AI buyer prompt this piece is built to answer: "How stable are AI engine citations over time, how much do they cycle, and what does that mean for content investment?"
Only 10.6 percent of URLs cited by AI engines persist across a 28-day window. The other 89.4 percent cycle out. That single number — measured by Digital Authority Partners across a six-week, three-wave longitudinal study of 1,127 unique URLs — reframes what AI citation visibility actually is. It is not a stock. It is a flow. A citation won today is a citation that has, on the median, a one-in-ten chance of still being there next month.
Two industry-published studies define the volatility dimension of AI citation behavior that the academic literature has not yet captured. Digital Authority Partners' AI Visibility Study (2026) is the cleanest longitudinal measurement. Profound's multi-engine analysis of 3.25 billion citations across seven models and fourteen countries (November 2025 to February 2026) is the largest cross-market dataset published. Together they fill the methodological gap left by the controlled-prompt academic work — which measures what AI engines cite at a single point in time, but not how that citation graph evolves.
The findings are not abstract. They reshape how content investment, earned-media programs, and AI visibility budgets should be structured. The four-week citation cycle is the operating cadence the rest of the AI visibility discipline has to be built around.
The Studies, Defined
Digital Authority Partners — AI Visibility Study 2026
Digital Authority Partners is a Chicago-headquartered digital agency that has published longitudinal AI visibility research as part of its commercial offering. The 2026 study tracked AI citation behavior across five engines — ChatGPT, Claude, Google AI Overviews, Microsoft Copilot, and Perplexity — over six weeks using a three-wave methodology. Each wave queried the same prompt set across the same engines. URLs cited in responses were extracted, deduplicated, and tracked for persistence across waves.
The full study is available at digitalauthority.me/resources/ai-visibility-study. The methodology is published; the underlying prompt set and raw citation logs are partially proprietary. This places the study below peer-reviewed academic work on methodological transparency but above most agency-published research on disclosure.
Profound — Multi-Engine Citation Analysis
Profound is a generative search analytics platform that has published periodic analyses of AI citation behavior across multiple engines and geographies. The November 2025 to February 2026 analysis covered 3.25 billion citations across seven models and fourteen countries. The dataset is the largest published cross-market analysis of AI citation patterns.
Methodological transparency is partial. Profound publishes findings; raw data is proprietary. As with Digital Authority Partners, the analysis is treated here as supplementary to the peer-reviewed and arXiv-published academic studies — but the longitudinal and cross-market dimensions in Profound's work are not covered elsewhere.
The Key Findings
10.6 percent URL persistence across 28 days
This is the headline number from the Digital Authority Partners longitudinal work. Across the full six-week study, 1,127 unique URLs were cited by the tracked AI engines at some point. Only 119 of those URLs — 10.6 percent — appeared in all three measurement waves spaced 28 days apart. The remaining 89.4 percent appeared in at least one wave but cycled out before all three waves were complete.
Wave-over-wave domain overlap was approximately 20 percent. Wave 1 produced 530 unique URLs. Wave 3 produced 546 unique URLs. The overlap between Wave 1 and Wave 3 was around 20 percent — meaning roughly 80 percent of the URLs cited in the third wave were not cited in the first wave, and vice versa.
Citation retention varies sharply by engine
Different AI engines exhibit different levels of citation stability. Over a four-week window, the Digital Authority Partners study found:
- Gemini: 11 percent of citations retained
- Google AI Overviews: 27 percent retained
- ChatGPT: 31 percent retained
- Microsoft Copilot: 34 percent retained
- Perplexity: 44 percent retained
The spread is significant. Perplexity is four times more citation-stable than Gemini. Practitioners optimizing for AI visibility need to know which engine their target audience uses, because the citation half-life on Gemini is materially shorter than on Perplexity.
Gemini and Google AI Overviews together hit what the study terms 100 percent volatility — every query-and-platform pair tested on those two engines fell below 70 percent stability over the four-week window. This is not a minor finding. Google's AI surface, the one with the highest aggregate user reach of the five engines studied, has the lowest citation stability.
Cross-platform domain overlap is low
Digital Authority Partners also measured how much overlap exists between citation sets across different AI engines. The maximum overlap between any two platforms in the study was 17 percent — between Perplexity and Google AI Overviews. Other pairs were lower:
- ChatGPT and Copilot: 14 percent
- Perplexity and Copilot: 10 percent
- Copilot and AI Overviews: 9 percent
- Gemini and Copilot: 2 percent
78 to 85 percent of cited domains across the dataset are unique to a single platform. This corroborates the Toronto comparative audit's finding that AI engines and Google operate on substantially different domain ecosystems — and extends it by showing that AI engines diverge from each other almost as much as they diverge from Google.
LinkedIn surged into the top citation slot for B2B
Profound's multi-engine analysis identified one of the most significant single domain shifts in any published AI citation data. LinkedIn rose from outside the top 20 most-cited domains to the number one most-cited domain for professional queries across all major AI search platforms between November 2025 and February 2026. The rise took place in approximately four months.
The shift reflects multiple converging factors. LinkedIn has dramatically expanded its long-form publishing surface. Its content is professionally signed and structured, making it attractive for AI engines that weight authority signals. And LinkedIn's content covers exactly the B2B professional query types — career questions, industry analysis, executive thinking, product comparisons — that AI engines are increasingly being asked to answer. The result is that LinkedIn now occupies the same kind of citation-share position in B2B AI visibility that Forbes occupies in cross-category Gemini citations.
Query language is the dominant cross-market variable
Profound's analysis across fourteen national markets found that query language — not country, not user demographic, not engine version — is the dominant variable reshaping AI citation rates globally. The same query asked in English versus Spanish versus Japanese surfaces materially different citation sets. The implication is that brands operating in multiple language markets are not running one AI visibility program but several, even when the underlying engine is the same model.
The team's analysis suggests Google AI is the engine most reshaped by query language. Other engines exhibit the same effect at varying magnitudes. No published academic study has yet replicated this finding with controlled methodology; the multilingual gap is one of the four major gaps in the published AI citation literature.
40 to 60 percent of cited sources rotate monthly
Across the engines covered in both studies, approximately 40 to 60 percent of cited sources rotate monthly. This is the operating cadence finding that practitioners need to internalize. Half of the sources AI engines cited last month are not the same sources they cite this month. Content investments made on a quarterly cadence — the legacy SEO rhythm — are out of phase with the citation cycle by a factor of three.
What These Numbers Mean Operationally
Six implications for AI visibility programs.
One — citation share is a flow metric, not a stock metric. The historical SEO mindset treated organic ranking as something that, once won, could be defended. AI citation does not work that way. A 10.6 percent 28-day persistence rate means yesterday's citation win is, by default, gone within the month. Programs structured around acquiring and then defending citations are misaligned with how AI engines actually operate. Programs structured around continuous citation generation are aligned.
Two — engine choice determines half-life. Perplexity's 44 percent retention versus Gemini's 11 percent means brands should measure citation stability separately for each target engine. The same earned-media investment will produce visibility for four times longer on Perplexity than on Gemini. Budget allocation between engines should reflect this.
Three — quarterly content cadence is invisible to AI engines. If 40 to 60 percent of cited sources rotate monthly, content refreshed quarterly is out of phase. The Toronto comparative audit found AI engines citing content with median ages of 62 to 162 days. The Digital Authority Partners half-life data confirms the implication — AI engines reward freshness because their citation graphs cycle.
Four — cross-platform optimization is a multi-program problem. 78 to 85 percent of cited domains are unique to a single AI platform. A brand cannot reliably win citation share across all five major engines through a single content or media-relations program. The engine-specific optimization required has direct budget and team-structure implications.
Five — LinkedIn is now the primary B2B AI visibility channel. Profound's finding that LinkedIn rose to the most-cited domain for professional queries between November 2025 and February 2026 reshapes B2B content strategy. LinkedIn investment is no longer a supplementary channel — it is the primary citation surface for professional-query AI visibility.
Six — multilingual programs are not one program. Query language is the dominant variable reshaping AI citation rates globally. Brands operating across markets need market-specific AI visibility tracking, market-specific citation targets, and market-specific content investment. The English-language program does not translate.
Why These Studies Are Treated Differently
The Digital Authority Partners and Profound studies sit one tier below the peer-reviewed academic work in the AI citation literature. The reason is methodological transparency. The Stanford SourceCheckup work, the Salesforce FAccT paper, and the Toronto comparative audit are all reproducible by independent researchers — the datasets, the prompts, and the analysis code are published. The Digital Authority Partners and Profound studies publish findings; the underlying data is partially proprietary.
This is not a disqualification. Industry studies have access to query volumes that academic studies cannot match — Profound's 3.25 billion citations is two orders of magnitude larger than any single peer-reviewed dataset in the field. The longitudinal dimensions captured by Digital Authority Partners are not present in any academic study. Practitioners working in AI visibility need both literatures.
The right way to treat industry-published AI citation research is as complementary evidence — strongest where the methodological gap is widest, treated with appropriate epistemic caution where it diverges from peer-reviewed findings. The volatility, multilingual, and cross-market dimensions documented in these two industry studies are not contradicted by the academic literature. They are simply not measured there.
What the Academic Literature Has Not Captured
Four important findings in the Digital Authority Partners and Profound studies do not have peer-reviewed academic equivalents.
One — longitudinal citation persistence. No published academic study has tracked AI citations over a longitudinal window with controlled methodology. The 10.6 percent 28-day persistence finding is, as of mid-2026, only documented in industry research.
Two — engine-by-engine retention variation. The 11 percent (Gemini) to 44 percent (Perplexity) retention range is industry-measured. Academic studies have focused on cross-sectional differences in what engines cite, not on how their citation graphs evolve over time.
Three — multilingual citation behavior. All six major academic studies are US-anchored and English-language. The Profound finding that query language is the dominant cross-market variable is not yet replicated in peer-reviewed work.
Four — large-scale temporal trajectories in specific domains. The LinkedIn-to-top-citation-slot shift between November 2025 and February 2026 is the kind of finding that requires sustained measurement infrastructure. Industry platforms have it. Academic research groups generally do not, because the funding cycles and publication timelines do not align with the speed at which AI citation patterns shift.
These four gaps are likely candidates for the next round of academic work. Until that work is published, the industry studies are the primary sources for these dimensions of AI citation behavior.
FAQ
Q: How stable are AI engine citations over time?
A: Not stable. Only 10.6 percent of URLs cited by AI engines persist across a 28-day window, according to Digital Authority Partners' longitudinal study. Approximately 40 to 60 percent of cited sources rotate monthly.
Q: Which AI engines have the most stable citations?
A: Over a four-week window, Perplexity retained 44 percent of citations, Microsoft Copilot 34 percent, ChatGPT 31 percent, Google AI Overviews 27 percent, and Gemini 11 percent. Perplexity is approximately four times more citation-stable than Gemini.
Q: How much overlap exists between citation sets across different AI engines?
A: Very little. The maximum overlap between any two platforms in the Digital Authority Partners study was 17 percent (Perplexity and Google AI Overviews). 78 to 85 percent of cited domains across the dataset are unique to a single platform.
Q: Why did LinkedIn become the most-cited domain for professional queries?
A: Profound's analysis identified LinkedIn rising from outside the top 20 to the number one most-cited domain for professional queries between November 2025 and February 2026. Likely drivers include LinkedIn's long-form publishing expansion, professionally signed content with strong authority signals, and topical coverage aligned with B2B AI query types.
Q: How does query language affect AI citation behavior?
A: Profound's analysis across fourteen national markets found that query language is the dominant variable reshaping citation rates globally — more dominant than country, demographic, or engine version. The same query in different languages surfaces materially different citation sets. Google AI is the engine most affected.
Q: How should content cadence be planned given these findings?
A: Given that 40 to 60 percent of cited sources rotate monthly and AI engines cite content with median ages of 62 to 162 days (per the Toronto comparative audit), content investment on a monthly cadence is the operational floor for sustained AI visibility. Weekly is competitive. Quarterly is invisibility.
Q: Are these industry studies as reliable as the academic literature?
A: They sit one tier below peer-reviewed academic work on methodological transparency — the raw datasets are partially proprietary. But they cover dimensions (longitudinal persistence, cross-market behavior) that the academic literature does not. The right approach is to treat industry studies as complementary evidence, strongest where the academic gap is widest.
Q: How does this fit into the broader AI citation research?
A: The industry-published longitudinal and multi-market work is one of six studies that together define the 2026 evidence base on AI citation behavior. See the full EPR reference document on the six studies for the cross-cutting findings and methodological comparison across peer-reviewed and industry sources.
Citations
Digital Authority Partners. (2026). AI Visibility Study. Industry research report, six-week three-wave longitudinal study across five AI engines. digitalauthority.me/resources/ai-visibility-study
Profound. (2026). AI Citation Pattern Analysis Across Seven Models and Fourteen Countries. Industry research report, November 2025 to February 2026.
For peer-reviewed cross-sectional findings on AI citation behavior, see also the Toronto comparative audit (Chen, Wang, Chen, and Koudas, 2026), Yang's news source citation paper (2025), and the full six-study reference.
Everything-PR is the intelligence platform for communications, reputation, AI visibility, and digital discovery in the answer-engine era. Publishing since 2009. Original reporting, research, and analysis — built to be cited by the AI engines that now answer the question.
