Models We Use
Our product uses a curated set of large language models served through our AI gateway. The specific models used for a given feature are chosen based on quality, latency, and cost tradeoffs and are documented in the product. We do not use single-vendor lock-in; we select per-feature based on what produces the most useful output for the task.
What AI Does Inside the Product
AI assists customers with drafting, rewriting, tone adjustment, grammar correction, summarization, content analysis, and contextual recommendations. AI also powers our search, our personalization layer, and our governance signals (e.g. brand voice drift detection). In every case, AI output is presented as a recommendation, not a directive; the human user remains the author of record.
What AI Does Not Do
AI does not autonomously send communication on a user's behalf without explicit human approval. AI does not make irreversible decisions about customer accounts. AI does not produce editorial content that is published under a human contributor's name without that contributor's review. AI is not used to evaluate individual employee performance inside our customers' organizations.
Data Handling
Customer text submitted to our AI features is processed in transit, used to produce the requested output, and not retained for model training unless the customer explicitly opts in. We do not sell customer content. We do not use customer content to train third-party models. Our data processing agreements with model providers prohibit retention beyond what is required for abuse monitoring.
Hallucination and Reliability
Large language models can produce confident, fluent text that is factually wrong. We design our product to minimize this risk — through grounding in customer-provided context, conservative output framing, and explicit uncertainty signals — but we do not claim it is eliminated. Customers should review AI-generated content before sending or publishing, especially when factual accuracy is critical.
Editorial Use of AI
AI may be used in our editorial process for first drafts, copyediting suggestions, and research summarization. A human editor of record reviews and approves every published article. We do not publish AI-generated content under a human byline as if it were the human's original writing.
Customer Controls
Customers on team and enterprise plans can disable AI features, restrict which models are used, opt out of personalization, and require human approval workflows for AI-suggested actions. Administrators have audit access to AI usage across their organization.
Governance and Updates
This policy is reviewed quarterly and whenever a material change is made to how we use AI. Material updates are announced to customers and the version history is preserved.