AI Prompts for Sales: 14 LinkedIn Prompts for ChatGPT and Claude
Most AI prompts for sales ask the assistant to work from text you paste in. Prompts that read your own LinkedIn account – who someone is now, what they posted, how you're connected, what you last said – return answers you can act on. Below are 14 copy-ready prompts grouped by job, two of them shown as full runs, each with what it reads, what it returns, what happens when data is missing, and where you approve.
The short version. A prompt is only as specific as what the assistant can see. Connect ChatGPT or Claude to your own LinkedIn account through Linked API, and these prompts find people by role, place and connection degree along with their recent posts, show who you already know at an account, tell you who wrote to you and who's still waiting for an answer, list who commented on a post, and tidy up old invitations. Every prompt ends with "send nothing" – messages and invitations go out only after you approve them.
Here is the first prompt in the library, run end to end:
Used Linked API: search_people, fetch_person
- title "Head of Revenue Operations", location Germany, industry Software Development, 2nd degree, 12 results
- posts from the last 30 days
12 found; 7 posted in the last 30 days, 4 of them about forecasting or CRM data. Notes are drafted for those four and for a fifth person whose recent post was about pipeline reviews – a weaker fit, marked as such. The other seven are listed without drafts. Nothing has been sent.
| Person | Company | Posted about | Degree | Draft note |
|---|---|---|---|---|
| Lena Hoffmann | Northlake Software | Moving the forecast from quarterly to monthly | 2nd | "Hi Lena – your post on a monthly forecast rhythm matched what we see in RevOps teams your size. Would be glad to connect." (121) |
| Jonas Weber | Brightpath Cloud | Cleaning duplicate accounts before a CRM migration | 2nd | "Hi Jonas – your CRM migration post was spot on about duplicates. I work on the same problem with RevOps teams; happy to connect." (128) |
| Miriam Kraus | Tessel Data | Forecast accuracy after a pricing change | 2nd | "Hi Miriam – interesting take on forecast accuracy after a pricing change. Would love to follow how it plays out. Glad to connect." (129) |
| Felix Brandt | Orbitel | Who owns CRM data quality: RevOps or sales | 2nd | "Hi Felix – your question on who owns CRM data quality got a lot of good answers. Happy to connect and compare notes." (116) |
| Sara Yilmaz | Kestrel Apps | Running weekly pipeline reviews (weaker fit) | 2nd | "Hi Sara – liked your post on weekly pipeline reviews. I spend most of my time with RevOps leads on the same rhythm; glad to connect." (132) |
invitations go out only when you approve them – at most 10 a day by default
Why most sales prompts return generic output
Sales teams already lean on AI for this work: Salesforce's State of Sales survey of 4,050 sales professionals (fielded August–September 2025) found that "55% of sales professionals are using AI for prospecting, with another 38% planning to do so", and that "sellers expect agents to cut prospect research time by 34%".
The output is only as good as the input. A prompt like "write a LinkedIn note to this VP of Sales" gives the model a job title and nothing else, so it writes what fits every VP of Sales. Pasting a profile helps, but it's one person at a time and already out of date. When the assistant can read the account you sell from, it sees the person's current role, what they posted this month, whether you're connected and what you last said – and the same prompt works for twelve people as well as one.
Where these prompts run
You need a Linked API account with your LinkedIn account connected to it. Linked API works through a dedicated cloud browser on your own account, at a human pace, and never stores your password. The setup steps cover each assistant, and what the connection can do lists every tool.
- ChatGPT. Per OpenAI's help, "Custom MCP server connections and full MCP support are available for ChatGPT Business and Enterprise/Edu customers on ChatGPT web", and "Pro users can connect custom MCP servers with read/fetch permissions". By that rule and Linked API's read-only marks, on Pro these prompts can research, draft and list pending invitations, and read messages that are already captured – but they can't turn on inbox monitoring, load missing conversation history, send or withdraw anything.
- Claude. Per Anthropic's help, custom connectors are "available on Claude, Cowork, and Claude Desktop for users on Free, Pro, Max, Ultra, Team, and Enterprise plans", and "Free users are limited to one custom connector".
- Claude Code, Cursor and Codex. Use the same MCP server, the agent-friendly CLI or a ready-made skill – the LinkedIn AI agent guide compares the routes.
For the full call-prep workflow – a brief before the call and the follow-up after it – see ChatGPT for sales and Claude for sales. This page is the library of single jobs.
AI prompts for sales prospecting
1. People by role, place and degree, with their recent posts
Find [12] people with the title [Head of Revenue Operations] in [Germany] in the [Software Development] industry who are 2nd-degree connections of mine. Read each one's posts from the last [30] days and rank them by whether they posted about [forecasting or CRM data]. Return a table: name, company, what they posted about, degree, and a connection note under 200 characters for the top [five]. If someone hasn't posted in that window, write "no posts since [date]" – don't invent a topic. Send nothing.
Reads: a people search by title, location, industry and connection degree; each person's recent posts. Returns: the ranked table with drafts. If data is missing: "no posts since …", never a padded topic. Approval: nothing is sent. The industry has to be LinkedIn's exact label – see the LinkedIn industry list – and connection degrees explains 1st, 2nd and 3rd+.
2. Former colleagues at your target accounts
Find people who used to work at [my former employer] and now work at [Account A] or [Account B], and who are 1st- or 2nd-degree connections of mine. Return a table: name, current role, account, the years they were at [my former employer] if their profile shows them, and my degree. If no one matches at an account, write "none found" for it. Send nothing.
Reads: a people search by past and current company and degree; experience dates. Returns: one table per account. If data is missing: "none found" per account. Approval: nothing is sent. A shared employer is a reason to write, not proof you know each other.
3. Run your saved LinkedIn search
Run this LinkedIn people-search URL: [URL]. Return the top [25] with their headline and my connection degree, and flag anyone I already have a conversation with. If a conversation hasn't been loaded, write "not loaded" – not "never talked". Send nothing.
Reads: the search exactly as you set it up in LinkedIn, so any filter you can click there works; your conversation with each person, where loaded. Returns: a table. If data is missing: "not loaded". Approval: nothing is sent. For the full process from an ideal-customer profile to a first invitation, see AI sales prospecting on LinkedIn.
Research one person before you write
4. Is now a good time to write?
Check [profile URL]: time in their current role, their posts and comments from the last [60] days, whether we're connected, and our last exchange if our conversation has been loaded. Give me a one-line verdict – write now, wait, or skip – with the reason. If our conversation hasn't been loaded, say so instead of assuming we've never spoken. Send nothing.
Reads: the profile, recent posts and comments, the connection status, the loaded conversation. Returns: a verdict and its reason. If data is missing: "conversation not loaded". Approval: nothing is sent.
5. What they engage with
Read [profile URL]'s comments and reactions from the last [90] days. Group them by topic, with one example link per topic and how often each topic comes up. If there's little activity, say how much there is rather than stretching it. Send nothing.
Reads: up to 20 recent comments and 20 reactions. Returns: topics with examples. If data is missing: the actual count, however small. Approval: nothing is sent.
Account context
6. Who you know at the account
At [company page URL], list people with [sales] or [revenue operations] in their title who are 1st- or 2nd-degree connections of mine. Return a table: name, title, degree. These are people I can see, not an introduction path – don't suggest who could introduce me. If no one in those roles is a 1st- or 2nd-degree connection, say so. Send nothing.
Reads: the company's employees, one title keyword at a time, filtered by your connection degree. Returns: a table. If data is missing: "no one in that role at 1st or 2nd degree". Approval: nothing is sent. To map everyone in the deal, see mapping a B2B buying committee.
7. What changed at the account this month
For [company page URL], read the company's posts from the last 30 days and find [sales or revenue] roles it posted in the past month. Return two short lists with dates. If a list is empty, say so. Don't guess funding, tech stack or buying intent – they aren't in what you read. Send nothing.
Reads: company posts and job postings. Returns: two dated lists. If data is missing: "nothing in the last 30 days". Approval: nothing is sent.
Inbox and replies
8. Who wrote to you this week
Which of my LinkedIn conversations got a new message from the other person in the last 7 days? For each, tell me what they asked for and whether I've answered, and draft a reply where one is owed. If nothing has been captured, say so – don't conclude no one wrote. Send nothing.
Used Linked API: get_inbox, fetch_person
- messages from the last 7 days
- the name and headline of each person who wrote
4 conversations had a new message from the other person this week. You've answered one; replies are drafted for the other three. Nothing has been sent.
| Person | Their last message | What they need | Answered? | Draft reply |
|---|---|---|---|---|
| Daniel Ortiz | "Can you send the security questionnaire before Friday?" | Your completed questionnaire | No | "Hi Daniel – sending it over Thursday morning, along with our SOC 2 summary." |
| Priya Menon | "Thursday works, what time?" | A time for the demo | No | "Great – does 10:00 your time work? I'll send the invite." |
| Tom Lindqvist | "We're pausing this until Q1." | Nothing now | No | "Understood, Tom – I'll check back in January. Good luck with the quarter." |
| Aisha Rahman | "Thanks, got it." | Nothing | Yes | – |
each reply goes out only when you approve it
Reads: messages captured since inbox monitoring was turned on (once per account). Returns: the table with drafts. If data is missing: "nothing captured", never "no replies". Approval: each reply.
9. Who you're keeping waiting
Find conversations where the other person wrote last and I haven't answered in [3] or more days, from the last [30] days of captured messages. Oldest first: name, their last message, days waiting, and a one-line draft reply. If nothing has been captured, say so. Send nothing.
Reads: captured messages from the last 30 days. Returns: a waiting list, oldest first. If data is missing: "nothing captured". Approval: each reply.
10. Load the history, then draft the follow-up
Read my conversation with [profile URL]. If the earlier history isn't loaded, load it first. Summarize where we left it in three lines and draft a follow-up. If the history can't be loaded or comes back incomplete, tell me that instead of drafting from an empty conversation. Show me the draft and wait.
Reads: your conversation with one person, loading the earlier history once if it's missing. Returns: a summary and a draft. If data is missing: the assistant says the history is unavailable or incomplete rather than drafting from nothing. Approval: the follow-up is sent only after you approve it. For reply handling at scale, see how to automate LinkedIn messages.
Engagement on posts
11. Who commented on a competitor's post
From this post: [post URL], list the [100] most recent commenters whose headline says [sales or RevOps], with my connection degree and what they said in one line. Skip company pages. If no commenters match, say so. Send nothing.
Reads: the post's comments. Returns: a filtered table. If data is missing: "no commenters match". Approval: nothing is sent. To watch a topic or competitor over time, see LinkedIn social listening.
12. Who reacted to your last post
Find my most recent post and list the people who reacted: name, headline, my connection degree, and a suggested next step – thank, connect, or skip. If you can't find a recent post, say "no post found"; if the post has no reactions, say "no reactions returned". Send nothing.
Reads: your latest post and its reactions. Returns: a table with next steps. If data is missing: "no post found" or "no reactions returned" – two different answers. Approval: nothing is sent.
13. Who's talking about your problem this week
Find posts from the past week about [topic] by authors in the [industry] industry. LinkedIn widens narrow searches, so drop posts that aren't really about [topic]. Return the author, headline, my connection degree and a one-line summary. If nothing on-topic is left, say so. Send nothing.
Reads: a post search by topic, date and author industry. Returns: on-topic posts only. If data is missing: "no on-topic posts this week". Approval: nothing is sent.
Invitations
14. Pending invitations older than three weeks
List my sent invitations that have been pending for more than [three weeks], oldest first, with name, headline and how long ago LinkedIn says each was sent (for example "1 month ago"). Keep borderline ones like "3 weeks ago" in a separate list instead of guessing an exact date. If none are older than three weeks, say so. Don't withdraw anything until I mark which ones.
Reads: your pending sent invitations, with the age LinkedIn displays. Returns: a list sorted by that age, borderline cases apart. If data is missing: "no invitations older than three weeks". Approval: only the rows you mark are withdrawn. The full clean-up is in LinkedIn pending invitations.
Prompts that need no LinkedIn data
Some sales prompts don't need your account at all – the assistant only needs what you tell it:
- Objection handling. "My prospect said: '[objection]'. Give me three ways to answer it in one or two sentences each, using only these facts about our product: [facts]."
- A first line from your notes. "Write the first line of an email to [name] using only these notes from our call: [notes]. No compliments and no claims I haven't given you."
- Discovery questions. "Write eight discovery questions for a first call with a [role] at a [company type] about [problem], grouped by situation, impact and decision."
For LinkedIn message templates, see LinkedIn cold outreach strategies.
How to write your own
Every prompt above has the same four parts. Keep them when you write your own:
- The input – who, which list or which URL, and the time window ("the last 30 days").
- The output shape – a table with named columns, or a one-line verdict.
- The missing-data rule – what to write when something isn't there ("not loaded", "none found"), so the assistant reports a gap instead of filling it.
- The approval line – "send nothing" or "show me and wait", so drafts stay drafts until you say otherwise.
What these runs don't include
- Emails, phone numbers and intent data. These prompts read what your LinkedIn account can see; use your enrichment provider for contact details.
- Funding-stage or tech-stack filters. Bring your own account list and paste the company pages in.
- CRM updates. Writing results back needs a CRM connection in your assistant, or a workflow you build – see building workflows in n8n.
- Sequences, automatic stop-on-reply and alerts. Each prompt is a run you start; scheduled follow-ups and stop-on-reply need an automation built around the LinkedIn tools, such as the standard outreach sequence in n8n, or the sequencer you already use.
- Private profiles. A private or limited profile stays limited – the assistant reads what your account can see.
Frequently Asked Questions (FAQ)
The ones that give the assistant real context and a clear output: who to look at, which window, what table to return, what to write when data is missing, and "send nothing" until you approve. Prompts that read your LinkedIn account – people by role and degree with their recent posts, who you know at an account, who's waiting for an answer – are more specific than prompts that work from pasted text.
Yes. Once your LinkedIn account is connected through Linked API, the same prompts run in ChatGPT, Claude, Claude Code, Cursor and Codex. What changes is the plan: in ChatGPT, full use needs Business or Enterprise/Edu, and Pro is limited to reading.
Per OpenAI's help, custom MCP connections with full support are available on ChatGPT Business and Enterprise/Edu on the web, and Pro gets read/fetch permissions. On Pro, these prompts can research, draft and read messages already captured, but can't load missing history or send – details are in where these prompts run.
Yes, once your account is connected: it reads conversations Linked API has captured. New messages are captured after inbox monitoring is turned on, and earlier history with one person can be loaded on request. An empty result means nothing has been captured yet, not that no one wrote.
Not with these prompts – each one ends with "send nothing" or "show me and wait", and nothing goes out until you approve. ChatGPT and Claude also ask for confirmation before tools that send, depending on your settings.
As many as your limits allow. By default, Linked API allows 300 people searches a month, 80 profile views a day and 10 invitations a day – your settings can differ, see limits. Prompt 1 uses one search and 12 profile views.
Build it in, or hand it to an agent
Get it out of the box: connect the MCP server as described above, the agent-friendly CLI or a ready-made skill, and paste these prompts in plain language. Or build the same jobs 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 – Salesforce's State of Sales announcement and OpenAI's and Anthropic's help articles on custom connections checked on that date.