What you can do with your network inside Claude and ChatGPT

The thing an assistant does not have

Ask Claude or ChatGPT who you know at a particular company and the honest answer is nothing. It has never seen your inbox. It does not know you had coffee with a fintech CFO in March, that a founder you passed on two years ago has been quiet since, or that you promised someone an introduction last Tuesday. An assistant that has read most of the internet has read none of your relationships.

That gap is the reason "AI for networking" has mostly meant "AI that writes a LinkedIn message", which is the least valuable part of the job. The valuable part is the memory: who you know, how well, what you last discussed, and what you owe them. Give the assistant that memory and the questions it can answer change completely.

The way to give it that memory is a connector: a small service the assistant can call in the middle of a conversation to search and act on your data. Anthropic calls them connectors, OpenAI calls them apps, and both speak the same open standard, MCP. You add one in settings, sign in once, and from then on any chat can ask about your network. What follows is what that looks like in practice, prompt by prompt, using the connector we build. Every prompt below is one it answers today.

Find who you know, in plain English

The first thing changes the moment it works. Instead of a search box that wants a name, you get a question.

Who do I know who has taken a company public, ideally in healthcare?

The assistant searches your contacts, your email history and your LinkedIn connections, and comes back with people and reasons: a CFO you met six times whose company listed in 2024, a board member who introduced you to two founders, a diagnostics CFO you sat on a panel with and have not spoken to since. The reasons are the point. A list of names you have to re-research is a search engine. A list of names with the history attached is a colleague.

The same prompt works for anything the history can answer. Who do I know at a Series A fintech in New York. Who have I met who has hired a first sales leader. Which of my contacts moved to Europe this year. Who introduced me to the most founders last quarter.

Draft the introduction, in your voice

This is the use case that pays for the rest. The pillar on warm introductions explains why a good one takes twenty minutes and three emails: ask both sides first, write the reason in a paragraph, send from your own address, follow up if nothing happens. An assistant with a connector does the drafting for all three emails from your own past messages, so the ask sounds like you and not like a template.

Draft a double opt-in intro between Marcus at Halden Robotics and Priya. He wants an hour with someone who has run an IPO from the finance seat.

You get the consent ask to Priya, the introduction that goes out only once she says yes, and the follow-up for four days later if she is quiet. You edit whichever you like and send; the emails leave from your Gmail, and the replies come back to you. The assistant asks before anything is sent. Nobody is copied in before they have agreed.

Know who is going quiet before they do

Human memory is good at faces and bad at dates. The relationship you would most want to protect is usually the one you last touched fourteen months ago without noticing.

Which of my warm relationships are going quiet?

The connector holds a model of how each relationship fades, built from your actual email and meeting history, so the answer is a short list with a reason attached: warm in 2024, three meetings and twenty-two emails, nothing since; he introduced you to two of your clients; you said you would follow up after her raise and did not. Ask for the reconnect and it drafts one for each, in your words, with the specific reason it is not awkward.

The guide to reconnecting after a long silence is the human version of the same routine. The assistant's job is to make sure the list exists.

Find the introductions your clients are asking for

If you advise companies, the asks arrive in investor updates, meeting notes and the last paragraph of a founder's email. "We are looking for a head of sales who has sold into hospitals." "Does anyone know someone at Stripe's partnerships team." They are easy to read and easy to forget.

Read Halden's last investor update and tell me which asks I can help with.

The assistant reads the update, pulls out the asks, searches your network for each, and comes back with the pairs it could introduce and how warm each path is. What used to be a mental note that expired by Friday becomes a list you can act on in ten minutes. The guide to what to track for each company you advise covers what to keep on hand so this works.

See who changed jobs, with the congratulations written

A job change is the single best excuse to write to someone, and it goes stale in about a week.

Who in my network changed jobs since my last LinkedIn refresh?

You get the moves, the old role and the new one, whether the relationship is warm, and a draft congratulations for each. What to say when someone changes jobs is the longer answer; the short one is that the note should be specific and early, and an assistant that already knows the history can make it both.

What it will not do, and should not

Three limits worth knowing before you expect too much.

It will not send without asking. Every send and every edit is confirmed in the chat. Read-only lookups run on their own. That is a design choice, and you should not use a connector that makes a different one.

It only knows what it has been connected to. Gmail, Google Calendar and LinkedIn today, in our case. A relationship that lives entirely in WhatsApp is invisible to it, the same way it is invisible to any tool that reads your inbox.

It does not decide who to introduce. It finds the candidates and drafts the words. Whether the introduction should happen at all is the judgment the guide to when not to make an introduction is about, and it stays yours.

How to get there

You need two things: an assistant, and a connector that holds your network.

If you want to wire it up yourself, Zapier's MCP server exposes thousands of app actions to Claude and ChatGPT, and it is the fastest way to give an assistant a Google Sheet or a HubSpot list. Clay's Claygent is the strongest option for enriching a list of people with public data. HubSpot's own assistant works well inside HubSpot. All three assume the network already lives in a structured tool somewhere; none of them reads your inbox to find out who you actually know.

If you want it to work from your real relationships, NetworkOS is the connector we build for this. It reads Gmail, Calendar and LinkedIn to learn who you know and how well, answers the questions above, drafts the double opt-in in your voice, sends from your own Gmail and watches the replies, and flags the relationships going quiet. It is a custom connector in Claude and an app in ChatGPT, and the same network is behind both. It is Google-only today, and it has no shared team directory the way a fund tool does.

Either way, start with one real question you have this week rather than a demo prompt. "Who do I know who could talk to Marcus about hiring a first sales leader" will tell you within a minute whether the memory is there.

The short version

Claude and ChatGPT know nothing about who you know until something connects them to your history. Connected, they can find who you know in plain English, draft the introduction in your voice, tell you who is going quiet, read a founder's update and find the introductions in it, and spot who changed jobs. The prompt is always the question you already had. The connector is the memory that lets the assistant answer it.