Trusting AI-generated results¶

datasight uses AI to translate natural language into SQL. The query confidence features reduce the risk of errors, but AI-generated queries and their results should always be treated as unverified drafts. Keep the following in mind before relying on a result.

Queries may be incorrect¶

The AI can misinterpret your question, choose the wrong columns, apply incorrect filters, or use flawed aggregation logic — even when the results look plausible.

Results can be misleading¶

A query that runs without errors is not necessarily a correct query. Spot-check row counts, totals, and edge cases before drawing conclusions.

Verify before publishing¶

Never include AI-generated queries or results in reports, dashboards, or decision-making without independent verification by someone who understands the underlying data.

Different runs may produce different SQL¶

The same natural-language question can yield different queries across sessions, potentially with different results.

The AI has no domain expertise¶

It relies on column names and schema descriptions to infer meaning. If the schema is ambiguous or incomplete, the AI may make wrong assumptions about what the data represents.

What to do about it¶

• Enable SQL approval and explanations to review every query before it runs.
• For high-stakes analyses, use the query verification workflow to validate results against known expected outputs.
• Be specific when asking questions — specify temporal granularity, include counts, name the metric, use explicit date ranges. See Why clarifying questions appear.