Blog/AI

Draft LinkedIn and X Posts With ChatGPT or Claude (Not Generic)

AI assistants can draft posts fast, but the default output sounds like everyone else. Here is how to get drafts that sound like you, and how to send them straight to a review link.

PThe PostFounder team· 5 min read· Nov 17, 2026

ChatGPT and Claude can turn a messy note into a clean post in seconds. The catch is that, left to their defaults, they tend to produce the same polished, slightly empty voice everyone has learned to scroll past.

The problem is rarely the model. It is the input. If you ask for "a LinkedIn post about our new feature," you get the average of every post about every new feature. If you give it your actual material and clear constraints, you get something much closer to what you would have written yourself, only faster.

This guide covers how to prompt for drafts that sound like you, how to edit them, and how to connect an assistant to a scheduler so drafts land somewhere you can review before anything is posted.

Why AI drafts sound generic

Three habits cause most of it:

  • No raw material. The model fills the gap with stock phrases and safe generalities.
  • No voice reference. Without examples, it falls back to a default tone that is cheerful, balanced, and bland.
  • No constraints. It adds emoji, hashtags, bolded headers, and a closing question because those patterns are common in social posts.

Each of these is fixable in the prompt.

Give it real material

The single biggest improvement is pasting in the facts. Before asking for a draft, collect:

  • What actually happened (a changelog entry, a customer email with names removed, your notes from a call)
  • One specific detail: a number you are comfortable sharing, a quote, a screenshot description
  • Why it matters to you, in one or two rough sentences
  • Who the post is for

Then ask the assistant to work only from that material. A line like "Do not add facts, numbers, or claims that are not in my notes" goes a long way. It also protects you from the model inventing a statistic that sounds plausible.

Give it your voice

Paste two or three posts you have written and liked, and say what you like about them. For example:

Here are three of my past LinkedIn posts. Match their tone: short sentences, first person, plain words, no exclamation marks, no emoji. Notice that I usually open with a specific moment rather than a general statement.

If you do not have past posts, describe your voice in concrete terms. "Direct, a bit dry, explains things like a senior engineer talking to a peer" is more useful than "professional but friendly."

Give it constraints

Spell out the format rules for each network:

  • LinkedIn: up to 3,000 characters per post, per LinkedIn's help center. Short paragraphs. A first line that works on its own, since most readers only see the opening before "see more."
  • X: 280 characters per standard post, or a thread of posts that each stand alone.

And add your personal bans. Common ones:

  • No "excited to announce," "game-changer," or "let's dive in"
  • No hashtags unless I ask
  • No closing question unless it is a real one I want answered
  • No more than one emoji, or none

A prompt you can reuse

Putting it together:

I'm a founder building in public. Below are my notes from this week, and two past posts that show my voice.

Write one LinkedIn post (under 1,500 characters) and an X thread of 3 to 5 posts, each under 280 characters.

Rules: use only facts from my notes. Open with the most specific detail. Short paragraphs. No hashtags, no emoji, no "excited to share." End the LinkedIn post with a plain statement, not a question.

Notes: [paste] Past posts: [paste]

Ask for two or three versions of the opening line. Picking the best of three is easier than fixing one.

Edit the draft like it is someone else's

Treat the output as a first draft from a capable intern. A few passes catch most problems:

  1. Check every fact against your notes. Delete anything you did not supply.
  2. Cut the first sentence if it is a warm-up. The real opening is often the second line.
  3. Remove filler transitions like "Here's the thing" or "But here's what surprised me."
  4. Read it out loud. If you would not say a sentence to a peer, rewrite it.
  5. Add one detail only you know. That is the fastest way to make it yours.

Sending drafts to a scheduler with MCP

Copying text out of a chat window, pasting it into a scheduler, and then trimming it for X is where a lot of time goes. The Model Context Protocol (MCP) lets an assistant call tools in other apps, so it can hand a draft over directly.

PostFounder's MCP server does this without asking you to sign in, because it cannot post or touch your accounts. It has two tools:

  • draft_post fits your text to LinkedIn's 3,000-character limit and X's 280, splits the X version into a thread when needed, and returns a link. You open the link, review the draft, and schedule it yourself while signed in.
  • check_post is read-only. It counts characters the way each network does and suggests where to split a thread. It is useful when you just want to ask, "Is this short enough for X?"

The assistant never posts on its own. The draft only goes out after you open the link, approve it, and put it on your calendar.

Connecting it

The server URL is https://postfounder.com/mcp.

Claude: Open the Connectors section of your settings, choose to add a custom connector, give it a name, and paste the URL. Anthropic's guide to custom connectors covers the current steps and notes that custom connectors are available on all plans, with Free users limited to one. Claude asks for your approval before it calls a connector's tools.

ChatGPT: In Settings, open Apps & Connectors, go to Advanced and turn on developer mode, then create a connector, paste the URL, and choose no authentication. Menu names change, so check OpenAI's help center if you do not see these options.

Cursor and other clients: Add the URL to your mcp.json:

{
  "mcpServers": {
    "postfounder": { "url": "https://postfounder.com/mcp" }
  }
}

Prompts that work well

Once connected, you can ask things like:

  • "Turn this blog post into an X thread and a LinkedIn post for tomorrow at 9am."
  • "Write three LinkedIn posts announcing our new feature."
  • "Is this short enough for X?"

All the voice and constraint advice above still applies. The connector handles length and formatting; your notes and examples handle the substance.

What to keep for yourself

AI is good at the mechanical parts: structuring, trimming, splitting into a thread, offering alternate openings. Keep the parts that make building in public worth reading: deciding what is worth sharing, the specific details, and the opinions you actually hold.

A useful rule is that the assistant can write the sentences, but you choose the story.

Try it on your next post

Pick one thing that happened this week. Paste your notes, two past posts, and the constraints above into ChatGPT or Claude. Edit the result, then send it to a review link instead of copying it around. If you want the review, calendar, and scheduling side handled, PostFounder has a free plan.

Questions

Can the PostFounder MCP server post to my accounts?+

No. It needs no sign-in and cannot post. draft_post returns a link, and nothing is published until you open it, review the draft, and schedule it yourself.

How do I stop AI drafts from inventing facts?+

Paste your real notes and tell the assistant to use only facts from them. Then check every claim in the draft against your notes before posting.

Which assistants can use the PostFounder MCP server?+

Any client that supports remote MCP servers, including ChatGPT (with developer mode), Claude, and Cursor. Add https://postfounder.com/mcp as a connector.

Sources

Schedule LinkedIn and X posts together.

Write once, fit it to each network, and plan the week on one calendar. Free plan, or Pro for $4.99 a month.

Get started for free