and June 2026. Their finding: 44% of all cited pages appeared in AI answers exactly once before disappearing from the citation set permanently.

Nearly half of all AI citations are one-time events.

That's not a comment on content quality. Those pages may have been excellent. But they were picked up once -- maybe because they published timely content, maybe because a directory listing was refreshed, maybe because they were swept into a training window -- and then never appeared again.

This matters for how you interpret your own visibility data. If you run a point-in-time AI citation check and find you're not appearing in a query result, that doesn't mean you've never been cited. You may have been cited in the last rotation window and churned before the check ran. The snapshot tells you where you stand today. It doesn't tell you about last week.

The 4.5-week half-life and the 44% one-time rate are consistent when you break them down correctly: 44% of pages exit after a single citation event, and the remaining 56% stay in rotation for an average of 11-15 days before churning. The 4.5-week median applies to the durable minority.

Why distributed presence extends the half-life

The same June 2026 investigation (session 43) documented a follow-on Stacker study from April-June 2026 that examined source decay for distributed versus standalone content. Finding: content distributed across a network of publications appeared in AI citations for approximately twice as long as content published only on the originating domain.

The mechanism is not complicated. When a business appears on Yelp, in a local news mention, in a business directory, and on its own website, each source contributes independently to the citation pool. As individual sources churn out of the rotation, others are still active. The pool stays stocked even as individual URLs cycle in and out.

This is why directory presence matters for citation durability, not just citation breadth. A business with a Yelp profile, a Thumbtack listing, a BBB entry, and three local directory appearances has six independent citation sources, each on its own decay curve. When Yelp's page for that business churns out of the ChatGPT rotation, the BBB page may still be active. When Thumbtack's listing cycles out, the local news mention is still in the pool.

A business that appears only on its own website has one citation source. When it churns, nothing else is active.

What ChatGPT's freshness signal means for your content

The Seer Interactive study we documented in our June 2026 investigation tracked 5,000+ URLs from 42 organizations across ChatGPT, Perplexity, and AI Overviews. Finding: 79% of AI-cited URLs were from the last two years, and 65% from the last 12 months.

Combined with ChatGPT's 76.4% "recently updated pages" figure, the implication for content strategy is specific: ChatGPT's citation set is almost entirely drawn from content that was recently updated, not content that was recently published. An established service page that's been live for two years but was updated last month has a stronger position in ChatGPT's citation rotation than a brand-new page published last month.

This reframes the standard "publish more content" advice. For AI citation purposes, refreshing existing pages that already have signal is more efficient than publishing new ones. Service page descriptions, FAQ sections, location pages -- content that AI engines have already identified as relevant to your category. Updating that content resets the half-life clock. Publishing a new blog post starts a new clock with no established track record.

The practical threshold: content updated within 30 days consistently outperforms content that hasn't been touched in 12 months across multiple 2026 citation studies. The delta is not small -- multiple analyses put it at 5-6x more likely to be cited.

The monitoring problem this creates

If your citation set rotates on a 4-week half-life, any snapshot of AI visibility describes one rotation window. An audit from six weeks ago describes a different state from where you are now. The composition of your citation sources will have turned over substantially in that time.

We see this in Sourcepull audit data. Citation rates on the same domain checked at different points in time vary more than most clients expect. That variance is not measurement noise -- it reflects the rotation behavior from the Scrunch/Stacker study. The score is a property of the current rotation window, not a fixed property of the domain.

A one-time audit answers the question "where am I today?" It doesn't answer "am I on a rising or falling trajectory?" For that, you need multiple data points over time, spaced at intervals shorter than the 4.5-week half-life.

The standard approach -- an annual SEO review that includes AI visibility as one line item -- is essentially guaranteed to miss the half-life dynamics entirely. You'd be sampling the citation pool once per 12 rotation cycles.

What to actually do

Three things follow directly from the half-life data.

Build and maintain distributed presence before worrying about your own site content. Each third-party listing is an independent citation source on its own decay curve. Five active directory listings give you five overlapping windows. One brand website gives you one.

Refresh existing pages on a 30-day or shorter cycle. The update doesn't need to be dramatic -- updated service descriptions, current pricing, new FAQ answers, recently-added photos. Any substantive update resets ChatGPT's freshness threshold for that page.

Measure your citation rate over time, not at a single point. The businesses that show up consistently in AI answers aren't doing something permanent -- they're maintaining a portfolio of citation sources that cycle in and out of rotation in a way that keeps the pool continuously stocked.

If you want a current baseline, Sourcepull's Signal Check runs a live four-platform citation check and returns a score in under a minute. That snapshot is a single rotation window -- but it's where the decay curve actually is right now, which is the starting point for everything else.