CLAIR AI Output Disclaimer

Effective October 10, 2026 · Version 1.0

This disclaimer is part of the CLAIR Terms of Use.

1. How CLAIR produces answers

CLAIR uses open-weight AI language models, made by third parties and running on your computer, to interpret your questions, write database queries, summarize data and describe results. It also uses statistical and machine-learning methods to test hypotheses, find patterns and make predictions.

2. AI output can be wrong

AI models can misunderstand a question, choose the wrong columns or filters, write a query that runs but answers a different question, leave out relevant data, invent facts that are not in your data, or describe correct numbers incorrectly. How reliable a model is depends on the model, the task, your data and your computer, and the models that fit on smaller computers are generally less capable.

Text inside your data can also influence the AI. If a file contains written instructions (for example, in a comments column), the model may follow them instead of your question. Be especially careful with files from sources you do not trust.

3. What CLAIR’s answer labels mean

CLAIR labels answers to help you judge them.

When CLAIR shows that an answer was calculated or verified from your data, it means the number came from a database query CLAIR ran on your data, and CLAIR’s automatic checks found no sign that the query left out part of your question (for example, a filter or grouping you asked for). Those checks catch common mistakes. They cannot prove the query matches what you meant. The query is usually written by the AI. Open it and review it.
When CLAIR labels an answer as estimated, it comes from a sample of your data or from a statistical estimate. It is not exact.
When the AI writes a summary or explanation, the wording comes from the model, based on data CLAIR retrieved. It has not been checked against all of your data.
When CLAIR says it could not confirm an answer, it may still be right or wrong. Treat it with extra care.

4. Correct calculations on the wrong data

A calculation can be done correctly and still mislead. Check the import preview: a column read as the wrong type, rows skipped during import, codes such as 99 meaning “missing”, or duplicated records will all change results. CLAIR warns about some of these problems, but not all of them.

5. Statistics and predictions

Statistical tests and machine-learning models depend on the quality, size and representativeness of your data, and on assumptions that may not hold. A statistically significant result is not proof of cause and effect, and running many tests makes some significant results appear by chance. Predictions and forecasts are estimates and can be badly wrong, especially for situations unlike the data they learned from, and a model’s accuracy on your data may not hold for new data. Features that rank “what drives” an outcome show association, not causation. Columns such as zip code or school can stand in for protected characteristics like race. Treat findings as a starting point to check, not a conclusion.

6. Cleaning and changes to data

Some features propose changes to your data, such as filling missing values or removing duplicates. Review each proposal before you apply it. Many changes can be undone in CLAIR, but not every operation can, so keep your own copy of your original files.

7. Not professional advice

CLAIR does not provide medical, clinical, legal, financial, investment, tax, statistical consulting, academic or other professional advice. Its outputs are not a substitute for a qualified professional. The Acceptable Use Policy limits using CLAIR for decisions about people.

8. Research and publication

If you use CLAIR in research, you are responsible for the validity of your methods and results, for meeting the requirements of your ethics or IRB approval, and for disclosing AI use as your institution, funder or publisher requires. Exports that contain AI-written text name the model used, to help you disclose it, but that note is not a full record of your methods. Keep your own record of the data version, the queries and the settings you used. Do not suggest that CLAIR or its developer reviewed or endorsed your results.

9. No guarantee

We do not guarantee that any output is accurate, complete or fit for your purpose. See Sections 18 to 20 of the Terms of Use.

In-app notices

These short notices appear in the app. Text in curly braces is filled in by the app.

Sign-in screen, shown with the agreement checkbox:

CLAIR uses AI that runs on this computer. It can misread your question, write the wrong calculation, or describe results incorrectly. It is a tool to help you explore data, not professional advice. Check results before you use them.

Chat, below the message box:

CLAIR’s AI can make mistakes, even when an answer is calculated from your data. Check the rows and calculation behind an answer before you rely on it.

Prediction and forecast screens, and exported predictions or forecasts:

Predictions are estimates based on patterns in your data. They can be wrong, especially for new situations, small datasets or data that has changed.

Footer of reports, slide decks and PDFs that contain AI-written text:

Contains AI-generated analysis produced by CLAIR on {date} using {model name}. Check figures against the source data before relying on them.