genai-disclaimer

GenAI Disclaimer

Standard disclaimers and attribution notices for AI-generated or AI-assisted data analysis outputs.

When to Apply

Add a GenAI disclaimer to any output that:

• Was generated or substantially shaped by an AI/LLM
• Contains AI-interpreted statistics, summaries, or conclusions
• Includes AI-generated Data Dictionaries, Descriptions, or Tags (from describegpt)
• Was validated, profiled, or cleaned with AI-assisted workflow decisions
• Contains visualizations where chart type, aggregation, or framing was AI-selected

Standard Disclaimer Templates

Short (inline footnote)

*Analysis assisted by AI. Results should be independently verified before decision-making.*

Medium (report footer)

---
**AI Disclosure:** This analysis was generated with AI assistance using qsv data wrangling
tools and Claude. All statistical computations were performed by qsv (deterministic,
reproducible); interpretations, summaries, and recommendations were AI-generated and should
be reviewed by a domain expert before acting on them.

Full (formal deliverable)

---
## AI-Generated Content Disclosure

**Tools used:** qsv (data processing), Claude (analysis and interpretation)
**Date:** [YYYY-MM-DD]

This document contains AI-generated content. The following aspects were AI-assisted:
- [ ] Data profiling interpretation and quality assessment
- [ ] Statistical summary and narrative
- [ ] Data Dictionary, Description, and Tags
- [ ] Chart type selection and visualization design
- [ ] Recommendations and conclusions

**What is deterministic:** All row counts, statistics, frequency distributions, and data
transformations were computed by qsv — a deterministic, open-source tool. Given the same
input data and parameters, these results are exactly reproducible.

**What is AI-generated:** Narrative interpretations, quality assessments, chart design
decisions, and recommendations were generated by an LLM. These outputs may contain errors,
hallucinations, or biases inherent to the model. They should be reviewed by a qualified
analyst before use in decision-making.

**Reproducibility:** The qsv commands and parameters used are documented in this analysis.
Re-running them on the same input data will produce identical numerical results. AI-generated
narratives may vary between runs.

Choosing the Right Level

Output Type	Recommended Level
Quick Slack message or internal chat	Short
Internal report or dashboard annotation	Medium
Stakeholder presentation	Medium or Full
Regulatory or compliance deliverable	Full
Published research or external report	Full
Exploratory analysis (own use)	None needed

What to Attribute to AI vs. qsv

Component	Attribution	Reproducible?
Row counts, statistics, frequencies	qsv (deterministic)	Yes — exact same results every run
Data type inference, null counts	qsv (deterministic)	Yes
Data cleaning operations (dedup, trim, safenames)	qsv (deterministic)	Yes
SQL query results via sqlp	qsv/Polars (deterministic)	Yes
Data Dictionary, Description, Tags	AI-generated via describegpt	No — may vary between runs
Quality assessment narrative	AI-generated	No
Chart type recommendations	AI-generated	No
Conclusions and recommendations	AI-generated	No
Validation methodology review	AI-generated	No

Integration with Commands

When these commands produce output, consider appending the appropriate disclaimer:

• /data-profile → Medium or Full (contains AI interpretation of statistics)
• /data-describe → Full (describegpt output is entirely AI-generated)
• /data-validate → Medium or Full (methodology review is AI-generated)
• /data-viz → Medium (chart type selection and title framing are AI decisions)
• /data-clean → Short or None (operations are deterministic; only the choice of operations was AI-guided)
• /data-join → Short or None (join execution is deterministic)
• /csv-query → Short or None (SQL results are deterministic)
• /data-convert → None (purely mechanical format conversion)

Regulatory Considerations

Some jurisdictions and industries require explicit AI disclosure:

• EU AI Act: Transparency obligations for AI-generated content
• Financial services: Regulatory expectations around AI-assisted analysis and model risk
• Healthcare: HIPAA considerations when AI processes PHI; disclosure requirements for AI-assisted clinical decisions
• Government: Executive orders and agency policies on AI use in federal data analysis

When in doubt, use the Full disclaimer. Over-disclosure is always safer than under-disclosure.