AI Citation Half-Life Is 11 Days: How Long AI Citations Last, With 2026 Data by Engine
Profound tracked 883,000 pages across 7 AI engines for 12 months and found that citation share fell to half its peak in a median of 11 days. This page covers how the number varies by engine count and content type, and how to plan refresh cycles around it.
How long does an AI citation last? Much less time than most teams assume. The median time for a page's citation share to fall from its peak to half that level is 11 days (Profound, 2026)[1]. Getting into a ChatGPT or Perplexity answer once does not mean you will stay there for months. This page defines citation half-life, explains how it is measured, breaks it down by engine count and content type, and shows how to turn the number into a refresh plan.
30-second definitions: citation half-life, citation share, rise
- Citation share is the percentage of all citations produced by the tracked AI engines on a given day that point to a specific page (Profound, 2026)[2].
- Peak is the point where citation share, smoothed with a 14-day moving average, reaches its highest value (Profound, 2026)[2].
- Rise is the number of days from a page's first citation to its peak (Profound, 2026)[2].
- Citation half-life is the number of days from the peak until share drops below half the peak and stays there for 14 consecutive days (Profound, 2026)[1].
- Citation decay is the full process of citation share declining after the peak; Profound released a product feature under the same name in September 2026 (Profound, 2026)[2].
AI citation half-life: key numbers at a glance
From September 8, 2025 to September 8, 2026, Profound tracked daily citations for 883,000 pages across 7 engines: ChatGPT, Perplexity, Claude, Google AI Overviews, Google AI Mode, Google Gemini, and Microsoft Copilot (Profound, 2026)[1]. Counted as page and engine pairs, that comes to 1,188,816 observations (Profound, 2026)[1].
| Metric | Value | Basis | Source |
|---|---|---|---|
| Median half-life | 11 days | Peak to half of peak share | (Profound, 2026) |
| Median from first citation to half-life | 42 days | First citation date to half-life point | (Profound, 2026) |
| Share of pages halving within 2 weeks | 78% | Within 14 days of peak | (Profound, 2026) |
| Share of durable pages | 22% | Pages that lasted beyond 2 weeks | (Profound, 2026) |
| Median half-life of durable pages | 27 to 29 days | Varies by engine | (Profound, 2026) |
| Cross-engine duration correlation | 0.03 to 0.09 | One engine compared with another | (Profound, 2026) |
| Correlation among Google's 3 surfaces | 0.12 to 0.17 | AI Overviews, AI Mode, Gemini | (Profound, 2026) |
Don't skim past the 78%. It means most citations are a brief spike, and only a minority last for weeks (Profound, 2026)[1]. Short-lived pages have a median rise of about 20 days, while durable pages take 36 to 46 days (Profound, 2026)[1]. Pages that climb slowly also fall slowly.
By engine count: the more engines cite a page, the longer it lasts
The clearest factor behind a longer half-life is how many engines cite the same page. Pages cited by 5 or more engines are 5.5 times more likely than single-engine pages to hold at least half their peak share after 8 weeks (Profound, 2026)[1].
| Citing engines | Median rise | Holding half of peak after 56 days | Source |
|---|---|---|---|
| 1 engine | 19 days | 1.2% | (Profound, 2026) |
| 2 engines | 31 days | 2.0% | (Profound, 2026) |
| 3 to 4 engines | 33 days | 3.5% | (Profound, 2026) |
| 5 or more engines | 46 days | 6.8% | (Profound, 2026) |
The absolute numbers are small. Even in the 5-plus-engine group, only 6.8% of pages last 8 weeks (Profound, 2026)[1]. The gap from the single-engine group (1.2%) is still clear. Add the 0.03 to 0.09 cross-engine correlation and the picture is plain: staying power on ChatGPT tells you nothing about staying power on Perplexity (Profound, 2026)[1]. That is why each engine needs its own measurement. Our multi-engine measurement guide covers how to set that up.
Engines also start to diverge after about 4 weeks. Among durable pages, a Claude citation was 2.8 times more likely than a ChatGPT citation to still be active after 8 weeks (Profound, 2026)[1]. Perplexity (62%) and Claude (63%) were also more likely to spread their citations over several weeks before the peak than the other 5 engines (18 to 26%) (Profound, 2026)[1]. For context on what sources each engine prefers, read this alongside AI engine citation source patterns.
Half-life by content type: comparisons last longest, news fades first
Profound sorted 133,070 durable pages by content type (Profound, 2026)[1]. The gaps are modest, but they all point the same way.
| Type | Pages included | Median half-life | Holding half of peak after 56 days | Source |
|---|---|---|---|---|
| Decision | Comparisons, lists, reviews | 29 days | 11.2% | (Profound, 2026) |
| Explainer | Explainers, guides, docs, references, tools | 28 days | 10.2% | (Profound, 2026) |
| Commercial | Product, pricing, company pages | 28 days | 9.8% | (Profound, 2026) |
| Reporting | News, announcements, research | 27 days | 7.0% | (Profound, 2026) |
Reporting content, tied to a date or an event, drops first; the other three types hold on at rates 1.4 to 1.6 times higher (Profound, 2026)[1]. Reporting pages also have a shorter median rise, 34 days versus 41 to 43 days for the other types (Profound, 2026)[1]. The common hunch that announcement posts spike fast and fade fast now has numbers behind it.
Reading half-life alongside freshness data
Separate from half-life, there is evidence that AI engines cite newer content more often. Profound reports that half of the content answer engines cite is less than 13 weeks old (Profound, 2026)[2]. Ahrefs analyzed 17 million citations and found that URLs cited by AI assistants averaged 1,064 days old, 25.7% younger than those in traditional search results (1,432 days) (Ahrefs, 2025)[3]. Among engines, ChatGPT tended to cite the newest pages and Google AI Overviews the oldest (Ahrefs, 2025)[3].
Reading both datasets as "just change the date often" gets it wrong. Ahrefs notes that low-quality content updated daily does not improve AI search visibility, and it quotes Google's warning against changing the publish date without real content changes (Ahrefs, 2025)[3]. Half-life gives you a reason to rewrite, not a reason to fake dates. For how AI picks its sources in the first place, see how AI chooses citations.
Caveats when interpreting half-life
- The unit is share. Half-life tracks a page's proportion of all citations, not its citation count. If competing pages multiply, your share falls even when your count stays flat (Profound, 2026)[2].
- It uses a 14-day moving average. Daily values are smoothed over two weeks to reduce noise (Profound, 2026)[2].
- There is a minimum citation threshold. Only pages first cited more than 28 days before the end of the study window, with at least 5 citations in the 14-day peak window, were included (Profound, 2026)[1].
- It is a sample. The dataset is not every cited URL but a deterministic 1-in-128 sample (Profound, 2026)[1].
- The public data covers 7 global engines. The materials include ChatGPT, Perplexity, Claude, Google AI Overviews, Google AI Mode, Google Gemini and Microsoft Copilot (Profound, 2026)[1]. They do not break out Korean-language queries or domestic engines.
That last point sets the scope. On the public materials alone, this is data on citation decay across major global engines. It does not show that Korean prompts, domestic services such as Naver, or local industry queries follow the same pattern. Korean brands need to measure separately with their own prompt sets.
A 5-step refresh plan built around citation half-life
| Step | What to do | Decision criteria |
|---|---|---|
| 1. Baseline measurement | Log your pages' citation share for core prompts, per engine, every day | Peak date and share per engine |
| 2. Half-life calculation | Compute peak, rise, and half-life point for each page | Based on 14-day moving average |
| 3. Refresh queue sorting | Sort by shortest half-life instead of a fixed cycle and work from the top | Pages past their peak first |
| 4. Substantive update | Actually update figures, comparison tables, and sources | Excludes changing only the publish date |
| 5. Content mix adjustment | Produce more of the types with both long rise and long half-life | Share of comparison and explainer pages |
Step 3 comes straight from Profound: skip the fixed cycle, sort your pages by half-life, and work from the top down (Profound, 2026)[2]. Step 5 comes from the same source, which recommends making more of the formats that have both a long rise and a long half-life (Profound, 2026)[2]. The research report is blunter. Citations expire within weeks, so content has to ship on a regular cadence to stay in the candidate pool (Profound, 2026)[1].
Your measurement tool sets the quality of step 1. Profound offers filters for your own domain, all domains, and individual models in its Citation Decay feature (Profound, 2026)[2]. If you also need to track Korean-language prompts and domestic engines, a Korean solution such as BOIDA is worth adding to the shortlist. We cover why tools report different numbers in AI visibility tool measurement discrepancy and compare tools in AI visibility monitoring tools comparison. For KPI design, see GEO KPIs and prompt monitoring.
Frequently asked questions
What is the half-life of an AI search citation? The median is 11 days, based on 12 months of tracking 883,000 pages across 7 engines (Profound, 2026)[1].
Once a page earns an AI citation, does it keep it? Usually not. 78% of pages lost half their share within two weeks of peaking (Profound, 2026)[1].
Which pages keep AI citations the longest? Pages cited by several engines at once, and decision pages such as comparisons and reviews (Profound, 2026)[1].
Does citation duration differ by engine? Medians are similar, but engines split after about 4 weeks. Among durable pages, a Claude citation was 2.8 times more likely than a ChatGPT citation to still be active after 8 weeks (Profound, 2026)[1].
How often should content be refreshed? No published data gives a correct cycle. Refresh pages in order of half-life and avoid changing only the publish date (Profound, 2026)[2]. Ahrefs also points out that date-only updates have no effect (Ahrefs, 2025)[3].
Key takeaways
An AI citation is exposure that lasts only a short time. A median half-life of 11 days, with 78% of pages halving within two weeks, means the "optimize once and move on" approach to GEO does not hold up (Profound, 2026)[1]. The citations that lasted came from pages cited by several engines at once, and from comparison and explainer pages less tied to a date. Measure each engine separately and refresh in half-life order. For GEO fundamentals, see What is GEO; for the wider market, continue with the 2026 global GEO and AEO landscape.
Related companies
- 보이다 (BOIDA)생성형 검색 최적화(GEO) 솔루션, AI 가시성 측정
- AhrefsSEO 툴킷, AI 가시성
- ProfoundAI 가시성 모니터링 플랫폼
Frequently asked questions
- Profound tracked 883,000 pages across 7 engines for 12 months. The median time for a page's citation share to fall from its peak to half that level was 11 days (Profound, 2026).
- Usually not. 78% of cited pages lost half their share within two weeks of peaking. Even the 22% that held on longer had a median half-life of only 27 to 29 days (Profound, 2026).
- Pages cited by several engines at once. Pages cited by 5 or more engines were 5.5 times more likely than single-engine pages to survive past 8 weeks. By content type, comparison and review pages lasted longest and news and announcement pages dropped first (Profound, 2026).
- Median half-life is similar across engines, but the curves split after about 4 weeks. Among long-lasting pages, a Claude citation was 2.8 times more likely than a ChatGPT citation to still be active after 8 weeks (Profound, 2026).
- Public data does not set a fixed cycle. Profound recommends sorting your own pages by half-life and refreshing from the top of that list instead of following a calendar (Profound, 2026). Ahrefs also warns that changing only the update date does nothing (Ahrefs, 2025).
Q.What is the half-life of an AI search citation?
Q.Once a page earns an AI citation, does it keep it?
Q.Which pages keep AI citations the longest?
Q.Does citation duration differ by engine?
Q.How often should content be refreshed?
Sources
Related documents
- What Content Does AI Cite?: How Generative Engines Choose CitationsHow generative engines like ChatGPT and Perplexity pick the sources behind an answer, explained as a three-step process: retrieval, grounding, and synthesis, plus the conditions that make content citable: extractable chunks, semantic density, source credibility, and freshness.
- Which Sources Do AI Engines Cite? Citation Tendencies by EngineA qualitative look at how generative engines like ChatGPT, Perplexity, and Gemini differ in the sources they pick for their answers, why those differences arise mechanically, and how to respond from a multi-engine perspective.
- What Is AI Search Share of Voice: Definition, Measurement Formula, and Brand Visibility GuideAI Search Share of Voice (AI SOV) is the percentage of AI-generated answers from ChatGPT, Perplexity, and Gemini that mention a specific brand. This page covers the formula, how it differs from traditional SOV, per-engine measurement methods, and a tool comparison: all in one place.
- Multi-Engine Measurement: How to Measure Visibility Across ChatGPT, Gemini, Perplexity, and ClaudeWhy every engine answers differently, the trap of single-engine measurement, and a multi-engine GEO methodology for measuring AI visibility through prompt sets, repetition, and share of voice.
- GEO KPI Performance Measurement: Citation Rate, AI SoV, and Prompt MonitoringA step-by-step guide to the core GEO KPIs: citation rate, AI Share of Voice, and prompt monitoring. Covers the arXiv-grounded statistical framework, brand-scale benchmarks, and a side-by-side comparison of domestic and global measurement tools for quantifying AI search visibility.
- Why AI Visibility Tools Report Different Scores for the Same Brand: Causes and a Verification ProcedureWhy the same brand scores differently in every AI visibility tool, split into four layers (engine variance, collection design, metric definition, aggregation), with a side-by-side table of the published formulas from Profound, Peec AI, and Ahrefs, sample-size confidence interval thresholds, and a six-step verification procedure.
- Content Structure That Gets Cited in AI Answers: Writing for ExtractabilityThe writing AI cites is not the same as writing that reads well. How to raise extractability through citable units, answer-first placement, question-answer structure, and tables, lists, and definitions: grounded in GEO research and a practical checklist.
- GEO & AEO Key Statistics 2026, With SourcesA citation magnet that gathers verifiable statistics on the adoption of AI search, citation, and generative search, each with its source URL. It spans everything from the visibility lift reported in the GEO paper to zero-click rates and AI-summary click-through.
- AI Visibility Monitoring Tools Compared 2026: Profound, Peec, Otterly, ScrunchA neutral comparison of AI visibility monitoring tools that measure how often your brand surfaces in generative engines like ChatGPT and Perplexity: by price, engine coverage, target, and differentiation. Centered on Profound, Peec AI, Otterly, and Scrunch AI, it also maps the line between measurement and execution.
- What Is GEO: The Definition of Generative Engine Optimization and How It Differs From SEOGEO (Generative Engine Optimization) is the strategy of getting your content cited in answers produced by generative engines like ChatGPT and Perplexity. Here is the definition, how it differs from SEO, and how it works.
- Global GEO/AEO Player Landscape 2026, Monitoring Tools, Agencies, and PlatformsA 2026 landscape that sorts GEO/AEO players into monitoring tools, specialist solutions and agencies, enterprise platforms, and regional players. We compare the leading vendor in each category, founding, headquarters, tracked engines, pricing, and differentiation, against primary sources.