Writing blog posts is hard work. Furthermore, writing threads to promote those articles is even harder. Its because a completely different writing style from blogging (punchy hooks, brevity) is required.

Therefore, I decided to automate this workflow using OpenAI (GPT-4o) and Twitter API . Here, I introduce how I built a script that analyzes my content and generates perfectly formatted threads.

The Problem: Just Post a Link and Pray Doesnt Work

Just posting a link to X (formerly Twitter) results in zero engagement. The algorithm gives the cold shoulder to posts with links. Its clear that threads that keep users on the platform perform better, but rewriting a 3, 000-character article into five tweets takes at least 30 minutes.

• Goal: Convert Markdown content → 5 impactful tweets.
• Tools: TypeScript, twitter-api-v2, Vercel AI SDK.

The Script: post-to-x-thread.ts

I created a CLI script that takes a file path as input.

# Usage looks like this
npx tsx scripts/utils/post-to-x-thread.ts --file src/content/blog/my-post.mdx --url https://...

The Viral Prompt

The secret sauce is in the prompt, not the code. I instructed the AI to follow specific constraints:

1. Tweet 1 (Hook): Include a paradoxical opinion or a specific promise. Reporting I wrote a post is a no-go.
2. Tweets 2-4 (Value): One idea per tweet.
3. Last Tweet (CTA): Link to the full article.

const prompt = \`
You are a professional social media manager.
Please convert the following blog post content into a high-engagement Twitter/X thread.
Constraints: Each tweet must fit within 280 characters (standard length).
Structure: Hook -> Value -> Value -> CTA
\`;

Automatic Image Processing

A text-only thread is boring. I updated the script to automatically fetch the heroImage from the articles frontmatter, upload it to X, and attach it to the first tweet.

// Auto-upload hero image
const mediaId = await client.v1.uploadMedia(heroPath);
await client.v2.tweet("text, { media: { media_ids: ["mediaId"] } });

Deep Dive: JSON Schema for Thread Generation

Allowing AI to output in a “free format” leads to parsing failures. The pro technique is to fix the structure using the Vercel AI SDK’s object mode.

const threadSchema = z.object({
  tweets: z.array(z.object({
    content: z.string().max(140),
    hashtags: z.array(z.string())
  })).length(5)
});

This ensures you always receive an “array containing exactly 5 tweets,” so the automated posting system never crashes.

Results

Now I can generate thread drafts in seconds. I added a --dryRun flag, so I can check the AI output before actually posting. 90% of the work is done; all that’s left is to refine the hooks and hit the send button.