# How to use Claude or ChatGPT to write LinkedIn connection requests
> Stop sending generic invites. Give Claude or ChatGPT the lead's recent posts and your goal, then draft short LinkedIn connection requests people answer.
- HTML: https://getkinlead.com/connection-request-notes-for-mcp-clients
- Markdown: https://getkinlead.com/connection-request-notes-for-mcp-clients.md

**Short answer:** To use Claude or ChatGPT for LinkedIn connection requests, give the AI the lead's recent post or profile detail and your reason for contacting them. The fastest route is an MCP server connected to a lead source, so the AI can pull the detail and draft a short note without copy and paste.

## Why generic AI prompts fail

Most people use ChatGPT as a template generator. They ask for a professional invite for a job title, and the AI returns something formal and obviously written by a machine. The reader can tell you did not look at their profile. That is the main reason they ignore it.

A connection request needs one specific anchor. An anchor is a detail only someone who read the profile would know: a phrase from a recent post, a project in the About section, a comment they left on a shared connection's post. When the AI has that detail, it stops guessing and writes from facts.

The prompt matters less than the input. A clever prompt with no context produces a polished stereotype. A plain prompt with a real post and a clear reason produces something a person can answer.

The fix is not a longer prompt. It is a smaller, more specific input. Give the AI one real detail and one clear reason, then ask for a short note. You will still edit it, but you will start from something that could only be sent to that person.

**Generic prompt**

- Prompt: Write a professional LinkedIn invite for a VP of Sales.
- Result: I came across your impressive profile and would love to connect to explore possible synergies.
- Outcome: The reader sees a template and ignores it.

**Context-rich prompt**

- Prompt: Write an invite for a VP of Sales who posted last week about hiring SDRs in Chicago. My goal is to ask about her hiring plan.
- Result: Saw your post about hiring SDRs in Chicago. I work with sales leaders on the first 90 days of a new SDR team and had one question about your plan.
- Outcome: The note gives her a reason to accept and a topic for the reply.

## Gathering the right context before you prompt

AI gives you a stereotype if you only provide a job title. If you provide a recent post, you get a conversation starter. Collect the detail before you open the chat. This takes two minutes per lead manually, which is why most people skip it and get generic drafts.

Keep the context short. You do not need the whole profile. One post, one line from the About section and your reason for contacting them is enough. Too much context makes the AI summarize the profile back at the person, which reads like a research report.

If you are working from a list, put the details in a spreadsheet first. Columns for post URL, post topic, profile line and your goal make the prompt repeatable. The same structure also lets you check drafts later without reopening every profile.

Do not paste the whole profile. The AI will summarize it back at the person, and the note will read like a research report. One post, one line from the About section and your goal is enough. If you cannot find one specific detail, the lead is not ready for a custom note. Use a simpler message or skip them.

| | Lead detail | Why it helps | Example |
| --- | --- | --- | --- |
| Recent post | Gives the AI a real topic to mention | Their post on moving from founder-led sales to a sales team |
| About section | Shows how they describe their work | A line about building outbound for a 20-person agency |
| Comment or reaction | Shows what they care about now | A comment on a competitor's post about pricing |
| Your goal | Keeps the note from becoming a pitch | Ask whether they are reviewing their SDR process |
| Shared ground | Gives a natural opener | Both worked at a company with a long sales cycle |

## The manual drafting process

If you draft manually, use AI for the first touch, not the whole sequence. The job of a connection request is to get the person to accept. The job of the message after that is to start a conversation. Do not mix them.

Set constraints that stop the AI from using its corporate voice. Tell it to write like a person sending a short note, not like a company announcing a partnership. Ask for one sentence of context and one sentence of reason. If the draft includes a compliment with no detail, delete it.

Use the AI as an editor after the first draft. Ask it to cut the note to one reason and one question. Ask it to remove any sentence that could apply to another lead. That second pass is where most of the quality comes from.

1. **Feed the profile detail.** Paste the lead's About section and one recent post. Add the reason you are contacting them in plain words.
2. **Set negative constraints.** Tell the AI to avoid corporate words, long compliments and phrases like 'I came across your profile'. Ask for short sentences.
3. **Enforce the note length.** The connection note field is short. Keep the draft to one or two sentences and check the character count in LinkedIn before you send.
4. **Request variations.** Ask for three versions: one direct, one curious and one based on the post. Pick the one that sounds like you, then edit it by hand.

> Never send the first AI draft. Read it out loud. If it sounds like a brochure, cut the adjectives and add the specific detail.

## Automating data retrieval with MCP

Copying and pasting profiles is slow. The Model Context Protocol (MCP) lets AI clients like Claude Desktop, ChatGPT clients and Cursor connect to external data sources. Instead of you bringing data to the AI, the AI asks the server for what it needs.

That turns the AI into an agent that can find a lead, read the recent post and draft a note in one motion. You still need to define the lead profile and the reason for outreach. The server does not know your offer.

Use this when you have a repeatable search, not for five random people. A good use is 'find heads of sales at logistics companies in the Midwest who posted about hiring this quarter, then draft a note based on the post.' A bad use is 'find me leads' with no profile.

Keep a human check in the loop. An MCP query can return 20 notes that look fine at a glance and three that mention the wrong company. Read the batch before it goes into a campaign. The cost of a bad note is higher than the time it takes to scan.

1. **Install an MCP client.** Set up a client that supports MCP, such as Claude Desktop, Cursor or a ChatGPT client with MCP support.
2. **Configure the lead source.** Add the server details to the client's configuration file. Use the provider's instructions for the exact fields.
3. **Run a narrow query.** Ask for a specific role, industry and recent activity. Broad queries return broad drafts.
4. **Check the output.** Read every note before it goes into a campaign. Remove any line that could apply to anyone.

```
Find 20 heads of sales at logistics companies in the Midwest. For each person, use their most recent post about hiring or team design. Draft a LinkedIn connection note with one sentence about the post and one sentence about why I am reaching out. Keep each note to two short sentences. Do not use compliments without a detail. Do not mention my product in the note.
```

## What to do when the AI draft sounds like a brochure

The first draft often sounds like a brochure because the AI defaults to praise. It tells the person they have an impressive background and asks to connect. That is not a reason to accept. Replace praise with a detail and a question.

A good test is to swap the person's name with another lead. If the note still works, it is too generic. The note should break when you change the post or the company.

If the reply is short, do not force a long answer. Match the length. A one-line question gets a one-line reply, and that is fine. The goal is to keep the conversation moving, not to show how much you know.

- **draft:** Hi Dana, I saw your post about moving from founder-led sales to a sales team. I work with founders on that transition and had one question about your hiring plan.
- **them:** Thanks. We are hiring two AEs this quarter. What is the question?
- **you:** Are you keeping founder-led outbound for enterprise, or moving it to the new AEs first? I ask because the handoff usually decides the first quarter.
- **them:** Good question. Founder-led for enterprise, AEs on mid-market. That is the plan for now.

> The note did not pitch. It named the post and asked about a decision. The reply gave a real detail. That is the pattern to copy.

## Doing this with Kinlead

Kinlead provides an MCP server and REST Agent API, so Claude, ChatGPT, Cursor and other MCP clients can find leads, check replies and draft answers. The [Agent API and MCP](/features/agent-api-and-mcp) page covers connection details.

The Steal Customers agent watches competitor company pages and optional founder or employee profiles, reads comments on their recent posts, and turns commenters who look like buyers into leads. Each lead keeps the post URL, what the post was about and what they commented. Connection requests can carry an AI-written note under 200 characters that opens with why that person became a lead. Queued campaign messages can be read and edited on the Actions page before they send, and each campaign sends inside limits and a send window you set.

## Related

- [AI SDR playbook for cold leads](https://getkinlead.com/ai-sdr-playbook-for-cold-leads.md)
- [AI outreach strategy for niche industries](https://getkinlead.com/ai-outreach-strategy-for-niche-industries.md)
- [AI SDR follow-up cadence playbook](https://getkinlead.com/ai-sdr-follow-up-cadence-playbook.md)

## Common questions

**Can I use ChatGPT for this without an API?**

Yes. You can paste the lead's profile and one recent post into the chat and ask for a short note. It is slower, but it works for a small list.

**How long should a LinkedIn connection request be?**

Keep it to one or two sentences. The note field is short, and the reader needs one reason to accept, not a summary of your offer.

**Will using AI to write invites get my account restricted?**

AI writing is not the main risk. LinkedIn can restrict accounts for many reasons, including sending too fast or too many pending invites. Keep outreach inside whatever limits your account shows and use a send window.

**What is the best way to personalize invites at scale?**

Use a lead source that keeps the post URL, the post topic and the comment or profile detail with each lead. Then the AI can draft from a real event instead of a job title.

[Create a Kinlead account](https://getkinlead.com/signup) or [book a call](https://getkinlead.com/demo) to talk through pricing and setup.
