GES-C02 Tests Which Cortex Feature Fits the Data — Not Generic LLM Theory
GES-C02 replaced GES-C01 in 2026 and swaps the old Document AI for AI_EXTRACT and AI_PARSE_DOCUMENT. Expect builder and admin scenarios where you pick the Cortex function, role or parameter that fits, inside Snowflake.
Check Your Readiness →Most candidates understand SnowPro Specialty: Gen AI (GES-C02) concepts — and still fail. This exam tests how you apply knowledge under pressure.
Weights come from secondary reproductions of the official guide and still need checking against the PDF. Items are short scenarios: identify per-row versus aggregate work, structured versus unstructured data, and whether Cortex hosts the model, then choose the most specific function or the least-privilege control.
Switch to a larger model with a longer context window
Cortex models are stateless between calls. The application must resend prior turns in the messages array, with roles and content, on every request; a bigger model changes nothing if the history is never passed.
Schedule a nightly task that parses every staged file
Use a stage with server-side encryption and a directory table, put a stream on the directory table, and run a task that calls AI_EXTRACT only on the stream's new rows, writing fields plus file path to a table.
Enable cross-region inference so requests route elsewhere
Cross-region inference is disabled by default and exists for model availability, but it sends data to other regions. Under a residency rule, choose a model offered in the local region and leave the account setting off.
Concatenate all rows into one string and call AI_COMPLETE
A single prompt would overflow the context window. AI_AGG with an instruction, or AI_SUMMARIZE_AGG without one, is built to reduce many rows to one insight without being bound by a single context window.
Cortex Search retrieves text passages with hybrid vector and keyword ranking; it cannot total revenue or count tickets. Numeric questions over governed tables belong to Cortex Analyst, which generates SQL from a semantic view or YAML semantic model, and mixed needs go to a Cortex Agent.
Granting the database role to one team changes nothing while PUBLIC still holds it. Revoke CORTEX_USER from PUBLIC, grant the narrowest role needed (CORTEX_EMBED_USER, CORTEX_ANALYST_USER or CORTEX_AGENT_USER), and restrict models with the CORTEX_MODELS_ALLOWLIST account parameter.
Sentiment, labels and yes/no conditions have dedicated functions: AI_SENTIMENT, AI_CLASSIFY and AI_FILTER return typed output at lower cost with nothing to parse. Fall back to AI_COMPLETE with a JSON-schema structured output only when no task-specific function produces the required result.
AI_PARSE_DOCUMENT in OCR mode returns plain text; LAYOUT mode preserves tables, headings and reading order, which also chunks better for RAG. When only a few named fields are needed, skip full parsing and call AI_EXTRACT with a response format.
Cortex Guard filters unsafe responses from AI_COMPLETE; AI_REDACT strips personal data from text you pass in. ACCOUNT_USAGE views report token and credit consumption, while answer quality such as groundedness comes from AI Observability evaluations and TruLens traces.
Strengthen weak areas with exam-style practice questions and detailed explanations.
Simulate the real exam experience and assess your readiness under timed conditions.
Review key concepts, objectives, and exam topics in one place.
Get personalized explanations, learning recommendations, and instant answers.
Follow a structured learning path designed to help you prepare efficiently.
GES-C02 rewards picking the exact Cortex function, role or mode. Find your gaps before exam day.