Intro (above the fold)
AI visibility advice is everywhere. Peer-reviewed evidence is rare.
The Capston Core Research Lab is where we separate the two. Every claim in the Capston methodology is traced back to either an empirical experiment we ran ourselves, or to a peer-reviewed academic paper that documents the underlying engine behaviour.
Two foundational papers anchor the lab. Ten research-grounded deep dives extend them into the specific decisions premium brands need to make about content, citations, freshness, language, and source authority.
This page is the index. Each linked article stands on its own evidence base.
Run a research-grounded baselineRead the methodology
Why research grounds the Capston Core method
Most GEO advice still circulates as folklore. “Write more FAQs.” “Add schema.” “Get cited by Reddit.” Some of it is partially true. Much of it is wrong once you test it across engines.
Capston Core was built to escape that loop.
The method only accepts a recommendation when it survives two filters: empirical evidence from our own measurement work, and convergence with at least one peer-reviewed study on how generative engines select and absorb citations. If a tactic fails either filter, it stays out of the playbook — no matter how popular it is on LinkedIn.
That is what the lab is for. It is the place where claims get tested, sourced, and either promoted into the methodology or discarded.
The result is a smaller, slower, more defensible set of recommendations. Premium brands cannot afford to chase signals that do not move answers. The lab keeps the work honest.
The two foundational papers
Two academic papers form the backbone of the Capston Core research base. Every deep dive in the lab cites one or both.
Chen et al. (2025) — “Generative Engine Optimization: How to Dominate AI Search”
Chen, Wang, Chen and Koudas published the first large-scale empirical study of how content variables affect citation likelihood across major generative engines. Their work tested formatting hypotheses, source-type effects, freshness signals, and authority proxies against real engine outputs.
The headline finding for our work: surface-level formatting tactics (Q&A blocks, bullet lists, schema) explain far less variance than source authority, evidence density, and semantic alignment. The paper is the reason Capston Core does not sell “AI-ready content templates.”
Zhang, He and Yao (2026) — “From Citation Selection to Citation Absorption”
Zhang, He and Yao extended the measurement frame in a critical direction. Earlier work asked, “What gets cited?” They asked, “What gets cited and then absorbed into the answer text itself?” The distinction matters because a cited source that contributes no language to the answer delivers far less brand value than one whose phrasing is reused.
Their measurement framework — citation selection vs. citation absorption — is now embedded throughout the Capston scoring system. It is also the title of one of the ten deep dives below.
Full references at the end of this page.
The ten research-anchored deep dives
Each deep dive takes one finding from the foundational papers, stress-tests it against our own measurement work, and translates it into a concrete decision for premium brands.
- Earned Media Bias — Why generative engines over-weight earned third-party coverage relative to owned content, and how to build a citation surface that reflects it.
- Citation Selection vs Absorption — The two-tier measurement frame from Zhang et al. (2026), and why being cited is only half the prize.
- The Q&A Formatting Myth — What the Chen et al. (2025) data actually says about FAQ blocks, and where the marginal return goes negative.
- Evidence-Container Design — Page-level patterns that make a passage extractable, attributable, and absorbable across engines.
- Engine Citation Behavior — How ChatGPT, Perplexity, Gemini and Google AI Overviews differ in source selection, and what that means for prompt-by-prompt strategy.
- The Big Brand Bias — Why entity recognition compounds, and how challenger brands close the gap without buying their way in.
- Machine Scannability — Page structure variables that move citation rates, and the ones that do not despite the folklore.
- Cross-Language Visibility — How generative engines handle bilingual brand content and where translation alone underperforms.
- Freshness Signal — When recency moves citations and when it does not, by engine and by query intent.
- Semantic Alignment — Why topical match at the passage level beats keyword density, and how to measure it.
Read in order if you want the full argument. Jump to the one that matches your current decision if you do not.
What this means for premium brands
Three implications carry across the ten deep dives.
First, owned content is necessary but not sufficient. The engines that matter most for premium discovery — ChatGPT, Perplexity, Google AI Overviews — lean heavily on third-party sources for the entity-defining passages. A brand site that does not earn citation from credible third parties will be cited less, absorbed less, and recommended less.
Second, format optimization has a low ceiling. Adding FAQ blocks and schema is fine. Treating them as a growth lever is not. The variance is in source authority and evidence density.
Third, measurement has to be longitudinal. Engine behaviour drifts. A page that earns citation absorption in March may lose it by September without a single content change, because the model rebalanced its source weighting. Quarterly retests are the floor, not the ceiling.
How this fits into Capston Core
The research lab is one layer of the Capston Core system, not the whole thing.
It feeds the Capston Core methodology, informs the AI visibility scoring dimensions, and provides the evidence trail that the data and evidence layer requires. It also defines what certified partners must understand before they can sell work under the Capston method — see Capston QA standards and certified Capston partners.
The lab does not replace execution. It disciplines it.
→ Back to Capston Core
FAQ
Is the research lab a publication channel?
No. It is the public-facing index of the evidence base behind the Capston methodology. We publish synthesis and decision implications, not raw datasets.
Are the foundational papers the only sources?
No. They are the two that most directly shape the measurement frame. Each deep dive cites additional empirical work and our own measurement results where relevant.
How often is the lab updated?
New deep dives are added when a finding has survived both filters — own measurement and peer-reviewed convergence. Existing pages are revised when engine behaviour shifts materially.
References
Chen, M., Wang, X., Chen, K., & Koudas, N. (2025). Generative Engine Optimization: How to Dominate AI Search. arXiv:2509.08919v1. https://arxiv.org/abs/2509.08919
Zhang, K., He, X., & Yao, J. (2026). From Citation Selection to Citation Absorption: A Measurement Framework for Generative Engine Optimization Across AI Search Platforms. arXiv:2604.25707v2. https://arxiv.org/abs/2604.25707
Final CTA block
Get a research-grounded baseline for your brand.
See how generative engines currently select and absorb sources about you, where the citation surface is weakest, and which lab finding applies to your next decision.
Run a baseline
Read the methodology