1. System, roles and purpose
Preferium AI Edge is developed/supplied by Preferium AS for technical SEO, content assistance and visibility measurement. Preferium may be AI-system provider and, in Managed delivery, a deployer on Customer’s behalf. Customer, Agency and End Customer roles depend on actual use, rebranding and modification.
This is not a regulator classification. Intended SEO use normally appears outside Annex III high-risk categories, but role/risk must be reassessed when purpose, function, data or market changes.
2. What it does
The system can read public site content and supported sources, generate proposals and—within selected controls—publish Edge HTML changes such as titles, descriptions, headings, structured data, alternative text and links. Generation primarily uses models available through Cloudflare Workers AI under active configuration.
ChatGPT/OpenAI, Claude/Anthropic, Perplexity and Gemini/Google may measure citations/visibility when enabled; they do not necessarily generate Customer SEO content. Active flows appear in the Subprocessor Register.
3. Controls and limits
Supported deployment flows may use guardrails, validation, logs, approval modes, verification and rollback. Not every control applies to every function/configuration and none guarantees all errors are detected. Customer must select risk-appropriate human oversight and review regulated/fact-sensitive claims. Preferium may reject/roll back unlawful, harmful, deceptive or platform-breaching output.
4. Prohibited/high-risk purposes
No prohibited AI practices, biometric/emotion recognition, social scoring or decisions on employment, education, credit, insurance, health, essential services, law enforcement, migration or justice without written approval and documented assessment. No substantial modification, intended-purpose change or rebranding that creates new provider obligations without written allocation.
5. Transparency, marking and literacy
Each party must meet transparency duties for its actual role/output. We do not categorically claim machine-generated material never needs marking. Machine-readable detectability, deepfake and public-interest text duties require output-specific analysis, including human editorial control. Customer must ensure appropriate AI literacy; Preferium provides reasonably needed operating information and investigates complaints.
6. Data, models and geography
We do not train general models for other customers on Customer Data or visitor personal data. AI/measurement data may process globally or in the US, not only EEA. See Privacy, DPA and Subprocessors.
7. Output rights and responsibility
Output may be wrong, incomplete, non-unique or not copyright-protected. Customer receives rights Preferium can lawfully grant but no guarantee of exclusivity, registrability or non-similarity. Customer remains final publisher. No ranking, indexing, citation, traffic or revenue guarantee. Report issues to post@preferium.no.