AI processing disclosure
Useful context, not perfect truth.
AI helps WordHalo explain language without interrupting the reader, but its output has limits.
Last updated: July 25, 2026Where AI is used
WordHalo may use an OpenAI model for natural translation, context-aware definitions, pronunciation guidance, examples, passage summaries, and explanations of how selected text functions in nearby text. English dictionary-only results may come from a separate dictionary provider without consuming an AI allowance.
What is sent
After user consent, WordHalo sends the intentionally highlighted text, up to 1,000 characters of nearby context, the user's selected interface language, and the webpage's language hint when available. WordHalo detects the highlighted text's language to provide the requested result. The extension does not send the complete URL, complete webpage, saved-word library, or browsing history.
Accuracy and cultural limits
AI can choose the wrong sense, mistranslate tone, omit nuance, hallucinate, or produce culturally unsuitable wording. A result should be treated as reading assistance. Important decisions require an authoritative source or qualified human.
Provider settings
Requests use application storage disabled where supported. Provider abuse-monitoring retention may still apply. WordHalo must verify its OpenAI organization data settings before public store submission and must not claim Zero Data Retention unless formally approved and enabled.
Human review and training
Humans do not routinely read highlighted text. WordHalo does not use reading selections to train its own model. Access may occur only for a user-requested support case, security investigation, or legal obligation, with access limited to what is necessary.