Ask for a schema, not prose

Pass Output.object({ schema }) instead of parsing the model's text.

AI SDK docsai-sdk.dev

The schema is also used to validate the generated data, ensuring type safety and correctness.

AI SDK Core · Structured DataRead docs ↗

From ai-sdk and the Generating Structured Data docs.

Three invoices, three replies

Press Extract 3 invoices and see what gets saved.

JSON.parse(text)Bad

Saved right 0Saved wrong 0Crashed 0

const invoice = JSON.parse(text)

Output.objectGood

Saved right 0Saved wrong 0Caught 0

output: Output.object({ schema: invoice })

import { generateText, Output, NoObjectGeneratedError } from 'ai';
import { z } from 'zod';

const invoice = z.object({
  vendor: z.string(),
  total: z.number().describe('Grand total in USD'),
  dueDate: z.string().describe('ISO date, YYYY-MM-DD'),
});

try {
  const { output } = await generateText({
    model: 'openai/gpt-5.5',
    output: Output.object({ schema: invoice }),
    prompt: `Extract the invoice fields:\n${pdfText}`,
  });
  await saveInvoice(output); // typed { vendor, total, dueDate }
} catch (error) {
  if (NoObjectGeneratedError.isInstance(error)) flagForReview(error.text);
  else throw error;
}

Why it matters

Prose is loosely shaped. The schema tells the model the shape and checks the reply before your code sees it.

Pros

output is typed from the schema.
Bad replies throw one named error, not a bad row.
.describe() guides each field.

Cons

Valid shape is not correct content; still check values.
Strict schemas can be slower or unsupported on some models.
With tools, the output counts as a step in stopWhen.

Three invoices, three replies

Why it matters, pros and cons

Why it matters

Pros

Cons