44% of AI Citations Appear Exactly Once. What Makes Others Stick?

The first thing most businesses discover about AI visibility is that they're not cited. The second thing -- which tends to take longer to learn -- is that getting cited once isn't the same as staying cited.

Two studies filed to our research vault in our 2026-10-14 investigation put specific numbers on the scope of the problem. Writesonic's analysis of 23 million citation events found that 44% of cited pages appeared exactly once and never returned across the full tracking window. And Yotpo's 14+ weeks of consecutive citation measurement found that citation sets turn over 70-89% every single week -- with a URL cited by ChatGPT today carrying roughly a 15% chance of still appearing in the same query's answer three weeks from now.

That 15% figure is worth sitting with. Most AI citations aren't wins that accumulate. They're temporary inclusions in a pool that gets replaced almost entirely, every week.

The Weekly Replacement Rate by Platform

Not all platforms rotate at the same speed. Our June 23, 2026 citation drift research (session 56), drawing on SISTRIX's analysis of citation behavior across hundreds of thousands of prompts, documented 4-week source pool retention by platform:

- Perplexity: 44% of the citation pool is the same after 4 weeks - ChatGPT: 31% - Google AI Overviews: 27% - Gemini: 11%

Perplexity holds its source pool most consistently. This comes down to architecture: Perplexity's RAG retrieval follows stable ranking logic -- freshness plus Google organic rank plus domain trust -- which produces relatively consistent retrieval for identical queries. The same sources tend to recur.

Gemini is the opposite. Its citation set draws from Google Maps data, the live Google index, and model updates simultaneously. All three change frequently and independently. A business that appears in Gemini's citation pool this week may be gone next week for reasons entirely disconnected from anything that business did.

This creates an underappreciated prioritization implication. Fix plan work that improves Perplexity citations tends to produce something that persists. Gemini citation wins, absent strong entity infrastructure, may evaporate in the next weekly rotation cycle.

What Content Type Holds Longest

Yotpo's study tracked not just turnover rates but what kind of content produces the most stable citations. The finding: high-intent content -- specifically product pages and policy pages -- produced the most stable citation behavior across all three platforms studied (ChatGPT, Gemini, and Google AI Mode). These page types anchor presence more reliably than editorial content or news-style posts.

The mechanism is consistent with what we see across audits. AI platforms retrieve pages that directly match what they're trying to answer. A service description page that clearly states what a business does, where it operates, and what it offers has a persistent retrieval signal because the underlying query doesn't change. A "top 10 tips" blog post matches different queries in rotation and gets displaced as newer content covers the same territory.

For local service businesses, this means the most durable citation targets are service pages, not content marketing. Not because content marketing is useless -- it contributes to broader entity signals -- but because it doesn't produce the stable, high-intent retrieval anchor that service pages do.

What Actually Drives Recurring Citations

Our June 10, 2026 investigation of new business citation acceleration (session 43) identified the input side of this question: what infrastructure predicts whether a business enters the citation pool as a recurring fixture rather than a one-time inclusion.

Two of the largest citation source studies available -- Muck Rack's analysis of over 1 million AI prompts (May 2026) and 5W PR's citation source study (May 15, 2026) -- both found that 84-85.5% of AI citations reference earned media sources: third-party directories, publications, and review platforms, not brand-owned websites. Distributing content across multiple third-party publications increases AI citations by 325% versus publishing the same content on the brand site alone.

More specifically: domains with profiles on review platforms (Yelp, Healthgrades, Houzz, Trustpilot) show a 3x higher citation rate than domains without them, per SE Ranking's November 2025 study. The key is not review scores. It's entity presence -- the business existing as a recognized, consistent entity on platforms that AI engines already trust and cite.

The pattern this points to is entity depth: the degree to which a business exists as a coherent, verifiable entity across the platforms an AI engine draws from. Businesses with thin entity footprints -- a website, a GBP listing, inconsistent name-address-phone data across directories -- appear in citation pools occasionally, when retrieval happens to select them. Businesses with deep entity presence -- consistent directory coverage, corroborating third-party signals, schema that clearly identifies the entity and links to its established profiles -- appear recurrently.

What This Means for Your Fix Plan

The one-time citation problem is a Phase 1 infrastructure problem, not a Phase 2 content problem. Most published AI visibility advice is Phase 2 advice: schema refinements, content freshness tactics, dateModified updates, YouTube presence. Those tactics produce movement for businesses already in the citation pool with established entity signals. For businesses with thin infrastructure, they do nothing -- there's no foundation for the retrieval to anchor to.

The conclusion from the vault data: NAP consistency, directory coverage, and LocalBusiness schema with sameAs links to established profiles build the baseline that drives recurring citations rather than one-time appearances. Infrastructure depth is what separates the 56% of cited pages that recur from the 44% that don't.

If you're seeing sporadic AI citations -- appearing occasionally across platforms and then disappearing -- the fix is probably not another content push. It's entity infrastructure: verifying which directories list your business, confirming name-address-phone consistency across all of them, and ensuring your schema gives AI engines a clean entity signal to anchor to.

Signal Check (free at sourcepull.ca) shows your current citation state across the main platforms. A full Sourcepull audit shows why specific gaps exist and what infrastructure work closes them.