AI_CLASSIFY

There is no hard limit, but accuracy tends to decrease with too many categories. Up to 15-20 categories works well. Beyond that, consider creating a hierarchical classification: first classify into broad groups, then sub-classify each group with a second AI_CLASSIFY call.

Yes, descriptive category names produce better results. "Billing Issue" works better than "Type A". The AI uses the semantic meaning of category names to understand what belongs in each. Use clear, human-readable labels.

Yes, AI_CLASSIFY works with text in any language supported by the underlying AI model. You can classify Spanish text into English categories, or use categories in the same language as the input text. Both approaches work well.

AI_CLASSIFY uses predefined category labels — you tell it the categories upfront. AI_FILL learns categories from examples — you show it what category each example belongs to. Use AI_CLASSIFY when you know your categories and have clear labels. Use AI_FILL when the categorization pattern is complex and better demonstrated through examples.

AI_CLASSIFY returns only the category label for simplicity. If you need confidence scores, use UNLIMITED_AI with a prompt like "Classify this text and include a confidence percentage: [text]. Categories: [list]".

The function will always return one of the provided categories, choosing the closest match. To handle edge cases, add an "Other" or "Uncategorized" option to your category list. This gives the AI a safe fallback for text that does not clearly match any specific category.

Even short texts (a single sentence) can be classified accurately when the meaning is clear. However, very short or ambiguous texts (1-2 words) may produce inconsistent results. For best accuracy, provide at least a full sentence of context.