Skip to content
Barkhausen AI

Reference

Annotated bibliography

The field's primary literature — measurement studies, statistical foundations, corpus and crawling research, adoption evidence, and the reporting-guideline tradition — each with a short appraisal and its source.

Last reviewed 2026.

Foundational measurement studies

The small body of work that measures visibility in AI-generated answers directly, rather than by proxy.

Aggarwal, Murahari, Rajpurohit, Kalyan, Narasimhan & Deshpande · KDD 2024 (arXiv:2311.09735) · 2024

GEO: Generative Engine Optimization

Introduces the term and the first controlled benchmark — 10,000 queries — for how content-level edits change a source's share of a generated answer, establishing that adding quotations, statistics, and citations raises that share while keyword stuffing scores below the unmodified baseline. Its limit is scope: the primary engine is a GPT-3.5 simulation and the headline number is the authors' own position-adjusted word-count metric, so the magnitudes do not transfer to production engines.

Jack, Lehman, Maloney & Xu · arXiv:2605.27440 · 2026

Paraphrase Brittleness in Production Retrieval-Augmented Commercial Recommendation: Reproducibility Below the Rerun-Stability Baseline

Measures how much a production retrieval-augmented system's output changes when a query is reworded without changing its meaning, and reports reproducibility falling below the system's own rerun-stability baseline — evidence that paraphrase-level variation is a first-order source of measurement noise. The effect size is specific to the single commercial system studied.

Schulte, Bleeker & Kaufmann · arXiv:2604.07585 · 2026

Don't Measure Once: Measuring Visibility in AI Search (GEO)

Argues that a single query to an answer engine is an unreliable visibility measurement and quantifies the variance a one-shot reading hides, making the methodological case for repeated sampling and interval estimates over point readings. It is a methods argument on sampled engines, not a standing benchmark.

Statistical foundations

The estimation, change-detection, and multiplicity results the measurement and sampling conventions rest on.

Wilson, E. B. · J. Amer. Statist. Assoc. 22(158), 209–212 · 1927

Probable Inference, the Law of Succession, and Statistical Inference

Derives the score confidence interval for a proportion that carries its name — the interval this publication's conventions use for a visibility probability estimated from repeated sampling, because it behaves well at the small counts and extreme rates that naive intervals mishandle.

Efron, B. & Morris, C. · J. Amer. Statist. Assoc. 70(350), 311–319 · 1975

Data analysis using Stein's estimator and its generalizations

Demonstrates empirical-Bayes shrinkage — that estimating many related quantities jointly beats estimating each one alone — which is the statistical basis for pooling across query cells rather than reading each in isolation. The gains depend on the units being genuinely related.

Gelman, A. & Hill, J. · Cambridge University Press (book) · 2007

Data Analysis Using Regression and Multilevel/Hierarchical Models

The standard applied treatment of hierarchical models, and the reference for the multilevel estimator the sampling convention specifies when pooling visibility across cells. As a textbook it standardizes practice rather than establishing a new result; being a print book, it carries no URL to cite.

Benjamini, Y. & Hochberg, Y. · J. R. Stat. Soc. Ser. B 57(1), 289–300 · 1995

Controlling the false discovery rate: a practical and powerful approach to multiple testing

Introduces false-discovery-rate control, the multiplicity correction the Barkhausen Criterion invokes when many query cells are monitored at once — less conservative than family-wise control and appropriate when a bounded rate of false positives is tolerable. It is the standard tool for the multiplicity disclosure the Criterion requires.

Page, E. S. · Biometrika 41(1-2), 100–115 · 1954

Continuous inspection schemes

The CUSUM change-detection method, and the classical basis for flagging the engine-change points the Criterion excludes — a broad, synchronous shift across unrelated queries. It establishes sequential change detection as a frequentist counterpart to the Bayesian method below.

Adams, R. P. & MacKay, D. J. C. · arXiv:0710.3742 · 2007

Bayesian online changepoint detection

Gives an online Bayesian method for detecting when a data stream's generating process changes, applicable to distinguishing a genuine visibility shift from an engine-change artifact in a running monitor. It is a preprint, and its behavior depends on the priors chosen.

Wald, A. · Ann. Math. Statist. 16(2), 117–186 · 1945

Sequential Tests of Statistical Hypotheses

Founds sequential analysis — testing as evidence accrues rather than at a fixed sample size — which is the theoretical root of the sustainment logic behind checking a visibility change across successive windows. It establishes that a stopping rule changes the inference that follows.

Lan, K. K. G. & DeMets, D. L. · Biometrika 70(3), 659–663 · 1983

Discrete sequential boundaries for clinical trials

Provides the alpha-spending approach to interim analyses, controlling error when a result is inspected repeatedly over time — the discipline a sustained-monitoring visibility claim needs so that looking often does not inflate significance. It was developed for clinical trials; the transfer here is by analogy.

Webber, W.; Moffat, A. & Zobel, J. · ACM Trans. Inf. Syst. 28(4), Article 20 · 2010

A similarity measure for indefinite rankings

Defines rank-biased overlap, a top-weighted similarity for comparing two rankings that may differ in length and membership — the appropriate measure for the ranked source lists AI answers return, where the top positions carry most of the weight. It is a principled alternative to naive list overlap.

Corpus & crawling

The datasets and protocols that determine what a model can have seen and who may fetch a page.

Penedo et al. · NeurIPS 2024 Datasets & Benchmarks (arXiv:2406.17557) · 2024

The FineWeb Datasets: Decanting the Web for the Finest Text Data at Scale

Documents the construction and filtering of a large open web-text training corpus derived from Common Crawl, with ablations showing how filtering choices change downstream model quality — a concrete measure of how much curation sits between a raw crawl and training data. It is one corpus's recipe, not a general law.

Koster, Illyes, Zeller & Sassman · IETF, RFC 9309 · 2022

RFC 9309: Robots Exclusion Protocol

Standardizes robots.txt and states plainly that its rules "are not a form of access authorization" — the boundary every crawler-access analysis in this publication rests on. It defines a request mechanism, not an enforcement one.

Le Pochat, Van Goethem, Tajalizadehkhoob, Korczyński & Joosen · NDSS 2019 · 2019

Tranco: A Research-Oriented Top Sites Ranking Hardened Against Manipulation

Provides a reproducible, manipulation-resistant ranking of popular domains for use as a research sampling frame — the standard answer to "which sites" when a census needs a defensible population. It also establishes that popularity lists used as frames must themselves be audited.

Adoption & behavior evidence

Primary sources on whether and how audiences use AI assistants to make decisions. Most are self-published by organizations with a commercial or advocacy interest in the finding; each appraisal notes the publisher's vantage, which a reader weighs alongside the number.

Carnegie Higher Ed · Summer Research Series · 2025

AI Use in the College Search

A survey of prospective US college students reporting a substantial minority already using AI tools in their college search. Carnegie is an enrollment-marketing firm serving universities, so the finding aligns with its service; read as one self-published data point.

EAB · Newsroom / press · 2026

Nearly Half of High School Students Now Use AI in College Search

Reports that close to half of surveyed high-school students use AI in the college search. EAB is an education research-and-consulting firm whose clients are institutions responding to these shifts; a single-publisher survey, with the headline stated in the title.

INTO University Partnerships (T. O'Brien, SVP), via ICEF Monitor · ICEF Monitor (trade press) · 2026

The ChatGPT generation: how AI is quietly rewriting the global student search experience

A pathway-provider executive's account, published as a guest article in an international-education trade outlet, of AI's role in student recruitment. It establishes practitioner perception; the vantage is a recruitment business and the venue is trade press, not a peer-reviewed study.

UCAS (J. Richards, Senior Insight Lead), via Wonkhe · Wonkhe (commentary) · 2026

Three ways prospective students are using AI when applying to higher education

The UK's central admissions service, writing in a UK higher-education commentary outlet, on how applicants use AI. UCAS observes application behavior directly, which strengthens the vantage; the piece is commentary drawing on that data rather than a released dataset.

IDP Education (Emerging Futures survey), via ICEF Monitor · ICEF Monitor (trade press) · 2025

Growing use of AI for study abroad decisions highlights importance of multi-channel marketing strategies

A large recurring survey of prospective international students, reported through trade press, finding rising use of AI in study-abroad decisions. IDP is a student-placement company, so the survey serves its market intelligence; the Emerging Futures instrument is sizeable, weighed against that commercial vantage.

Pew Research Center · Internet & Technology · 2026

How Teens Use and View AI

A US survey of teenagers' AI use and attitudes from an established nonprofit research organization with no commercial stake in the result — among the strongest-vantage sources here. It remains a stated-behavior survey rather than observed use.

China Internet Network Information Center (CNNIC) · Official report (title translated) · 2025

Report on the Development of Generative Artificial Intelligence Applications (2025)

The official Chinese internet-statistics body's report on generative-AI adoption, and the main public source for user counts in that market — establishing scale where Western surveys have no reach. Read with the framing that its methodology and definitions differ from those surveys.

ACT · Research blog · 2023

Half of High School Students Already Use AI Tools

An early 2023 survey from the US testing organization finding half of high-school students already using AI tools — useful as a dated baseline showing how substantial adoption was even then. The vantage is an assessment company, and the figure is now old enough to read as a floor rather than a current rate.

Everspring · 2025 Higher Ed Trend Report · 2025

AI Has Changed How Students Search — And Universities Are Paying the Price

A trend report arguing that AI-mediated search is diverting traffic universities previously captured. Everspring is an online-program-management firm whose services address exactly this problem, so the framing serves its offering; read for its claim, not as an independent measurement.

Cues.ai · Vendor analysis · 2025

ChatGPT trends: referral traffic to UK universities

Reports referral-traffic patterns from ChatGPT to UK university sites. Cues.ai is a vendor in the AI-visibility space, so this is a metric its product concerns; the underlying signal is observational referral traffic rather than survey response — a different and harder-to-game measure, weighed against the single-source vantage.

Reporting-guideline tradition

Cross-disciplinary precedents for the disclosure and pre-registration discipline this publication adopts.

Chambers, C. D. · Cortex 49(3), 609–610 · 2013

Registered Reports: A new publishing initiative at Cortex

Launches the Registered Reports format, in which a study's methods are reviewed and accepted before its results are known, decoupling publication from outcome. It is the precedent for pre-registering a visibility study's analysis before any sampling.

Nosek, Ebersole, DeHaven & Mellor · Proc. Natl. Acad. Sci. 115(11), 2600–2606 · 2018

The preregistration revolution

Argues for preregistration as the remedy to analytic flexibility and the file-drawer problem, distinguishing prediction from postdiction. It is the methodological rationale for fixing a visibility study's design in advance.