Generating Text and Structured Objects

The Vercel AI SDK provides two primary server-side functions for generating non-streaming content: generateText and generateObject.

1. Using generateText

Use generateText for simple question-and-answer exchanges or summary requests that do not require streaming:

// src/services/aiService.ts
import { generateText } from "ai";
import { defaultModel } from "../lib/models";

export async function askQuestion(prompt: string) {
  const response = await generateText({
    model: defaultModel,
    prompt: prompt,
    system: "You are a senior code reviewer. Give short and direct answers.",
  });

  console.log("Raw output text:", response.text);
  console.log("Total tokens consumed:", response.usage.totalTokens);
  
  return response.text;
}

2. Using generateObject with Zod

To guarantee that the model outputs structured JSON that matches your database schema, combine generateObject with the Zod library:

import { generateObject } from "ai";
import { z } from "zod";
import { defaultModel } from "../lib/models";

// Define output structure rules using Zod
const UserProfileSchema = z.object({
  fullName: z.string(),
  companyRole: z.string(),
  skills: z.array(z.string()),
});

export async function parseUserProfile(bioText: string) {
  const { object } = await generateObject({
    model: defaultModel,
    schema: UserProfileSchema,
    prompt: bioText,
  });

  // object is typed as: { fullName: string; companyRole: string; skills: string[] }
  console.log("Parsed skills count:", object.skills.length);
  return object;
}

3. Selecting the Right Function

• Use generateText for general copywriting, text translation, markdown summarization, or whenever conversational dialogue is the target.
• Use generateObject for parsing search intent, formatting user profile fields, generating charts datasets, or populating relational database columns.