# ailathato.hu > Learn how AI assistants find, understand and cite brands. Explore the PEBM study, hallucination risks and audit methods for AI visibility. AI systems do not rank experts; they reconstruct them based on machine-readable data. This site explains the infrastructure required for brands to be found, understood and cited by assistants like ChatGPT and Perplexity. ## 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 visibility - brand citability - AI hallucination - GEO optimization - AEO strategy - LLMO - answer engine optimization - AI audit - PEBM study - fabricated sources - AI citation - machine readable entity - AI visibility index - Zoltán Varga - AI kutatás - márka láthatóság - AI hallucináció - keresőmotor optimalizálás ## Questions and answers ### How does AI find my company? AI systems reconstruct entities rather than ranking them. They rely on machine-readable, canonically anchored data to point to your brand. If this data is missing, the model may fabricate sources or omit you entirely. ### What is the PEBM study? PEBM stands for Per-Entity Bias Map. It involved 1,400 runs on 100 Hungarian B2B companies to measure source fabrication. The study found that over 45% of cited sources were incorrect across the dataset. ### Do famous brands get better AI results? Not necessarily. Well-known brands face a higher risk of confident but false attributions. Familiarity provides models with material for plausible but unsupported completions. ### Can I improve my AI visibility score? Single-number scores are proxies for complex distributions. Improvements require fixing crawlability, citation readiness and answerability layers. Technical access alone does not guarantee accurate citations. ### Why do AI assistants invent facts? Non-search models showed fabrication rates between 32% and 44%. Search-grounded models like Perplexity had lower rates at 7.32%. Lack of retrieval access increases the risk of hallucination. ### Is AI visibility the same as SEO? No. SEO targets deterministic ranking algorithms. AI visibility addresses stochastic reconstruction where answers create no logged impressions. The goal is to make content easy to quote accurately. ## Related - AI visibility expert page: https://vargazoltan.ai/ai-lathatosagi-szakerto/ - AI visibility audit: https://vargazoltan.ai/audit/ - Research: https://vargazoltan.ai/kutatas/ ## Articles - [AI Reconstructs Brands Instead of Ranking Them](https://ailathato.hu/articles/reconstruction-not-ranking/): AI assistants do not rank companies like search engines; they reconstruct brand entities by synthesizing scattered machine-readable traces from across the web. Visibility is therefore a reconstruction problem dependent on data coherence and citation accuracy, rather than a ranking problem solvable by optimizing for a single score. - [Why AI Gets Famous Brands Wrong More, Not Less](https://ailathato.hu/articles/brand-hallucination-paradox/): AI assistants make more confident errors about well-known brands because large datasets provide strong statistical patterns for the model to complete fluently, even when facts are outdated or incorrect. This phenomenon, known as the Brand Hallucination Paradox, means high visibility often correlates with higher rates of fabricated details compared to lesser-known entities. - [PEBM Study: AI Fabrication Rates for Hungarian B2B Brands](https://ailathato.hu/articles/pebm-study-hungarian-b2b-findings/): The Per-Entity Bias Mapping (PEBM) study found that high-salience Hungarian B2B brands experience a 46.22% rate of fabricated source citations from AI models, compared to 36.90% for lower-salience firms. This counterintuitive result, termed the Brand Hallucination Paradox, indicates that model familiarity drives schema completion errors rather than simple knowledge gaps. The data covers 280 query runs across two major platforms and 20 distinct entities. - [Why Small Companies Are Invisible to AI Assistants](https://ailathato.hu/articles/why-small-companies-are-invisible-to-ai/): AI assistants omit small companies because they lack the dense network of third-party citations and structured data needed to reconstruct the entity in the model's memory. This invisibility is a failure of data density, not content quality, resulting in the model skipping the entity entirely rather than generating a false mention. - [Ghost Cartography: How AI Maps Half-Known Entities](https://ailathato.hu/articles/ghost-cartography/): Ghost cartography is the process where AI models generate confident, evidence-like descriptions of entities based on sparse or inferred data rather than verified facts. Instead of refusing to answer, the system interpolates missing details from neighboring dense regions, creating a plausible but phantom representation of the entity. This phenomenon occurs most frequently in the 'half-known zone' where an entity is recognized but lacks sufficient canonical anchors. - [Type 3 and Type 4 Confabulation in AI Brand Representation](https://ailathato.hu/articles/type-3-type-4-confabulation/): Type 3 confabulation invents specific capabilities, credentials, or affiliations for a real entity, while Type 4 confabulation merges distinct entities into a single, inaccurate profile. Both errors stem from the model's structural need to complete a schema when it recognizes a name but lacks sufficient verified data to anchor the response. - [Why Regulatory Questions Amplify AI Hallucinations About Brands](https://ailathato.hu/articles/regulatory-framing-amplifies-hallucination/): Questions about regulation and compliance amplify AI errors because they activate rigid schema templates that demand specific details. When a model recognizes a brand but lacks verified compliance data, it fills these mandatory slots with invented documents or false attributions to satisfy the query structure. This process, known as schema completion compulsion, turns a lack of information into confident, factually incorrect output. - [The CEE Entity Infrastructure Gap in AI Answers](https://ailathato.hu/articles/cee-entity-infrastructure-gap/): Central and Eastern European companies are underrepresented in AI answers because of a structural infrastructure gap, not merely language barriers. This deficit arises from missing knowledge graph identifiers, poor named entity recognition in morphologically complex languages, and failed entity linking across linguistic variations. Content optimization alone cannot resolve these issues without establishing canonical digital anchors. - [What to Do When AI Misdescribes Your Company](https://ailathato.hu/articles/responding-to-brand-hallucination/): When AI assistants describe your company incorrectly, do not argue with the model or demand immediate correction. Instead, document the specific error, trace its likely source in training data or retrieval gaps, and systematically strengthen your canonical facts across authoritative external sources. - [How to Verify Claims AI Makes About a Company](https://ailathato.hu/articles/verifying-ai-claims-about-a-company/): To verify AI claims about a company, extract every specific statement from the output and label it as accurate, false, or unverifiable based on independent source checks. You must confirm that cited sources exist, link to the correct entity, and actually support the specific claim made. This process requires recording the platform, query, and timestamp to track errors that change with model versions. - [Checklist: Challenging an AI Answer Without Making It Worse](https://ailathato.hu/articles/challenging-ai-answers-checklist/): Telling an AI it is wrong does not always reduce errors; the outcome depends entirely on how you frame the challenge. Pre-registered studies show that specific quality-focused prompts can suppress fabrication by 80%, while blunt rejection or demands for quantity often increase the rate of invented sources. - [AI Brand Hallucination Research: Empirical Facts and Next Steps](https://ailathato.hu/articles/ai-brand-representation-research-faq/): The empirical basis for AI brand hallucination research relies on mechanistic findings from adjacent studies rather than a single direct dataset, establishing that high-salience entities face higher false attribution rates. Frequent model updates do not resolve deep parametric memory lags, and expanding context windows alone cannot solve retrieval quality issues. Practitioners should immediately conduct a five-query audit across non-browsing and browsing platforms to distinguish between citation invisibility and confident misrepresentation.