Most of the time, you don't just want text back from an LLM—you want that text in a structured format. Imagine generating a story about an imaginary planet, then extracting facts from it as a clean array of strings.
The AI SDK makes this incredibly easy. Instead of asking the model for text and then parsing it yourself (or validating it with a library like Zod), you can use structured output support built right into the framework.
There are actually two ways to do this in the AI SDK. One approach is more aligned with version 5, the other with version 6—but both work in version 6. Your task is to implement the structured output generation.
streamText()Start by streaming a paragraph of a story about an imaginary planet to the terminal.
generateText() with structured output supportAfter the text finishes streaming, call generateText() with a prompt asking for facts about the planet. Pass in the finalText from the previous step.
You'll need to import Output from the AI SDK alongside generateText.
Output.object() method to define structured outputPass an object with a schema property. The schema should be a Zod object with a facts field containing an array of strings.
Remember to use .describe() on your fields to guide the model toward better outputs.
Access the output from your result object and log it to the terminal.
Execute your code with pnpm run dev and verify that:
Once you've completed the exercise, check if you can implement the version 6 solution using generateObject() instead of generateText() with Output.object(). Both approaches work, but one is the newer pattern.
console.log(factsResult.output);
const stream = streamText({model,prompt:'Give me the first paragraph of a story about an imaginary planet.',});for await (const chunk of stream.textStream) {process.stdout.write(chunk);}const finalText = await stream.text;
// TODO: Replace this with a call to generateText, passing:// - The model, same as above// - The prompt, asking for facts about the imaginary planet,// passing in the finalText as the story// - The output, which should be Output.object({}), passing// the schema: z.object({// facts: z.array(z.string()).describe('The facts about the imaginary planet. Write as if you are a scientist.'),// })const factsResult = TODO;
output: Output.object({schema: z.object({facts: z.array(z.string()).describe('The facts about the imaginary planet. Write as if you are a scientist.',),}),});