Guardrails for data analysts

When you finish, you can use AI with data without exposing people's information or passing along wrong numbers.

What never to paste

Datasets often contain personal information: names, emails, addresses, account numbers, health details, salaries. Do not paste real rows into a consumer AI tool unless your employer has approved that tool for that kind of data. Privacy laws in many places, such as GDPR in Europe and HIPAA for health data in the US, set rules on how personal data may be handled.

You usually do not need real data to get help. Describe the structure: column names, types and a few invented rows. Replace identifiers with fake ones. If you are unsure whether a tool is approved, ask your data or security team before you paste anything.

Wrong answers that look right

An AI-written query can run without error and still answer the wrong question. A formula can return a number that looks fine and is off by a filter. Tools can also state a figure with total confidence that appears nowhere in your data. The danger is not the obvious failure but the believable one.

Build habits that catch it: test on a small slice, reconcile with a total you already trust, and spot check individual rows. Keep the query or steps saved so another analyst can repeat your work. If you cannot explain how a number was produced, do not present it.

Bias and missing people

Data reflects how it was collected. A survey sent only by email leaves out people without email. A model trained on past hiring may repeat past unfairness. AI summaries tend to smooth over these issues. Ask yourself who is missing from the data and what each metric rewards.

When analysis affects people, such as who gets a loan review, a job interview or extra attention, keep a human responsible for the decision, and check results across groups where you are allowed to do so. Say plainly what the data cannot show.

Disclosure and saying no

Follow your employer's rules about AI use, and any rules from clients or regulators. Many organizations ask you to note when a tool helped with an analysis. A simple line works: “Query drafted with AI assistance and verified against the source system.”

Learn to say no clearly. If someone asks you to find numbers that support a conclusion already chosen, or to share a file with personal data, you can reply, “I can do that safely if we remove identifiers, or I can show what the data actually says.” Protecting accuracy protects your name.

Practice now

Take a dataset you use and write a safe version of it for AI help: keep the column names, replace real values with invented ones, and note which columns are sensitive.
Write a one-line disclosure note you could attach to a report where AI helped draft the query.

Use any AI assistant you have. Do not paste private, client or patient information into it.

Check yourself

1. How can you get AI help with a table that holds customer details?

2. Why are believable wrong answers more dangerous than error messages?

3. A manager wants numbers that prove a decision already made. What is a good response?

Ask the AI tutor

Ask anything about this lesson, in any language. The tutor is an AI: it can be wrong, and it never needs your personal details.

Next lesson: Grow with AI: your 30-day plan

AI-drafted. Checked by an automated validator and a separate AI reviewer; a person has not read every lesson. Translations and tutor answers are AI-made. Found a mistake? Tell us. AI can be wrong. General information only, not professional advice. Not affiliated with any AI company named. Current University is not accredited.