ChatGPT for Recruiting: Shortlist and Reach Candidates on LinkedIn
ChatGPT for recruiting usually means writing job ads, Boolean strings and outreach drafts – useful, but ChatGPT on its own doesn't search from your LinkedIn account. LinkedIn's own plugin in ChatGPT looks a person up. Connected to your own LinkedIn account through Linked API, ChatGPT can also search with your network, read candidates' full experience against your must-haves, draft first touches you approve, and read the replies back.
The short version. Give ChatGPT one role and your must-haves. It searches LinkedIn from your account, reads the experience of the closest matches and returns a shortlist with the evidence for each – nothing is sent. Then it drafts a connection note or a message for the strong ones, and sends only what you approve. You need a Linked API account with your LinkedIn account connected, added to ChatGPT as a custom connection – full use on Business and Enterprise/Edu on the web; by OpenAI's rule and Linked API's read-only marks, the shortlist run also fits Pro, though that hasn't been tested on a Pro account yet.
Here is one role taken to a shortlist from a single ask – the setup is below:
Used Linked API: search_people, fetch_person
- "enterprise account executive", Chicago area, past companies CrowdStrike, Okta, Zscaler or Palo Alto Networks, up to 50 – each with your connection degree
- the 20 closest picked by headline
- experience with dates and descriptions
20 read: 6 strong, 9 possible with what's missing named, 5 excluded with the reason. Nothing has been sent. Strong means the profile shows 4+ years in quota-carrying roles whose descriptions name enterprise deals or accounts, at a security or identity company, in the Chicago area.
| Candidate | Now (role, time in it) | Security or identity sales | Enterprise closing evidence | Fit | Degree |
|---|---|---|---|---|---|
| Marcus Lee | Enterprise AE at an identity startup, 2 years | Yes – identity, 6 years in total | "Closed seven-figure deals with Fortune 500 banks" across 5 quota-carrying years | Strong | 2nd |
| Naomi Fischer | Senior Enterprise AE at a data-security vendor, 1 year | Yes – security, 5 years | "Ran 12 enterprise accounts in financial services" over 4 years | Strong | 1st |
| Hannah Cole | Enterprise AE at a security vendor, 3 years | Yes – security | Title only; no deal or account description | Possible – enterprise closing unknown | 3rd+ |
| Victor Ramos | Account Executive at a security vendor, 2 years 3 months | Yes – after an SDR start | 2 years 3 months in quota-carrying roles | Excluded – under 4 years closing | 2nd |
nothing is sent – the first touch is the next step, and you approve it
What ChatGPT already does for recruiters – and what your LinkedIn account adds
| Job | ChatGPT on its own | LinkedIn's plugin in ChatGPT | Your LinkedIn account through Linked API |
|---|---|---|---|
| Job ad and Boolean string | Drafts both from what you tell it | Not in its listing | Same drafts, plus a search run on your account |
| Finding people | Public web results only | Looks up a person you name | Searches by title, location, past companies and connection degree |
| Reading a profile | Pasted text or public web excerpts | Title, company, location and follower count | Full experience with dates and descriptions |
| First touch | A draft you copy | Not in its listing | A connection note or message, sent only when you approve |
| Replies | – | Not in its listing | Your conversations, read back |
LinkedIn's plugin listing describes it as a way to "Look up any professional on LinkedIn without leaving ChatGPT. See their title, company, location and follower count". Recruiting vendors also offer ChatGPT connections to their own data – SeekOut's candidate database, Leonar's stored talent pool, several ATS apps; those search the vendor's data, not your LinkedIn account.
Recruiters are already moving this way: in LinkedIn's Future of Recruiting 2025 report (1,271 recruiting professionals, surveyed in September 2024), 37% said their recruiting organization was "actively integrating" or "experimenting" with generative AI, up from 27%.
What you need first
You need a Linked API account with your LinkedIn account connected to it, added to ChatGPT as a custom connection. Linked API works through a dedicated cloud browser on your own account, at a human pace, and never stores your password. The setup steps for ChatGPT are in the installation guide, and what the connection can do is covered in the overview.
- Your ChatGPT plan. Per OpenAI's help, "Custom MCP server connections and full MCP support are available for ChatGPT Business and Enterprise/Edu customers on ChatGPT web". On Business, "Only admins and owners can create and publish plugins for custom MCP servers"; on Enterprise/Edu, admins can let selected members connect one.
- On Pro. "Pro users can connect custom MCP servers with read/fetch permissions." By that rule and Linked API's read-only marks, Pro can run the shortlist above and read messages already captured, but can't load a conversation's history or send anything – not yet tested on a Pro account.
- How a run starts. Each run is an ordinary chat: pick the app from the @ menu in the message, and pick it again on a later message that needs it.
ChatGPT is one way in. The same account also works from Claude, Cursor or Claude Code through the MCP server, the agent-friendly CLI or a ready-made skill – the LinkedIn AI agent guide compares the routes.
First touch, approved
Take the strong matches to drafts:
Used Linked API: check_connection_status
- 1 connected, 1 with your invitation already pending, 4 not connected
5 drafts – 1 message and 4 notes; nothing new for the pending one. A free account can add a note to three invitations a month: pick three, or send the fourth without a note. Nothing has been sent.
| Candidate | Status | Goes as | Draft |
|---|---|---|---|
| Naomi Fischer | Connected | Message | "Naomi – I'm hiring an Enterprise AE in Chicago for a Series B identity-security company. Your financial-services accounts are a close match. Worth a short call this week?" |
| Marcus Lee | Not connected | Note (177 characters) | "Hi Marcus – your move from a large identity vendor to a Chicago startup caught my eye. I'm hiring an Enterprise AE for a Series B identity-security company here. Open to a chat?" |
| Leo Brandt | Not connected | Note (164 characters) | "Hi Leo – five years closing enterprise deals in endpoint security stood out. I'm hiring for a Series B identity-security company in Chicago; happy to share details." |
| Grace Owens | Not connected | Note (164 characters) | "Hi Grace – your enterprise healthcare accounts at a security vendor match a Chicago role I'm filling for a Series B identity-security company. Open to hearing more?" |
| Amir Haddad | Invitation pending | Nothing new | – |
nothing goes out until you approve each draft – ChatGPT may also ask before it sends
Did they reply? A week later, read the conversation with every candidate you contacted – not only those who accepted. Per LinkedIn's help, a person can reply to an invitation message without accepting it. Ask ChatGPT to load each conversation once if it isn't loaded, then read it, and keep three outcomes apart: replied; no reply in the loaded history; and "reply not verified" where the history couldn't be loaded or is partial. An empty result is never proof of no reply, and the connection status doesn't answer whether someone wrote back. On Pro, only messages already captured can be read. For the follow-up cadence after the first touch, see how to engage passive candidates.
ChatGPT's own approvals. Per OpenAI's help on app permissions, with "Always ask", "ChatGPT asks before reading app information or making changes", and sending a message is one of its examples of an action that "may require additional review".
Prompts recruiters can reuse
Each prompt carries its own inputs, output and what to do when data is missing:
- A job ad from intake facts (ChatGPT alone): "Write a 250-word job ad and a one-line LinkedIn post for [role] from these intake notes: must-haves [list], nice-to-haves [list], location and hybrid terms [terms], salary band [band, if shared], three facts about the team [facts]. Where something isn't in my notes, leave [TBD] instead of inventing benefits or numbers."
- A Boolean string for LinkedIn search (ChatGPT alone): "Write a Boolean search string for [titles], with past employers [list], skills [list] and excluding [terms]. Explain in plain words what it matches. If a term has common synonyms, list them instead of guessing which one LinkedIn uses." Operators are explained in LinkedIn Boolean search.
- Read a profile against the must-haves (your account): "Read [profile URL] and return a table: must-have, the evidence on the profile, and met, unknown or not met. A job title or employer name alone isn't evidence – mark it unknown. Don't contact them."
- Interview questions from what's still unknown: "For [role], using these must-haves [list] and the 'unknown' rows from the profile read, write six interview questions – each tied to one unknown – and what a strong answer would show. If nothing is unknown, ask about the riskiest must-have instead."
- Reply check across everyone you contacted (your account): "For these candidates [list], read my conversation with each – load it once if it isn't loaded. Return replied, no reply in the loaded history, or reply not verified, with their last message. If a conversation can't be loaded or is partial, say 'reply not verified'. Send nothing."
What this does not do
- LinkedIn Recruiter features – Recruiter seats, projects, Recruiter InMail and Hiring Assistant are not part of this; the tier comparison is in AI candidate sourcing, and InMail in LinkedIn InMail.
- ATS updates – unless you also add an ATS app in ChatGPT, or build the write-back with n8n or Make.
- Emails and phone numbers.
- Screening decisions – the shortlist is profile evidence; a person decides, and searches never filter on protected characteristics.
- Alerts – nothing tells you when a candidate changes jobs; each check is a run you start.
- Private profiles stay limited – ChatGPT reads what your account can see.
Frequently Asked Questions (FAQ)
For job ads, Boolean strings, interview questions and outreach drafts on its own – and, connected to your own LinkedIn account through Linked API, for searching candidates with your network, reading their full experience against your must-haves, sending first touches you approve and reading the replies.
On its own it can write the search, but it doesn't run it on LinkedIn from your account. Connected through Linked API, it searches by title, location, past companies and connection degree, reads profiles and returns a shortlist with evidence – you still decide who's a fit.
LinkedIn's own plugin in ChatGPT looks up a person you name. To search LinkedIn the way a recruiter does – by title, location and past employers, from your own account and network – connect your account through Linked API.
Yes, and that works without any connection: give it titles, employers, skills and exclusions, and ask it to explain what the string matches. Connected to your account, it can also run the search and read the results.
Per OpenAI's help, full custom connections are on Business and Enterprise/Edu on the web, and Pro gets read/fetch permissions. On Pro the shortlist run fits by that rule, but sending and loading conversation history don't – see what you need first.
Not with these runs – each prompt asks ChatGPT to show the drafts and wait, and nothing is sent until you approve. ChatGPT's own permission settings can also ask before it sends.
No. It runs on a regular LinkedIn account without Recruiter's projects, InMail credits or Hiring Assistant. For how the tiers compare, see AI candidate sourcing.
Connected to your LinkedIn account, it does some of the work one does – searching, reading, drafting and reading replies – with you approving each step. For what a full AI recruiting agent is and the options, see AI recruiting agents.
Yes – the same LinkedIn account connects to Claude, Claude Code and Cursor; the LinkedIn AI agent guide covers the routes. For the sales version of these runs, see ChatGPT for sales.
Build it in, or hand it to an agent
Get it out of the box: connect the MCP server as set up above, the agent-friendly CLI or a ready-made skill, and ask in plain language. Or build it into your own tools with the REST API, the Node and Python SDKs and the shell CLI. Either way it runs on your own LinkedIn account. Plans are flat per seat, from $49/mo billed annually – see pricing.
Facts verified November 9, 2026 – OpenAI's help articles on custom connections and app permissions, LinkedIn's ChatGPT plugin listing, LinkedIn's Future of Recruiting report and LinkedIn Help on invitations checked on that date.