AI Filter Function

The AI_FILTER function evaluates a natural-language predicate or yes or no question and returns a boolean result.

Note:

For information about limitations with using AI Functions, see Known Issues with AI Functions.

The function returns TRUE only when the model responds with a “yes”. Any other response is treated as FALSE. There is no need to specify the model to answer yes or no. The function handles the output format and you only needs to specify the predicate or question. Because it returns a boolean value, you can use the AI_FILTER function anywhere a boolean expression is valid, such as:

• WHERE clause
• JOIN conditions
• CASE expressions

Use this function for:

• Sentiment-based filtering
• Intent or urgency detection
• Semantic matching
• Relevance evaluation

Where You Can Use the AI_FILTER Function

You can use this function in filter expression editors, for example:

• Workbook's filter expression editor
• Data Flow's filter step editor
• Data Preparation's filter expression editor
• Semantic Modeler's Data Filters editors

You can also use this function in join conditions, for example a semantic model's physical join expression.

You can only use the function as a CASE condition when you add new columns, transform existing columns, or edit column expressions. You can't use it as a column value by itself.

You can use this function in various editors, for example:

• Workbook's Create Calculation editor
• Data Preparation's Create Column editor
• Data Flow's Add Columns editor
• Data Flow's Transform Column editor
• Semantic Modeler's Create New Column editor

You can't use this function as a column value by itself when adding new columns, transforming existing columns, or editing column expressions. Some editors may allow you to save the AI_FILTER by itself, but results will encounter errors.

Syntax

AI_FILTER(<predicate>, <input>, <model_id>)

Parameters

Parameter	Description
predicate	String A yes or no question, or predicate for the model to evaluate. Must be a string literal or expression.
input	String Text data to evaluate against the predicate. Must be explicitly cast to string if derived from non-string columns.
model_id	String literal Identifier of the model asset to use.

Examples

This example filters rows that contain only negative reviews.

SELECT * 
FROM REVIEWS
WHERE AI_FILTER(
	'Is the review negative?',
	'Review: ' || REVIEW_TEXT,
	'model'.'id'
);

This example matches reviews with the products reviewed.

SELECT * 
FROM REVIEWS R
INNER JOIN PRODUCTS P
ON AI_FILTER(
	'Does the review describe the same product as the description?',
	'Review: ' || R.REVIEW_TEXT || '; Description: ' || P.DESCRIPTION,
	'model'.'id'
);

Best Practices

Keep these best practices in mind when you use the AI_FILTER function:

• Keep predicates and questions short, clear, and unambiguous.
• Phrase predicates as explicit yes or no questions.
• Avoid compound or multi-part questions that require nuanced answers.
• Provide descriptive labels for each input value to help the model understand what each piece of data represents.
• Ensure the prompt references the same labels used in the input.
• Use punctuation and separators to clearly distinguish between multiple input values.
• Cast non-string expressions / columns to string.
• Use consistent and predictable labels across queries to improve reliability.
• Keep input text focused because long inputs reduce classification accuracy.
• Remember that only "yes" evaluates to TRUE.