AI Content Policy

Content Policy

Guidelines for AI-generated content in ClientPulse

Last updated: April 2026

Next review: July 2026

Sections

Overview

ClientPulse generates AI-powered outputs to help agencies manage client relationships effectively. This policy defines what content we generate, our principles for generation, and what we explicitly will not create.

What ClientPulse generates:

  • • Health Scores: Composite 0–100 client health indicators
  • • Meeting Summaries: Automated transcription and extraction of action items
  • • Churn Predictions: Probability-based risk assessments
  • • Action Suggestions: Proactive recommendations for client engagement
  • • Monday Briefs: Weekly client summaries
  • • Save Plans: Structured churn-prevention proposals

Generation Principles

Accuracy over Speculation

We prioritize factual, data-grounded insights. We never guess or extrapolate beyond available signals.

Data-Grounded Only

All outputs reference concrete data points: financial records, meeting transcripts, engagement metrics.

No Fabricated Quotes

We never create fake client statements or meeting quotes. All quoted material is extracted from actual transcripts.

No Financial Advice

We do not generate financial recommendations, investment guidance, or tax advice.

No Legal Recommendations

We do not provide legal interpretations or contract advice.

What We Will Not Generate

Discriminatory Content

Content targeting protected classes (race, gender, religion, national origin, age, disability status, sexual orientation).

Manipulation & Coercion

Content designed to manipulate client relationships, coerce contract renewal, or exploit emotional pressure.

Fake Testimonials

Synthetic client feedback or fabricated success stories.

Unsubstantiated Claims

Claims about client intent, behavior, or future actions without evidence from data.

Human Oversight

All automated outputs requiring client communication are queued for human review and approval before delivery.

1

Content Generation

AI generates briefs, alerts, save plans, and meeting invitations

2

Approval Queue

Content held in approval dashboard for agency owner review

3

Human Review & Approval

Agency owner must explicitly approve each piece of content before it reaches the client

4

Optional Auto-Approve

Once confidence is established, agency owners may opt-in to auto-approval for specific action types

Full Agency Control

Agency owners maintain full control over all client-facing communications. No outbound content is sent without explicit human authorization in the current sprint.

Data Handling

Data We Process

  • • Meeting transcripts (audio → text via Whisper)
  • • Financial data from Stripe (invoices, payment history, disputes)
  • • Calendar data (meeting frequency and dates)
  • • Client metadata (name, contract value, engagement history)
  • • Agency owner contact information for communications

Data We Do NOT Store

  • • Raw audio files (deleted after transcription)
  • • Personally identifiable information of end-clients beyond business name
  • • Biometric data
  • • Health or employment records

Retention Policy

Client data retained for 24 months after contract termination; deletion available upon request. GDPR right-to-delete honored within 30 days.

EU Hosting

All data processed and stored in Frankfurt, Germany (AWS EU region) for GDPR compliance.

Accuracy & Limitations

Health Scores are Probabilistic, Not Deterministic

Scores reflect trends and patterns, not certainties. A client with a health score of 45 is at elevated risk, but not guaranteed to churn.

Churn Predictions Based on Available Signals

Predictions depend on data quality and historical coverage. Clients with fewer than 30 days of history will have lower-confidence predictions.

Sentiment Analysis Limitations

English-only in v1.0; known limitations with sarcasm, cultural context, and tone.

Transcription Accuracy Varies

Whisper transcription quality depends on audio input quality and speaker diarization. Heavy accents, background noise, and technical jargon may reduce accuracy.

Disputed Outputs

If an AI-generated output is inaccurate or misleading, users can flag it directly in the dashboard.

Correction Workflow

Flagged outputs reviewed by support team within 48 hours. Corrections applied to future outputs.

Feedback Loop

All corrections and disputes logged and used to improve model behavior and decision logic.

Escalation Path

Users may dispute any AI-generated output and request human review.

Contact for AI Decisions

hello@helloaurora.ai

Subject: “AI Output Dispute” + specific health score ID or output reference

Response Time

48-hour acknowledgment; 5-business-day resolution target

Right to Human Review

Any user may request human review of any AI-generated output. A member of the Aurora team will manually re-evaluate the decision.

Review Schedule

Quarterly Content Review

All output categories (health scores, predictions, summaries) evaluated for accuracy and bias every three months.

Annual Bias Audit

Comprehensive evaluation of potential fairness issues across demographic and business segments.

Changelog & Transparency

All policy updates, model improvements, and bias assessments logged and publicly disclosed.

Last Updated

April 2026

Next Review

July 2026

For more details on our AI systems and oversight, see our Model Card.