Frequently Asked Questions
Do you use our data — or your content — to train AI models?
No. CARAVAN does not license, sell, or supply its content library as training data for third-party AI models, and we do not use client or user data to train models. Our AI is applied to our own licensed library to build metadata and adapt content for the platforms we serve. That is the entire scope.
What does the AI do, and where does it run?
We use generative AI to tag, translate and adapt content from our own library. It doesn’t score, predict or make automated decisions about users. It runs on our side, before delivery: we enrich and adapt content within CARAVAN’s environment, then hand your platform a finished record or asset. Nothing about the integration requires you to run AI infrastructure of your own.
How do you prevent hallucinations and bias?
Our AI works only within CARAVAN’s expert-vetted library, never the open web, which limits the risk of hallucinations. Then a person checks everything. Our process is human-in-the-loop: CARAVAN’s expert reviewers check every mapping, tag, translation, and adaptation for factual accuracy, fairness and potential bias, and correct anything before delivery.
Does using AI change the compliance picture?
No. The same SOC 2, HIPAA, and GDPR commitments that govern our licensing apply here. No protected health information (PHI) or personally identifiable information (PII) enters our AI pipelines — enrichment operates on content, not on people. We use third-party AI models for defined tasks, CARAVAN remains accountable for everything we deliver, and we monitor output quality over time.
How is adapted content different from machine translation?
Machine translation swaps words. Adaptation changes reading level, cultural references, examples, foods, measurements, length, and format — then puts the result in front of a qualified reviewer who speaks the language and knows the subject. The content stays identical; everything around it is rebuilt for the audience.
What do you need from us to get started?
Your taxonomy or content schema, a set of sample records, and roughly an hour with whoever owns your content structure. We conform to your field names, enums and identifiers rather than asking you to adopt ours. No formal taxonomy? Our team will build one for you, designed around your platform and user segments, so your content is easier to find, recommend and personalize from day one.
What does it cost, and can we try it first?
Metadata enrichment and content adaptation are configured as part of our services, scoped to the size of your library and the number of variants you need. Before getting started, we provide a sample dataset mapped to your taxonomy so your content and engineering teams can see the quality, structure, and fit for themselves. Contact us to learn more.











