Model Card Conversational AI
This model card describes the conversational AI behind ask impact, our chat interface for understanding your questions and generating text-based responses and content.
AI technology
Anthropic Claude (primary): Claude Opus 4.8 via Google Cloud Vertex AI, with automatic fallback to other Claude models when needed. Google Gemini (supplementary): select features, available as an alternate provider.
Model type
Large Language Model (LLM), orchestrated through a multi-agent system
Provided by
Anthropic (via Google Cloud Vertex AI); Google (via Google AI / Vertex AI)
Classification
Generative AI
Description
This is the core generative AI layer behind ask impact. A routing model interprets your message and directs it to a specialized agent for the task. Each agent uses a large language model to understand natural language, reason over retrieved context and tool results, and generate human-like responses. We combine model output with real-time data from your account, impact.com documentation, and trusted marketing resources to answer questions and generate content such as newsletters and partner-facing copy.
Intended use
Primary use cases
Answering natural-language questions about your performance analytics.
Providing industry expertise and best practices from trusted marketing resources.
Generating content for newsletters based on your prompts and account data.
Offering step-by-step guidance on how to use the impact.com platform.
Recommending partners or programs to invite or apply to.
Helping publishers create product reviews, comparisons, roundups, and other promotional content.
Surfacing publisher insights such as social content search and website/social metrics.
Live support escalation: transferring to a live agent or logging a support ticket when appropriate.
Out-of-scope uses
Making final, binding business decisions without human review.
Providing financial, legal, or other professional advice.
Generating harmful, unethical, or offensive content. This is strictly prohibited and filtered.
How we use your data
Training data. These models were pre-trained by Anthropic and Google on vast, public datasets of text and code. They are not trained on your impact.com data.
Answering your questions. When you send a message, we provide the model with task-specific, real-time context from your account only, for example your account performance data for analytics queries or relevant help documentation or marketing knowledge. That context is used only to generate a response to your immediate query.
Data privacy commitment. In line with our internal governance and Google's Vertex AI data governance, your data is never used to train or improve the model, is not stored by the model after your query completes, and is never shared or mixed with other customers' data.
Safety screening. User prompts and model responses are scanned by Google Model Armor to detect and block content that violates our usage guidelines.
Limitations & ethical considerations
Potential for inaccuracy. Like all LLMs, these models can occasionally produce inaccurate information or "hallucinate." Use AI-generated insights as a starting point and verify critical data against your own records. We include in-product disclaimers as a reminder.
Routing and tool dependencies. Responses depend on which agent handles your request and what data or tools are available for your account type. Some capabilities are limited to brand or partner users.
Governance. Use of this AI is governed by our internal AI governance framework and responsible-by-design practices, which mandate Privacy, Fairness, Transparency, Accountability, Explainability, and Safety.
Last updated
Was this helpful?