AI Usage Policy

1. Overview

liaisonIQ includes AI-powered features including LiaisonIQ Assist (AI chat), route planning, visit prep, Liaison Coach, note enrichment, voice-note transcription, agent-assisted onboarding and target discovery, AI expense review, and the AI Territory Agent (automated competitor, relationship-risk, new-provider, and news monitoring). These features are powered by large language models (LLMs) and search APIs from third-party providers, currently including Google Gemini and, for the Territory Agent's news monitoring, Tavily.

AI features are designed to help field representatives work more efficiently. They are assistive tools — not authoritative sources of medical, clinical, legal, regulatory, or compliance information.

This policy explains how AI features work, what data is processed, what the AI can and cannot do, and what your responsibilities are as a user.

2. What Data Is Sent to AI

When you use an AI feature, the text you input — including your questions, visit notes, clinic names, territory descriptions, and workspace context — is sent to our AI provider to generate a response.

We also send contextual data from your workspace to help the AI give relevant responses. This may include: clinic names and addresses, visit history summaries, task lists, territory information, your company name and offerings, and relationship strategy tags.

The AI Territory Agent works differently from on-demand features: on a daily schedule per workspace (not only when you open the app), it sends competitor names and general workspace region information to Gemini, and to Tavily for public news search, to generate the Opportunity Feed and Activity Timeline.

We use prompt engineering and input filtering to limit what is sent to the minimum necessary to generate a useful response. We do not intentionally send your password, payment information, or full account credentials to any AI provider.

Before text is sent to an external AI model, we automatically mask obvious identifiers in it, such as email addresses, phone numbers, Social Security-style numbers, medical record and member ID numbers, labeled dates of birth, and long digit strings. This is a safety net, not a guarantee: it does not catch every identifier (for example, names), so you must still follow the no-PHI rule in Section 3.

Your data goes to an AI model only when you use an AI feature or when a workspace feature you or your owner turned on runs (such as the daily Territory Agent).

If you dictate a voice note, the audio is sent to a speech-to-text provider (currently Groq or Google Gemini) to produce a transcript; we do not store the audio.

LiaisonIQ Assist can look up your workspace data and, when you ask, take actions such as creating tasks, scheduling or logging visits, changing an account's stage, or drafting messages. It asks you to confirm changes before making them, caps how many changes it makes per message, and can still misunderstand a request, so review what it did.

3. No Protected Health Information (PHI)

You must NOT enter into any AI feature: patient names, dates of birth, Social Security numbers, medical record numbers, diagnosis codes, procedure codes, health plan beneficiary numbers, or any other information that constitutes Protected Health Information (PHI) under HIPAA or equivalent laws.

This prohibition applies to all AI-powered features: AI chat (LiaisonIQ Assist), visit prep, note enrichment, Liaison Coach, route planning, and any other AI-powered feature.

This prohibition covers all patient information, not only the identifiers listed above, and applies to voice notes, files, and screenshots you submit. liaisonIQ is not a HIPAA business associate and no BAA is in place for standard plans or pilots; we may delete suspected PHI without notice.

The purpose of liaisonIQ is to manage your field operations and provider relationships — not to process patient clinical data. Use de-identified, aggregate, or account-level descriptions when discussing your business with AI features.

If you require a HIPAA-compliant environment, contact us at support@liaisoniq.app to discuss enterprise arrangements.

4. AI Output Limitations

AI output may be incomplete, inaccurate, outdated, biased, or inappropriate for your specific context. Factors that affect AI quality include:

• The completeness and accuracy of your workspace data (clinic profiles, visit history, territory information)
• The specificity of your prompt — vague questions produce general answers
• The knowledge cutoff of the underlying AI model — recent events or regulatory changes may not be reflected
• Vertical-specific nuance — liaisonIQ serves healthcare, pharma, dental, cardiology, ASC, and other verticals; AI output is not specifically calibrated to the compliance requirements of each vertical

The AI Territory Agent's opportunity, risk, and news findings are inferred from public and workspace data and are not verified by liaisonIQ. Treat them as leads to investigate, not confirmed facts.

Always treat AI output as a starting point for your own thinking, not as a definitive answer.

5. No Professional Advice

AI output from liaisonIQ is not and should not be construed as:

• Medical or clinical advice
• Legal or regulatory advice
• Financial or accounting advice
• Compliance guidance for any specific regulatory framework (e.g., HIPAA, Stark Law, Anti-Kickback Statute, FDA regulations)
• HR or employment advice

For any decision with legal, medical, financial, or compliance implications, consult a qualified professional in the relevant field.

6. Your Responsibility

You are solely responsible for reviewing, verifying, and acting on AI output. liaisonIQ disclaims all liability for decisions, actions, or outcomes based on AI-generated content.

This includes but is not limited to: route plans used for driving, talking points used in physician conversations, coaching scripts, clinic recommendations, referral data interpretations, and AI-generated analytics summaries.

Before acting on AI output in a professional context, verify the information from authoritative sources and apply your own professional judgment.

7. Content Moderation and Safety

AI inputs are filtered for safety and appropriateness before being sent to our AI provider. We apply safety instructions designed to keep AI responses professional, relevant, and appropriate for a business context.

We may log and review AI queries that are flagged by safety filters to improve our filtering and prevent misuse. Flagged queries are reviewed by authorized liaisonIQ staff under our data access policy.

Prompts, responses, and performance metadata may also be captured in an access-restricted AI observability tool for quality monitoring and debugging.

You must not attempt to circumvent safety filters, "jailbreak" the AI, or use AI features for purposes other than legitimate field operations.

8. AI Provider and Model

liaisonIQ currently uses Google Gemini as its primary AI provider via the Google AI API. Google processes AI inputs under its API terms of service and privacy policy. For API (non-consumer) usage, Google states that it does not use API input/output data to train its models by default.

The Territory Agent's News & Market feature additionally uses Tavily's news search API to find publicly available articles relevant to your competitors or region; Tavily processes only competitor names and general region terms for this purpose.

We may change AI providers or models without notice if doing so improves quality, safety, or cost-efficiency. We will update this policy if we change to a provider with materially different privacy practices.

Prompt caching may be used for deterministic, read-only AI requests to reduce latency and API costs. Cached responses are not shared across users.

8a. Model Training and Data Use

We do not use your content (prompts, notes, workspace data, AI responses, or voice recordings) to train or fine-tune our own AI models, and we do not sell it or use it for advertising.

We send your content to third-party AI providers only to generate the response you asked for. We choose commercial API services whose terms state that customer API content is not used to train their models by default, and we will update this policy if that changes. Provider terms govern what each provider does with data, including how long it keeps it for abuse monitoring.

Our access-restricted AI observability tool stores prompts, responses, and performance metadata so we can monitor quality and debug problems. It is used for monitoring and troubleshooting and not for training models.

You can choose not to use any AI feature. Voice dictation, AI chat, and the other features described above are optional.

9. Feedback and Corrections

If AI output is factually wrong, harmful, inappropriate, or not useful, please report it via Profile → Contact Support in the app. Describe the feature used, the prompt you submitted (excluding any sensitive information), and the problematic output.

Your feedback helps us improve prompt engineering and AI quality. We do not use your feedback to directly train AI models; feedback informs our own engineering and filtering improvements.

10. Changes to This Policy

We may update this AI Usage Policy as AI technology evolves, as we change providers or models, or as regulatory guidance develops. Material changes will be communicated with at least 30 days' notice.

Questions? Contact us at support@liaisoniq.app.