The Half-Life of AI Citations
Every day for 10 months, we ran the same brand queries through 8 AI models and watched what changed. The same source rarely gets cited twice. Brand mention counts swing dramatically week to week. AI visibility, once earned, doesn't stay earned.
Two questions: when ChatGPT cites a New York Times article today, does it cite the same article tomorrow? When Perplexity mentions Nike five times this week, does it mention Nike five times next week? We ran the same brand queries every day for 10 months across 8 AI models to find out. At the URL level, citations rarely repeat - most sources appear once and don't come back. At the brand level, mention counts swing wildly week to week. AI visibility behaves nothing like SEO rankings - what you earn today isn't what you keep tomorrow.
Ghost Citations
Ghost Citations
We tracked 108,650 citation URLs across 960 brands to answer one question: when an AI model cites a URL, how long does that citation persist?
The answer is stark. 73.5% of citation URLs appear exactly once and don’t return in our observation window. Median lifespan: zero days. And the pool isn’t shrinking - AI models keep citing the same volume, just different URLs each time.
The citations your brand earned yesterday may already be out of rotation today.
The Half-Life
The Decay Curve
URL-level churn is one lens. But brands have many URLs, and some of them stay. Across 857K daily reports covering 10,991 brands over 10 months, we tracked how long brand-level citation counts hold their peak. The median brand drops to half in 31 days.
That is the decay from peak. Even without anchoring to a peak, a typical brand’s week-over-week citation count swings by 51.8%. Whatever baseline a brand earns, the next week rarely looks the same.
The Stability Premium
The Stability Spectrum
Of 1,314 brands tracked for 90+ days, only 173 hold a steady citation presence. The stable few average 74% presence; everyone else cycles through decay and rebound.
Stability improves sharply as brands move from the Small to the Medium tier, then plateaus. A Large-tier brand is no steadier than a Medium one, and the Mega tier sits in the same volatility band as Large. Past a certain scale, more citations stop buying more stability.
Stability is bought in the jump from Small to Medium. Past that, extra scale doesn’t translate into steadier visibility.
What Sticks
Sticky vs Volatile Sources
Not every citation source decays at the same rate. Some domains hold their position for weeks; others appear once and don’t return in our window. The stickiest domain we tracked averaged 32 days between first and last appearance, while the most volatile averaged under a day.
Source type matters. The persistent sources skew toward niche industry tools and specialized guides; the volatile list skews toward high-traffic marketplaces and social platforms whose content rotates quickly.
instagram.comThe Operating Shift
Visibility Is a Running Position
If most citations appear once, the brand half-life is a month, and only 13% of brands hold steady, then visibility is not a milestone to reach. It is a running position to hold.
That shifts the work. A single citation win is evidence, not a finish line. An AI audit done once a quarter sees two half-lives of drift before the next review. And the teams that stay in the stable 13% look less like campaigners and more like operators, running the surface continuously rather than launching at it.
Stop reading AI visibility as a ranking you reach. Read it as a position you run.
A quarterly audit cannot see a 31-day half-life. Run the same queries on a weekly or daily cadence, so losses and recoveries show up inside the window you can still act on.
One-and-done rate (73.5%) plus brand half-life (31 days) is the working pair. URL churn tells you whether individual wins are sticking; brand half-life tells you whether your position is holding.
A 51.8% week-over-week swing is normal noise, not a trend. Treat new citations as provisional and only count a win once it persists for multiple observation cycles.
Methodology
Data Sources
OpenAI
Claude
Gemini
Grok
Deepseek
Meta
Perplexity
AIOHonest Gaps
URLs first seen close to the end of our 177-day window had fewer days to reappear, so their measured persistence can look shorter than it would in a longer study.
Brand half-life is measured from each brand’s highest observed day. Some of the observed decline is regression from an unusually high peak, not an underlying loss of visibility.
Some brands stopped being tracked mid-window, and some teams changed their prompt sets part-way through, so long-term stats exclude a small slice of the data.
We have not yet matched this against a clean benchmark of traditional SEO ranking stability, so comparisons to search rankings are qualitative for now.
Is your position holding, or decaying?
Trakkr runs your brand across 8 AI models every day, so you can tell the difference between a visibility win that sticks and one that vanishes by next week.
Start monitoring freeFree trial included
Continue with this study
See how your brand performs in AI search