De-identify a dataset before sharing it
A plan to remove or mask personal data in a dataset while keeping it useful for analysis, including re-identification risk checks.
The prompt
Help me de-identify a dataset so it can be shared for analysis without exposing personal information.
Dataset description and columns: [PASTE column names, types, and a few masked example values]
Who will receive it and for what: [e.g. "external vendor building a churn model", "internal team for a workshop"]
What analysis must still be possible: [e.g. "join across tables by customer", "trends by region and age band"]
Rules that may apply: [e.g. GDPR, HIPAA, CCPA, internal data policy - or "unsure"]
Steps:
1. Classify each column: direct identifier (name, email, phone, ID numbers), quasi-identifier (date of birth, ZIP/postal code, gender, job title, rare attributes), sensitive attribute (health, finances), or safe.
2. Recommend a treatment per column: drop, pseudonymize with a keyed hash or random token (and keep the key separate), generalize (age bands, 3-digit ZIP, month instead of date), suppress rare values, add noise, or keep.
3. Check re-identification risk from combinations of quasi-identifiers: suggest a k-anonymity threshold and how to find small groups.
4. Check free-text fields and file names for hidden personal data.
5. Explain the trade-off: what analysis is lost by each treatment.
6. List controls for the receiving party: access limits, no attempts to re-identify, deletion date.
Return a column treatment table (Column | Category | Treatment | Analysis impact), code outline in [SQL / Python] to apply it, a re-identification checklist, and a sign-off checklist. This is general guidance, not legal advice: confirm requirements with my privacy or legal team before sharing.
Highlighted [brackets] are blanks to replace with your own details.
Fill in the blanks and use this promptMore Data & Analysis prompts
What to fill in
- PASTE column names, types, and a few masked example values
- e.g. GDPR, HIPAA, CCPA, internal data policy - or "unsure"
- SQL / Python
How this prompt is built
This prompt follows the Few-Shot structure (Few-shot prompting). A few-shot prompt includes one or more worked examples of the input and the output you want, so the model can copy the pattern instead of guessing it.
Related prompts
- Analyze my clinic's claim denials and revenue cycle KPIsTurns denial and A/R data into root causes, benchmark gaps, and a prioritized fix list for the front end, coding, and billing teams.
- Trace data lineage across SQL scripts and stored proceduresMap where each table's data comes from and goes to, so you can see what breaks before you change a table.
- Explain, debug, or speed up a SQL, Python, or DAX snippetExplains what an existing query or script does, fixes errors, and optimizes it.
- Find what my dashboard is not telling meReviews a dashboard for blind spots, missing comparisons, vanity metrics, and misleading views, and suggests what to add.