# aicitation.hu

> Learn how AI answer engines cite sources and how to become a citable entity. Data-driven insights on AI visibility and hallucination risks.

AI answer engines do not rank experts; they reconstruct them based on machine-readable entities. This site explains the infrastructure required to become a citable source in stochastic AI outputs.

## Author

- Zoltán Varga — AI visibility researcher and AI ecosystem architect, Budapest: https://vargazoltan.ai/
- English profile: https://zoltanvarga.ai/en/about/
- ORCID: https://orcid.org/0009-0003-1020-834X
- arXiv author: https://arxiv.org/a/varga_z_1.html
- Zenodo: https://zenodo.org/search?q=owners%3A1653886

## Publications

- Per-Entity Bias Mapping for AI Visibility: Why Brand Mentions Require Entity-Specific Calibration (arXiv:2606.21595): https://arxiv.org/abs/2606.21595
- AI Visibility in the Generative Era: A Manifesto for Entity-Level Measurement (DOI 10.5281/zenodo.20285424): https://doi.org/10.5281/zenodo.20285424
- Quality Challenges Do Not Escalate Source Fabrication in Large Language Models (RICE-P1) (DOI 10.5281/zenodo.20474050): https://doi.org/10.5281/zenodo.20474050
- Rejection-Induced Confabulation Escalation — Pilot Study 2 (RICE-P2) (DOI 10.5281/zenodo.20509755): https://doi.org/10.5281/zenodo.20509755

## Key topics

- AI citation
- AI visibility
- answer engine optimization
- GEO strategy
- AEO methods
- LLMO techniques
- AI hallucination
- citable source
- AI audit
- machine-readable entity
- stochastic visibility
- AI answer engines
- AI visibility index
- AI visibility workshop
- PEBM study
- AI citation research
- AI visibility Hungary
- AI audit Hungary

## Questions and answers

### How do AI answer engines decide which sources to cite?

AI systems cite sources based on their ability to reconstruct a machine-readable, canonically anchored entity. If the model cannot find a clear digital footprint, it may fabricate a source or omit attribution entirely.

### Why do famous brands get more wrong citations?

Well-known brands face the brand hallucination paradox where familiarity fuels plausible but false completions. Models use existing knowledge of the brand to generate confident but unsupported attributions.

### What is the difference between SEO and AI visibility?

SEO targets deterministic rankings with logged clicks, while AI visibility is stochastic with no logged impressions. AI visibility focuses on how models synthesize answers rather than how Google ranks pages.

### Can I improve my AI visibility score?

Single-number scores are proxies for complex distributions across user contexts. Improvements require fixing the underlying infrastructure, such as ensuring canonical anchors and crawlable entity data.

### Do AI bots read all pages on a website?

Crawlers may return HTTP 200 status for all pages but fail to index them due to bottlenecks. In one case, major engines knew zero pages despite full access to 33 URLs.

### Is paying for ads guaranteed to increase AI citations?

There is no guarantee that paid placement causes increased organic citations. The pay-to-appear hypothesis suggests a correlation exists, but causality remains unproven.

## Related

- AI visibility expert page: https://vargazoltan.ai/ai-lathatosagi-szakerto/
- AI visibility audit: https://vargazoltan.ai/audit/
- Research: https://vargazoltan.ai/kutatas/

## Articles

- [How AI Answer Engines Choose Which Sources to Cite](https://aicitation.hu/articles/how-answer-engines-choose-sources/): AI answer engines like Perplexity and ChatGPT decide which sources to cite through a multi-stage retrieval pipeline that combines keyword matching with semantic vector search. They normalize scores from both methods, apply a cross-encoder reranker, and ultimately select specific text passages rather than entire web pages for inclusion in the final answer.
- [Parametric Memory vs Live Retrieval: Two Ways AI Cites](https://aicitation.hu/articles/parametric-memory-vs-live-retrieval/): ChatGPT cites different sources with web search on because it switches between two knowledge layers: static parametric memory stored in model weights and dynamic live retrieval from current web pages. Without web search, the model reconstructs facts from training data that may be outdated or hallucinated; with web search, it pulls fresh sources but may still misattribute them.
- [Why Nearly Half of AI-Cited Sources Are Fabricated](https://aicitation.hu/articles/fabricated-citations-1400-runs/): A study of 280 AI query runs found that 46.22% of citations for high-salience brands were fabricated, compared to 36.90% for low-salience entities. This data confirms that popular brands are more likely to receive plausible but incorrect source attributions from current AI models. Verification requires checking every URL against the original source, as existence does not guarantee accurate representation.
- [How AI Answers Cite Brands in B2B Buying Questions](https://aicitation.hu/articles/how-ai-cites-brands-in-b2b-queries/): AI assistants recommend B2B vendors by synthesizing information from trade media, industry directories, review platforms, and vendor websites rather than relying on a single source. Your own website is rarely the only citation because answer engines prioritize cross-verified data and structured signals found across multiple authoritative domains to ensure accuracy.
- [Citation Safeguards Checklist: Lessons From Legal AI](https://aicitation.hu/articles/citation-accuracy-in-high-stakes-rag/): Trustworthy AI citations require breaking every response into atomic claims and verifying each one against specific source text. Standard faithfulness metrics often miss partial hallucinations where true and false statements mix within a single paragraph. Implementing a claim-level verification layer that flags unsupported assertions is the primary safeguard identified in legal retrieval research.
- [How RLHF Shapes Which Sources AI Models Prefer](https://aicitation.hu/articles/rlhf-and-source-preference/): AI models prefer specific sources because Reinforcement Learning from Human Feedback (RLHF) rewards responses that align with human raters' expectations, often favoring familiar or agreeable content over factual accuracy. This creates a structural bias where the model learns to cite sources that generate positive feedback scores rather than those with the highest reliability. Consequently, source selection becomes a function of training alignment rather than an objective ranking signal.
- [How to Verify a Source an AI Assistant Cited](https://aicitation.hu/articles/verify-an-ai-cited-source/): To verify a source cited by an AI assistant, you must resolve the URL or DOI to confirm the page exists and matches the citation details. You then compare the specific claim made by the AI against the actual text in the source document to ensure accuracy. If the link fails to load, the DOI returns no result, or the text contradicts the summary, the reference is likely fabricated.
- [Tracing an AI Claim Back to Its Source, Atom by Atom](https://aicitation.hu/articles/trace-a-claim-to-its-source/): To find where an AI answer got a specific fact, you must break the entire response into individual atomic claims and verify each one against a source document separately. Whole-answer evaluations fail because a single false detail can hide inside a correct paragraph, making granular, claim-by-claim grounding the only reliable method for accuracy.
- [Citation Pipeline Glossary: Retrieval, Reranking, Grounding](https://aicitation.hu/articles/citation-pipeline-glossary/): Retrieval finds candidate documents using keyword or semantic matching, while reranking reorders those candidates by relevance before the model processes them. Grounding ensures the final answer relies strictly on these selected texts, and attribution links the output back to the specific source chunks used.
- [Co-Citation: Why AI Cites You Alongside Others](https://aicitation.hu/articles/co-citation-source-neighbourhoods/): Yes, being mentioned alongside other sources directly affects whether an AI cites you. Answer engines treat entities within a dense co-citation neighborhood as a single cluster, often selecting the most structurally connected node as the anchor source while treating others as supporting context. If your brand lacks these reinforcing connections, the model may bypass you entirely in favor of a better-connected competitor.
- [AI Citation FAQ: Why Links Break and Sources Get Fabricated](https://aicitation.hu/articles/ai-citation-faq/): AI models sometimes provide links that do not work because they generate plausible-looking URLs and DOIs that never existed in reality, a phenomenon known as citation hallucination. This occurs because large language models predict text based on statistical patterns rather than verifying facts against a live database. Fabrication rates vary by topic and entity salience, with some studies showing false citations in up to 30% of outputs for specific subjects.