How-to
How to use ChatGPT for LinkedIn prospecting (with a connector, not copy-paste)
7 days free. No card. Then US$79/mo · See pricing
By the Bingle team · Published 29 October 2026 · Researched from the sources listed below, drafted with AI help and checked against those sources
Short answer
How do you use ChatGPT for LinkedIn prospecting?
On its own ChatGPT only works on what you paste in, so add a connector that holds your LinkedIn leads, such as Bingle's, as a custom app through Developer mode. Then ask ChatGPT who to contact today and why, have it draft the first note from that person's signal, and approve it yourself: Bingle keeps sending off unless you allow it when you sign in, and even then it shows what goes to whom and waits for a yes. OpenAI's developer docs say account and workspace policies decide who can add a custom MCP server, so check your own ChatGPT settings first.
How we researched this
On 29 October 2026 we read OpenAI's developer docs on connecting a custom MCP server to ChatGPT and LinkedIn's help pages on invitations. OpenAI's help centre and ChatGPT's pricing page refused our reader with a 403 error that day, so we don't state ChatGPT plan details as checked. Bingle's steps come from its Claude and ChatGPT page and user guide; we didn't run this workflow in a live ChatGPT account for this post.
What ChatGPT can and can't do for prospecting on its own
ChatGPT can only work with what it can see. Without a connector that means what you paste into the chat and, where it searches the web, public pages. Your LinkedIn notifications, your pending invites and the list of people who liked a competitor's post yesterday all sit behind your LinkedIn login, so it has none of them.
| ChatGPT with copy-paste | ChatGPT with Bingle's connector |
|---|---|
| Know who is showing interest this week | |
| No. You'd have to find them first and paste them in | Yes. Bingle's sources watch LinkedIn for buying signals and save the people who fit |
| Rank who to contact first | |
| Only the handful you paste, with no timing data | Leads scored 0 to 100 on fit and recent signals, with the reason for each |
| Explain why one lead beats another | |
| It can guess | explain_lead_score gives the signals and dates behind each score |
| Write a first message | |
| Yes, from whatever you pasted | draft_message writes from the lead's own signal, in the voice you set in Bingle |
| Find more people like your best leads | |
| No | add_leads takes pasted profile links; update_icp changes who Bingle looks for, after you agree |
| Send on LinkedIn | |
| No | Only if you ticked sending at sign-in, and then only after you say yes to the exact message and recipient |
Copy-paste has a second problem besides scale: the data is stale the moment you paste it. A profile you copied last Monday doesn't show the post the person commented on this morning. Bingle's guide defines intent as something dated that a person or their company did, shown with the date and a link, and that date is what makes a message timely.
How apps and connectors work in ChatGPT
A connector is an MCP server. MCP, the Model Context Protocol, is the standard ChatGPT and Claude use to talk to other apps. The app publishes a list of tools (search my leads, explain a score, draft a message), ChatGPT reads that list, and when your prompt needs one it calls the tool and reads back the answer.
OpenAI's naming has moved around. Its developer documentation, read on 29 October 2026, now calls a custom MCP server you add a plugin, and its steps start from ChatGPT Plugins: press the plus button, choose Add custom MCP server, give it a name, paste the server URL, set up sign-in, read the risk warning, tick "I understand and want to continue" and create it. To use it, start a new chat, type @ and pick the plugin.
| What we could confirm on 29 Oct 2026 |
|---|
| Who can add a custom MCP server? |
| OpenAI's developer docs: "Account and workspace policies apply to adding and using custom MCP servers." No plan list on that page |
| Which plans? |
| OpenAI's help article on Developer mode lists them. When Bingle's team read it on 24 October 2026, it gave full support on Business, Enterprise and Edu, with an admin turning it on for Business |
| Where is the switch? |
| Bingle's steps say Settings, then Apps & Connectors, then Developer mode under Advanced settings. OpenAI's developer docs start from ChatGPT Plugins instead |
| Do write actions ask first? |
| OpenAI's docs mention write actions that need authorisation or confirmation, but don't show what the prompt looks like. Bingle asks for a yes before any send whatever ChatGPT does |
The quickest check is your own screen. Open Settings in ChatGPT and look for Developer mode. If you're on a company workspace and can't find it, ask your admin before you spend time on the rest of this guide.
Step by step: connect Bingle to ChatGPT
You need a Bingle account first, because the connector works on your own workspace: your leads, your ideal customer profile and your sequences. The first 7 days are free with no card, and the connector is part of every plan, Starter at US$49 a month included.
- In Bingle, open Settings, then General, and copy the connector link.
- In ChatGPT, open Settings, then Apps & Connectors. Under Advanced settings, turn on Developer mode. If your menus say Plugins instead, follow OpenAI's route: Plugins, the plus button, Add custom MCP server.
- Choose Create, paste the link, pick OAuth and sign in with your Bingle account.
- On Bingle's sign-in screen, leave sending and spending unticked for your first week.
- Start a new chat, type @ and pick Bingle, then ask a read-only question to confirm it answers.
A good first question is the one Bingle suggests after you connect: catch me up, using activity_since and get_overview, without sending anything or spending credits. If the answer lists your recent leads and replies, the connection works. If ChatGPT says it has no such tool, it isn't calling the connector; start a fresh chat and pick it with @ again.
Prompts for a prospecting session
The four starter prompts on Bingle's Claude and ChatGPT page cover catching up, today's best leads, finding people like your favourites and tuning your ICP. The ones below are for the session after that, when you're working through actual accounts. Naming the tool in the prompt helps ChatGPT pick the right one.
| Prompt to paste |
|---|
| Shortlist |
| "Which five Bingle leads should I write to today? Use search_leads with view ready and give me each person's signal, its date and the link." |
| Choose between two |
| "Use explain_lead_score on [lead A] and [lead B]. Which one would you contact first, and which signal would you mention?" |
| Connection note |
| "Draft a connection note for [lead] with draft_message, from their latest signal. Keep it under 200 characters and don't send it." |
| Message after they accept |
| "Now write the first message for after they accept: open with the signal, say why I'm writing, and end with a question. Under 60 words." |
| Add your own research |
| "Here is [company]'s latest announcement: [paste]. Combine it with the Bingle signals for people at that company, and mark which sentences come from the page and which are your guess." |
| Check what's working |
| "Use signal_stats: which signal brought in the most leads this month,, and which brought in the fewest?" |
| Widen the net |
| "Watch people who comment on posts by [competitor 1] and [competitor 2], and add the ones who fit my ICP. Show me the plan before you set it up." |
The 200-character limit in the third prompt is LinkedIn's: its help page on personalising invitations, read on 29 October 2026, caps a note at 200 characters. The 60 words in the fourth is our own habit for a first message, not a LinkedIn rule. The fifth prompt asks ChatGPT to label its guesses because a model joining two facts can invent the link between them, and you want to see that before it reaches a buyer.
Work in that order: shortlist, choose, draft, then check. If a shortlist looks wrong, reject the leads in Bingle with a reason. Bingle learns from rejection reasons, and the Tune who I look for prompt later turns them into ICP changes you approve one at a time. If a draft feels generic, connection messages by signal has an example opener for each kind of signal.
What to check before anything sends
Controls sit in three places. Set each one on purpose rather than finding out later which was on.
| Setting | Effect |
|---|---|
| Bingle sign-in from ChatGPT | |
| Sending and spending unticked | ChatGPT can't send a message, an invite or a reply through Bingle, or spend credits on email finding or LinkedIn lookups |
| Bingle sign-in, sending ticked | |
| Sending allowed | Bingle still shows exactly what goes to whom and waits for your yes |
| Bingle sequences | |
| Review on, or Send LinkedIn steps automatically off | Message drafts wait in the Approval queue. With automatic sending off, each step waits in Today's sends for you to send by hand. Visits, likes and blank invites run on their own unless automatic sending is off |
Then read the draft itself. These are the checks we'd run on every first message, whichever tool wrote it:
- Open the signal's link and confirm the person really did the thing the message mentions, on the date it says.
- Check any fact about the company against its own page, not ChatGPT's summary.
- Read the first line out loud. If it would fit any prospect, rewrite it around the signal.
- Keep connection notes inside 200 characters, and leave the pitch for later.
- Paste it into our free [LinkedIn message checker](/linkedin-message-checker), which flags lines that read like a template or a pitch.
One more risk is specific to connectors. A LinkedIn post or a profile can contain text written to give an AI orders. Bingle passes text other people wrote to the AI as data, so a post saying "ignore your rules" changes nothing, and Bingle never sees your LinkedIn password.
Where this setup falls short
Bingle is LinkedIn only. It has no cold email or email sequences, no contact database and no self-serve CRM integration, so if your process runs through a CRM you'll copy outcomes across yourself. The connector works on your Bingle workspace, so ChatGPT can search the leads Bingle has found and set up new sources, but it can't browse LinkedIn at large for you.
If your ChatGPT plan or workspace won't let you add a custom app, Bingle's built-in Home chat works on the same data inside Bingle. You can describe who you want, paste profile links, ask for a draft or ask why a lead is hot, and anything that sends asks first with Go ahead or Not now.
Sends still go from your own LinkedIn account, in the working hours you set as the sender. If those are 9 am to 5 pm AEDT, that's the window, whichever chat you typed the request in. LinkedIn's invitation limits apply as normal: its help centre says a restriction for hitting them typically lasts a week. For the rules on contacting Australian buyers, see LinkedIn outreach in Australia; if you use Claude as well, how to use Claude for LinkedIn prospecting covers that side. Plans and prices are on the pricing page, and more guides are on the Bingle blog.
Questions
Can ChatGPT find leads on LinkedIn?
Not on its own, since it can't see behind your LinkedIn login. With Bingle connected, it can search the leads Bingle's sources have found and set up a new source that watches for the people you describe.
Can ChatGPT send LinkedIn messages for me?
Only through a connector with a send tool. Bingle's connector keeps sending off unless you tick it at sign-in, and even then shows exactly what goes to whom and waits for your yes.
Do I need a paid ChatGPT plan to add a connector?
Plan support changes, so check for Developer mode in your own Settings. When Bingle's team read it on 24 October, OpenAI gave full support on Business, Enterprise and Edu, with an admin turning it on for Business.
Is it safe to connect a LinkedIn tool to ChatGPT?
OpenAI shows a risk warning before you add a custom MCP server, so only add ones you trust. Start with sending and spending switched off, and read every confirmation before you approve it.
Should I use ChatGPT or Claude for LinkedIn prospecting?
Use the one you already work in. Bingle's connector link is the same for both; the steps for each are on Bingle's [Claude and ChatGPT page](/claude-and-chatgpt).
Sources
- OpenAI Developers, Connect and test your plugin. read 29 Oct 2026
- OpenAI Developers, Plugins quickstart. read 29 Oct 2026
- OpenAI Help, Developer mode and MCP apps. read by Bingle's team 24 Oct 2026; returned 403 to our reader on 29 Oct 2026
- LinkedIn Help, Personalize invitations to connect. read 29 Oct 2026
- LinkedIn Help, Invitation limit reached. read 29 Oct 2026
More from the blog
- How to use Claude for LinkedIn prospecting, step by step
- How to score LinkedIn leads by fit and intent (with a worked formula)
- LinkedIn connection request messages for each buying signal (with examples)
- Which LinkedIn outreach tools let you approve every AI message first?
- LinkedIn outreach in Hong Kong: language, timing and rules
- LinkedIn outreach in Malaysia: what works for B2B
See who's looking right now
Your first leads show up in minutes, each with its reason. Free for 7 days, no card.