AI-Use Disclosure

Grounded autocomplete can draw continuations from passages in your library when you have PDFs; without them it drafts from the model, and those suggestions are not citations. Rewrite, expand, polish, and abstract/title drafting transform text you select. Research chat answers with scoped sources: your library, this paper, or the web. Review passes flag uncited claims and check submission readiness. AI generates prose and suggestions, but AI does not write your citations: references are retrieved from Crossref, DataCite, Semantic Scholar, and arXiv metadata and formatted by a deterministic rule.

The AI provider is Google, through the Gemini API. When you trigger an AI action, the relevant text (your draft, the passage you selected, and any library excerpts used for grounding) is sent to Google to produce the result, and processed there solely to return the output you requested.

One case is not an AI action and is easy to miss, so it is stated separately. Adding a PDF to your library also sends its extracted text to Google, in passages, to compute the embeddings that make semantic search and grounded retrieval work. That happens on upload, on fetch by DOI or arXiv identifier, and on re-extract, without you pressing anything. If there is a manuscript you are not willing to send to Google at all, do not put it in your library.

Rivul does not use your content to train any model. On the paid tier of the Gemini API, which is what Rivul runs on, Google's own terms state that it does not use prompts or responses to improve its products, and processes them as a data processor. The Privacy Policy sets out that distinction in full, because it is the sentence that matters most here. The AI is included in your Rivul plan; you do not need, and are not asked for, your own API key.

Your account (email, hashed password), documents, references, and the library files you upload or fetch, in Rivul's database and file storage. The text extracted from each library PDF, plus the passage embeddings computed from it. Usage counts (AI requests and tokens) for quota, billing, and cost visibility. The Privacy Policy lists every company that receives any of it, by name.

Rivul does not train any AI model on your papers, drafts, or library. Rivul does not sell your content.

"Portions of this manuscript were drafted with the assistance of Rivul, an AI writing tool, using our managed AI models. All AI-generated text was reviewed and edited by the authors. All citations were verified by the authors against their original sources. The authors take full responsibility for the content." Always follow your venue's specific AI-disclosure policy; it takes precedence over this text.