The Half-Life of AI Citations
857K reports. 109K citation URLs. 11K brands. 8 AI models. 10 months of data reveal how quickly AI visibility decays.
Nobody in the industry has published data on how long AI citations actually persist. We have 10 months of longitudinal data across the same queries, the same brands, and the same AI models - enough to answer the question definitively. The findings challenge the assumption that AI visibility, once earned, is durable.
Ghost Citations
Ghost Citations
We tracked 108,650 citation URLs across 960 brands to answer a simple question: when an AI model cites a URL, how long does that citation persist?
The answer is stark. 72.8% of citation URLs appear exactly once and never return. The median citation lifespan is zero days. The mean is pulled to 6.8 days only by the 11.4% that persist beyond 18 days.
Nearly three in four AI citations are ghosts - they surface once, then vanish. The citations your brand earned yesterday may already be gone.
The Decay Curve
The Replacement Cycle
Of the citations that do persist past day one, the drop-off is steep. After 7 days, only 29% of original citations remain. After 14 days, 23%.
Critically, the total number of citations typically grows over time - AI models aren't citing less, they're citing different URLs. The citation pool is constantly churning.
The Half-Life
The 31-Day Half-Life
Zooming out from individual URLs to brand-level: across 857K daily reports spanning 10 months, we tracked how long brands maintain their peak AI citation counts. The answer: 31 days on average before mentions halve from their peak.
Worse: 47.1% of brands that peaked saw their citations halve within 7 days. The decline isn't gradual - it's a cliff.
Brand citations in AI search have a half-life of 31 days. Nearly half of all brands that peak lose half their citations within a single week.
Model Divergence
Eight Models. Eight Different Opinions.
Citation decay isn't uniform across AI platforms. Each model maintains its own citation behavior - and they diverge dramatically over time. Gemini dropped Nike's presence to 14% at the same moment OpenAI held it at 57%. A brand can be invisible on one platform and dominant on another.
| Model | 2025-07 | 2025-09 | 2025-11 | 2026-01 | 2026-03 | Change |
|---|---|---|---|---|---|---|
OpenAI | 100 | 100 | 71 | 57 | 71 | -29pp |
Anthropic | 100 | 100 | 57 | 29 | 57 | -43pp |
Gemini | 100 | 100 | 71 | 14 | 57 | -43pp |
Grok | 100 | 100 | 71 | 29 | 71 | -29pp |
Deepseek | 100 | 100 | 57 | 14 | 57 | -43pp |
Meta | 100 | 94 | 57 | 43 | 57 | -43pp |
Perplexity | - | 0 | 43 | 29 | 57 | +57pp |
Google AI Overviews | - | 0 | 43 | 0 | 14 | +14pp |
Perplexity grew brand presence by +22pp while Meta stayed flat. A brand's AI visibility depends heavily on which model you measure.
The Shape of Decay
Three Brands, Three Patterns
Citation decay takes different shapes. Boeing peaked at 171 monthly citations before dropping 66% in two months. Nike shows high volatility with partial recovery. Gymshark stays remarkably flat - a rare example of stable AI presence.
Decay patterns vary by vertical, brand strength, and how many AI models cite you. The shape matters more than the direction - brands that plateau (like Gymshark) are far better positioned than brands showing continued decline.
The Stability Premium
Only 13.2% Stay Stable
Of 1,314 brands tracked for 90+ days, only 173 maintain "very stable" citation presence. These stable brands average 74% presence - the rich stay stable, and the rest face constant churn.
Counterintuitively, mega brands (50+ citations) are less stable than large brands. The most stable tier is brands with 26-50 citations - big enough to have established presence, not so big that models constantly reshuffle recommendations.
The stability sweet spot is 26-50 citations at 31% volatility. Below that, presence is fragile. Above it, AI models actively reshuffle among well-known names.
What Sticks
Sticky vs Volatile Sources
Not all citation sources decay equally. Some domains maintain their position for weeks while others appear once and vanish. The gap between the stickiest and most volatile domains spans from 43 days to zero.
Citations that appear frequently across multiple queries tend to persist longer. A URL cited 31+ times averages 49 days of persistence vs zero for single-appearance citations. Frequency is the strongest predictor of citation longevity.
Getting cited once is noise. Getting cited repeatedly across queries is signal. Frequency is the best predictor of whether a citation will persist.
Methodology
Data Sources
857,138 daily visibility reports across 10,991 brands, spanning 2025-06-01 to 2026-03-30 (~10 months). Each report tracks citation counts (mentions), presence rates, and per-model breakdowns across 8 AI platforms.
108,650 unique citation URLs tracked across 200 brands (sampled from 960 brands with citation data), spanning a 177-day observation window. Each citation includes first-seen and last-seen timestamps, appearance counts, and source domains.
OpenAI, Anthropic, Gemini, Grok, Deepseek, Meta, Perplexity, Google AI Overviews. Queries are run daily against each model with identical prompts.
Honest Gaps
Pull the answers, facts, and trackers built from this data
Narrow-question answer pages, copyable reference facts, and live benchmark trackers drawn from the same underlying study.
AI visibility decays fast. Are you tracking yours?
Trakkr monitors your brand's AI citations daily across 8 models, so you see decay before your competitors do.
Start trackingSee how your brand performs in AI search